Just a few weeks ago the UC Berkeley School of Law announced a new set of policies regulating AI use for its students. While the full document offered some flexibility for professors who were more open to AI in their classes, the default policy for the school was vanishingly close to an outright ban:

The use of AI is prohibited for aid in conceptualizing, outlining, drafting, revising, translating, or editing any work submitted for credit. AI use is prohibited for any use for any purpose in any exam situation. Students may not upload course materials—including assignments, readings, slides, class recordings, or other class content—into generative AI systems. AI can be used for research on papers ONLY for the limited purpose of identifying sources, such as cases, statutes, or secondary sources. Students are responsible for the accuracy of their research and all other aspects of their submitted work. Citations to sources that do not exist will raise a presumption of prohibited AI use.

I suspect we are going to see an increasingly wide divergence in the coming months in the way that academic institutions integrate AI into the classroom, with many institutions following the default-ban approach taken here by Berkeley Law, while others embrace the new AI platforms (or at least some of them.) One example from the other end of the spectrum: the Harvard Law professor Larry Lessig made NotebookLM a foundational part of his constitutional law class this past semester, by creating 110 notebooks for each legal case that was covered in the syllabus. Students would then add other sources (the oral argument, newspaper reports) to deepen their understanding of how each decision was rendered. Even my old high school—which I had happened to visit for my fortieth reunion a few weeks before the Berkeley announcement—has adopted an approach that offers their teenage students significantly more agency over their AI use than Berkeley gives to its adult aspiring lawyers. Instead of a default ban, the school has introduced a tripartite system that teachers can adopt in their classes or for specific assignments, color coded along the lines of traffic signals: green means you are free to use AI in every step of the process; yellow limits you to conceptual feedback and editing suggestions; red means no AI altogether.

Let me say at the outset that I am not trying to disparage Berkeley’s policy here. It’s a complicated moment. I know the Berkeley faculty are genuinely trying to figure out an approach that minimizes learning loss and embeds solid thinking habits for their students. It undoubtedly makes sense to restrict the use of AI in assessment settings, like a final exam. And if you read the full statement announcing the policies, it’s clear that the school isn’t rejecting the utility of AI outright. It opens with these lines:

Future lawyers may need to use artificial intelligence (“AI”) fluently. But the current state of the technology requires that AI use be coupled with the cognitive skills necessary to strategically deploy the technology, to critically assess its work product, and to uphold ethical obligations to clients and to the legal system. In short, thinking remains the sine qua non of good lawyering (and of a quality legal education). This policy seeks to ensure that our courses focus on requisite cognitive skills by default.

Berkeley is alluding to the phenomenon known as cognitive offloading here—a phrase first coined only a decade ago that has begun to seep into everyday conversation in the past year or so. (Its use in academia has exploded as well; Google Scholar reports 6,000 papers using the phrase in the past five years, ten times the number of references in the preceding half decade.) At first glance you might think of cognitive offloading as just academic jargon for cheating, but the concept is more subtle than that and less universally pejorative. Some forms of cognitive offloading are indeed negative in their effects, as in a student who bypasses actually researching and writing a paper by handing the task over to Claude. But other forms are clearly beneficial. There was cognitive load in navigating the complexities of index cards and the Dewey Decimal system in the old days of analog libraries; offloading that mental work to a good search algorithm actually freed up our minds to focus on more nuanced problems. Calculators undoubtedly deepened our understanding of math and economics, even though technically we were shifting our computation workloads to the machines.

The problem with the Berkeley Law approach is that it assumes that there is a fundamental zero-sum game between AI use and human thinking. The logic is: if you are using AI in any fashion, by default you are thinking less. The Berkeley ban doesn’t entertain the possibility that AI, used properly, can actually make us better thinkers. In the interest of eliminating all possible forms of cognitive offloading it also eliminates all the ways that AI can get new ideas into your brain, or push your thinking in new directions, or assess your mastery of the material you are trying to understand.

In a conversation with Dan Blumberg on the Future Around And Find Out podcast, I referred to this as cognitive uploading. The phrase came to me spontaneously in the conversation almost as a joke. (In the podcast you can hear me cycle through variations: cognitive onboarding, cognitive inputting.) But the more I’ve turned it over in my mind since then, the more I think it points to an important element of the AI education debate that has been largely ignored. For understandable reasons, we’ve spent most of our time worrying about what happens when AI does our thinking for us. But we haven’t focused enough on all the ways AI gives us new things to think about.

By cognitive uploading, I mean something more than just traditional learning, where you have pre-defined material that you know you need to understand to pass the course or do your job successfully. It should go without saying that AI can be an enormous help in achieving that kind of knowledge mastery, particularly in a source-grounded environment like NotebookLM. Think about the learning journey that is now possible using Notebook’s Quiz feature, one of our most widely used Studio artifacts, developed in partnership with the amazing LearnX team at Google: you read any document you need to understand for your class, and then dynamically generate a quiz to test your comprehension; when you get an answer wrong, you hit the “explain” button which produces a chat response walking you through why the correct answer was more appropriate than your response, with direct links back to the original passages from the source material so you can review the key elements that you misunderstood. Just a year or two ago, that kind of on-demand learning support—with bespoke assessments generated for any text dynamically—simply wasn’t an option unless you could afford a personal tutor to generate the quiz for you, evaluate your answers, and assemble a digest of key passages based on your mistakes to review. Now it’s available to anyone with a web connection.

Using Quizzes in Notebook to improve your comprehension is an example of cognitive uploading where the tool helps you get “assigned information”—information you know you need to master—into your brain. But AI can just as easily steer you towards the information that you didn’t realize you needed, help you see around your own analytic blind spots, challenge your assumptions, or make novel connections that wouldn’t have otherwise occurred to you. One of my most common routines when I am working on an essay or a chapter outline is to share my latest thinking/writing with Notebook, and ask it: what am I missing? Notebook has access to all the sources I’ve used as part of the research for the project, along with my notes and other relevant things I’ve written in the past, and so it’s able to review all that material and suggest ideas or angles that hadn’t yet occurred to me, which inevitably sends me back to the original sources to explore the new lead. Or it reminds me of an earlier idea or passage that I’d forgotten about. In either case, though, the tool is giving me more to think about, not less.

Another example is using the AI as a kind of intellectual sparring partner. One of the very first prototypes we built after I joined Google was a feature we called “Contrarian.” You’d write a paragraph on any topic, and on demand the model would come up with a counter-argument to whatever you’d written. It was a bit of a party trick back in the day, and lacked any of the advanced thinking, research, and source-grounding that current AI platforms enjoy, but even then you could see the promise of it. Chatbots have a well-deserved reputation for sycophancy, but they are also skilled at adopting whatever persona you want from them. If you want the model to punch holes in your argument instead of lavishing you with praise, you can just ask for it.

AI also enables another form of cognitive uploading: triaging novel hypotheses to determine if they warrant our own deeper attention. So much of my creative thinking as a writer is a variation of hmmm.. I wonder if there’s something interesting here. By chance, I stumble across an article about ant colonies a few weeks after I’ve finished reading a book about urban development, and in my head I think: I wonder if there’s a useful connection to make between those two fields. The cost—just in time alone—of going down a rabbit hole like that has been reduced exponentially over my adult lifetime. When that ants/cities connection first occurred to me in the late 1990s, as I was gathering ideas for the book that became Emergence, it took months of ordering/reading books and visiting libraries to get enough confidence that this was an idea worth exploring. Now I can test an equivalent hypothesis—and get a high-level view of the existing thinking on the topic—in a matter of minutes using Deep Research in Notebook. It’s true that in a perfect world, the most cognitively engaged way to pursue stray hunches like this would be to assemble all the research yourself, but in practice, most of us don’t have the time for that kind of investment in an idea that is likely to be a red herring in the end. So we end up doing nothing with those hunches. The rabbit holes go unexplored. But now, by offloading the early stage exploratory research to the model, we can actually entertain more hunches and upload the most promising ones to our minds for deeper engagement.

To a certain extent, the concerns about cognitive offloading are an example of technological lag, where the mainstream discussion of AI and its impact is still framed in terms of the first-generation chatbots. The original Chat-GPT had none of the capabilities described above: source grounding, quiz generation, web research, hypothesis testing, and so on. But it was remarkably good at writing a paper for you with some clever prompting. So part of the conviction that AI is in a zero-sum battle with human thinking stems from people simply not following the progress since 2022, both in the underlying models themselves and the application frameworks we have built around them.

But I suspect the primary reason why we are so quick to resort to outright bans is we don’t have a clear enough picture in our heads yet of what an ideal engagement with AI would look like. What’s the partnership that helps you master the material but also make new connections? For understandable reasons, the conversation has mostly focused on the students who are tempted to use AI to bypass the hard work of thinking and learning, who simply want to create the illusion of understanding the material in order to pass the course and get on with their lives. But what about the students who genuinely want to deepen their understanding? How might AI actually assist them in that journey?

As a thought experiment, I tried to imagine what the optimal approach would be for writing an intensive research paper, assuming a student that honestly wants to maximize their understanding and create a compelling and original paper that will demonstrate their command of the material. It would look something like this:

• You read the assignment. In partnership with the AI, you make a plan to establish which primary sources you need to read to be able to write an accurate and original paper, and which sources you’ll need to consult but not read in their entirety. “Directed research”—where you dispatch an agent to find sources based on a prompt you write—is encouraged. (Find me the most relevant primary texts and secondary commentary that I need to understand the Watergate break-in.) That’s good cognitive offloading. And knowing how to prompt a complex research query is a valuable skill to master, at least for the 2026 rendition of the job market.

• You then read all of the essential sources, highlighting and annotating as you do. You can query the AI to help you understand specific passages or larger questions, but you don’t skip the crucial step of actually reading the material. The AI can draw on the supplemental sources (even if you haven’t read them) when answering your questions; on occasion, the AI will suggest highly relevant sections from the secondary material which you read and annotate as well.

• Once you’ve completed the primary reading, you decide on the overall approach and draft an outline, with the AI assistant providing feedback where needed. You can ask “negative search space” questions like: what am I missing? What’s the thing that I need to know to write this essay that I don’t know about yet?

• And then you sit down to write the paper. You draft everything yourself but consult with AI throughout: fact-checking based on the assembled sources, asking for stylistic alternative phrasing, chasing down surprising new leads as they emerge in the writing.

The rule of thumb is effectively to imagine that you have a professional-grade researcher, tutor, and editor at your side and just treat the AI the way you would treat those human collaborators. You wouldn’t ask your tutor to do your homework for you, but you would happily ask them to explain the cosmological constant or test your knowledge of the Dred Scott decision. I’ve been lucky enough to work with some brilliant editors over the course of my career. Do I lean on them to help me tighten or expand the outline for an essay I am writing? Do I ask them for alternate approaches to a particularly tricky section transition? Of course. (Is that cognitive offloading? No—it’s giving me more material to work with, expanding my sense of what’s possible.) But I would never ask my editor to write the whole essay for me. The same common sense principles should apply to working with AI.

I truly believe that following this kind of approach to AI in the classroom would, in the aggregate, produce better learning outcomes than the research and writing workflows available to students pre-AI. It would strengthen the students’ cognitive skills, not cause them to atrophy. I even suspect some of the faculty at Berkeley Law might embrace this approach if they felt it was a valid option. The harder question, though, is how you encourage or even enforce that kind of usage.

I could see a case for the Berkeley ban based entirely on the practical realities: today’s AI platforms could be terrific tools for thought, helping students deepen their understanding, but in reality, most students are going to take the low road and just one-shot their essay if you open the door for AI use even slightly. In other words, in theory AI is capable of providing brilliant intellectual scaffolding, but in practice, gravity always wins out in the end. I think that’s probably too cynical, or at least we’re too early in the development of these tools to make that call. So often with new software paradigms, what looks like inevitability turns out to be just design failure that can be solved with the right guardrails or affordances or system instructions. But I also think it is incumbent on those of us who are building these new platforms to figure out how to steer learners towards the deeper path, and away from the siren song of the one-shot paper. Certainly this is a project that we are deeply engaged with at NotebookLM. (Our original slogan for the product was “Do Your Best Thinking”—not “Let Us Do Your Thinking For You.”) But to move past this phase of outright bans, we’re going to need to build a new consensus—from educators and students alike—that a good-faith partnership with AI doesn’t just offer an escape route from cognition, but can actually make us better at the productive struggle of thinking.

Here’s the full conversation with Dan Blumberg:

And in other podcast news, I had a wonderful chat with Radical Candor author Kim Scott about my last book, The Infernal Machine:

[In the first three installments of this series, I explored how the widely accepted idea that agriculture was a sudden, catastrophic trap—a “colossal downgrade” popularized by writers like Jared Diamond and Yuval Noah Harari—misses a massive chapter of our history. Recent scholarship has revealed that humans actually spent roughly 4,000 years enjoying mix of foraging and managed resources without submitting to the grueling toil of full-time farming or the tax collectors of an early agrarian state.]

If the adoption of agriculture wasn’t the immediate catalyst for the miseries of the early state, if our ancestors spent four thousand years happily dabbling in part-time farming without being conscripted into forced labor, what finally caused the agrarian trap to snap shut?

Start with the basic facts of the timeline. The first agricultural states were established around 3200 BCE, most notably in the expanding Bronze Age city-state of Uruk, followed shortly thereafter by the neighboring city of Ur. Centralized authorities, typically embodied by priest-kings or secular rulers, emerged to manage the logistics of a newly complex economy, overseeing the collection of agricultural surplus to support a burgeoning class of non-agricultural specialists. Subsistence gave way to extraction; most of the population worked to generate food not just for themselves and their families, but to support the increasingly lavish lifestyles of the administrative elite.

The emergence of statehood was dependent on a few critical ingredients. One of them, as James Scott powerfully explains, was the biological nature of grain itself. It was harvested in regular cycles; it could be weighed, transported, and divided up in even units far more easily than rival crops in other places where agriculture developed. Because grain was, in Scott’s words, uniquely “visible, divisible, assessable, storable, transportable, and ‘rationable,’” it possessed an innate compatibility with a tax collecting regime that simply did not exist for the diverse staples of the earlier wetlands diet. A community subsisting on fish, waterfowl, and wild plants relied on common-property resources governed by a shifting web of natural rhythms. Even other domesticated foods like root crops and tubers proved remarkably resistant to the tax collector, because they grow underground and can safely be left in the soil for months, even years. All of which allows farmers to harvest them piecemeal as needed, making it virtually impossible for a tax collector to assess the yield or confiscate a surplus. Grains, on the other hand, grow visibly above ground, ripen all at once, and must be harvested and stored immediately, making them an easy target for centralized extraction.

But if there was something in the biological reality of grain that made it eminently “divisible” and “assessable,” our ancestors still needed a mechanism for recording and keeping track of those assessments. According to Denise Schmandt-Besserat’s influential “token-to-tablet” theory, the administrative origins of the state date back to a surprisingly simple Neolithic counting system that first appeared as early as 8000 BCE. Long before anyone drew an abstract symbol, these early record-keepers used small clay tokens shaped into basic geometric forms—cones, spheres, disks, and cylinders—to function as physical counters for specific goods: a small cone might represent a specific measure of grain, while a sphere represented a larger measure. For thousands of years, this tactile, three-dimensional system provided a primitive but remarkably effective means of resource tracking.

Around 3500 BCE, someone hit upon the idea of enclosing these tokens within hollow clay envelopes called bullae. To allow an observer to verify the contents without breaking the seal, the citizens of Uruk began pressing the tokens into the wet clay exterior of the envelope before it dried. It was a clever security hack, but pressing those shapes into the clay also opened a new door in the adjacent possible. By transforming a physical, three-dimensional counter into a two-dimensional sign, these early accountants made a momentous step towards history’s inaugural information technology: writing.

Read more

In part two, I explored how declassified Cold War spy satellite imagery allowed archaeologist Jennifer Pournelle to discover that the first cities were actually born in lush, resource-rich wetlands, completely overturning the idea that early states arose to irrigate a desert. Coupled with Melinda Zeder’s insights into how animals like dogs and pigs essentially domesticated themselves by sharing our habitats, this evidence points to a 4,000-year “Wetlands Interregnum”—a stable, resilient period of abundance that pre-dates the emergence of agrarian states.

The sheer length of the Wetlands Interregnum—a span of roughly four thousand years, encompassing more than a hundred and fifty generations—raises an obvious question: why did such a massive chapter in the human story go untold for so long? Part of the answer lies in the geographical misunderstanding that Jennifer Pournelle’s satellite archaeology finally cleared up. For decades, researchers literally couldn’t see the marshes beneath the modern Iraqi desert, assuming that the earliest settlers were forced to build centralized states just to survive in a desert ecosystem. But the obscurity of this period also stems from an interpretative bias that has long plagued archeological reconstructions of the past. The technical term for it is “taphonomic bias” — taphonomy being the study of how organisms and human artifacts decay over time. The concept is a relatively simple one: archeologists naturally build their interpretations of human history out of artifacts that survive the passage of deep time. Because traditional archaeology was founded by scholars looking for the next Parthenon or Great Pyramid, the discipline was structurally predisposed to ignore cultures that built their worlds out of biodegradable materials. If your society is built of stone, baked clay, and gold, you leave an enduring footprint. If your society is built of wood, earth, vines, and reeds, you eventually vanish into the soil, leaving virtually nothing for 19th-century antiquarians to put in a museum.

The most glaring example of the taphonomic bias at work is the entire chronological framework we use for early human history—the Stone Age, Bronze Age, and Iron Age. That now ubiquitous timeline names eras after the inorganic materials that happened to survive in the dirt. In reality, early humans were masterful engineers of wood, bark, skin, woven fibers, and more. They built complex traps, weirs, baskets, and dwellings, but because these materials rot, our museums are filled only with their stone arrowheads and flint scrapers. We end up with this distorted image of our ancestors where we measure progress exclusively by how well they can do things with rocks. Weavers and woodworkers, thatchers and tanners—all their innovations vanish from the historical record.

The wetlands interregnum occupied a similar blind spot. There were no pyramids, no stone-carved law codes, and no vast, state-directed irrigation projects in this era. Instead of building with stone, the humans who lived in these settlements built with marsh reeds; instead of subsisting on an irrigated monoculture of wheat, they thrived on a diverse “hortipiscoral” diet of fish, waterfowl, dates, and legumes.

To borrow a concept that Scott developed to great effect in Seeing Like a State, the genius of these communities lay in their profound “illegibility.” Because their subsistence was spread across a complex, shifting web of natural rhythms and common-property resources, it defied any simple system of central accounting, making it challenging for an aspiring elite to quantify, tax, or control. And what makes a society illegible to the tax collector also makes it illegible to the traditional archaeologist. Because this civilization was built from biodegradable materials and sustained by a fluctuating, amphibious delta, the physical evidence of this golden age was, quite literally, washed away by the tides, or silently reclaimed by the microbial life of the wetlands ecosystem.

One trace of it did manage to survive, though—not in ancient ruins but in scripture.

Read more

Over the past few weeks, I’ve been having a tremendous amount of fun curating the notebook for the new Planet of the Barbarians series. Notebook curation is a remarkably creative exercise in its own right: building out a comprehensive knowledge base on a specific topic, layering in scholarly papers, key quotes, and supplemental research—and then deciding how best to represent that knowledge through the studio artifacts like Audio Overviews and my new obsession, Slide Decks. The intellectual work that goes into it shares some properties with the kind of work you do when you are writing an essay or a chapter in a book: you have to assemble knowledge and figure out the best way to present it to others. But it also flexes different creative muscles, like art directing the Slide Deck presentations. To give just one example, in the latest addition to the Barbarians notebook, there’s an amazing deck about the history of the Corona spy satellite program, styled as though it were a collection of declassified documents from the CIA:

Curating is really the exact word for the work that goes into assembling these notebooks, given the existing association the word carries with museum curators—it feels like I am preparing an exhibit to be explored rather than writing an essay to be read in a linear fashion. I really encourage everyone to try it, even if you are just assembling a knowledge base for your own personal edification. (We are actively working on improving the publisher tools at NotebookLM for those of you who are interested in sharing your collections more widely—more on that as it develops over the next few months.)

In my post announcing the Barbarians series, I explained that while the essay series and notebook would only be available to paying subscribers at Adjacent Possible, I was planning on assembling a notebook for all subscribers based around my earlier series from 2022-2023 on creative workflows, which had previously been paywalled. But when I sat down to start porting over those old posts into NotebookLM, I realized that I actually had a much more extensive body of work, scattered across several sites, that covered the same general topic, including a collection of interviews with people like Liz Phair, Kevin Kelly, and Rebecca Skloot talking about their own creative workflows. So I ended up assembling a much more comprehensive notebook than I had originally planned, with about two dozen total sources. I think it’s maybe 30,000 words all in—roughly half the length of one of my books. In a way, it’s a continuation of the themes that I covered in Where Good Ideas Come From, just more explicitly focused on practical advice. I’ve jotted down a few more notes on the content below, but you can also just head over to NotebookLM and experience it yourself. The link is publicly available so feel free to share if you want.

The Architecture of Ideas notebook has a custom chat mode that instructs the AI to give you advice almost as a creativity coach would, drawing on the insights and stories in the assembled sources; you should be able to describe whatever you’re working on, and the model will devise novel instructions tailored to the needs of your project. But I’m sure there are many more inventive ways to engage with the notebook as well. (As with the Barbarians notebook, and indeed all our Featured Notebooks, there’s no way for you to create your own artifacts—hopefully that will change soon as we update the publisher tools.)

In one of the Slide Decks I created, Notebook came up with the title The Architecture of Ideas. I think it’s a pretty fitting description for the entire collection. While there is a little extra emphasis on the craft of writing in the notebook, in general the material is all about the intellectual infrastructure we build around ourselves to make discovery possible. In other words, how can we build the most effective work environment for creative thinking, in any field?

“Environment” is used broadly here. It encompasses the physical architecture of a workspace—like Dan Pink’s converted-garage office or the “bedroom composer” solitude Liz Phair seeks after midnight—but it also refers to the digital and conceptual infrastructures we inhabit: software stacks we use to manage our “outboard memory,” the specific guitar tunings used to trick the brain into new melodies, and even the social networks we cultivate to inject useful noise into our thinking. Whether it is a slip-box system, a years-long Spark File, or an AI-powered research tool like NotebookLM, these are all ways of designing a space where ideas can collide in unpredictable ways.

The notebook draws on three different threads of my work to give you a complete look at how ideas actually come together. The backbone of the collection is the original “Creative Workflows” series from Adjacent Possible. These essays move through the entire lifecycle of a project: how to capture those fragile “slow hunches” before they disappear, how to build a “serendipity engine” to force those ideas to collide, and how to use systems like Walter Benjamin’s “convolutes” to turn a mess of fragments into a coherent shape. I even get into the more physical side of the process, like the “thinking paths” Charles Darwin explored on foot while working on Origin Of Species.

I’ve also included a series of in-depth interviews I published at Medium with some of my favorite creative minds. You’ll find Dan Pink talking about his hyper-organized system of analog Redweld folders and his ritual of pruning his idea files every six months. There’s Rebecca Skloot explaining how she managed the dizzying complexity of The Immortal Life of Henrietta Lacks using a literal wall of index cards, and Liz Phair on why she captures raw melodic “noodles” on her phone in the middle of the night. It’s a great reminder that while the underlying principles are often the same, the actual tools we use can be delightfully idiosyncratic.

Lastly, I’ve pulled in pieces from an earlier series I wrote at Medium called The Writer’s Room that focused on the specific tactical habits and software stacks that make this work possible for writing. There’s a long piece about why I think standard word processors are actually comically unsuited for complex, long-form thinking, and why I prefer tools like Scrivener that let you treat text as moveable blocks. I also share some of my favorite defensive writing strategies—like the “Shadow Plot” (where you treat distractions as seeds for future books) and my “Don’t Look Back” rule, which is essentially a strategy for avoiding the trap of over-reading your own work, ensuring that when you finally sit down for a first full edit, you are able to approach the text with fresh eyes. And I’ve included an updated essay about using NotebookLM as a research platform that walks through the best practices for integrating with tools like Kindle books, ReadWise, and Google Keep.

If you’ve been following my work for a while, you may know one of my favorite quotes about the creative process, from Thomas Schelling: “One thing a person cannot do, no matter how rigorous his analysis or heroic his imagination, is to draw up a list of things that would never occur to him.” At the most elemental level, that is the goal of a creative work environment: it should help you have ideas that would have never occurred to you. I hope this notebook helps you build out your own “idea architecture”, and more importantly helps you stumble across a few gems that would have otherwise gone undiscovered. Let me know what you find there.

If you’d like to keep up with the Barbarians series, consider signing up for the paid edition of Adjacent Possible.

Subscribe now

In the meantime, here’s a preview of some of the Slide Decks from the Architecture of Ideas notebook:

Shortly after Karl Wittfogel published Oriental Despotism, the United States began secretly flying its Corona satellites over the Middle East to survey the Soviet Union’s interests in the region. Launched just two years after Sputnik, with the primary objective of spying on territory inside the USSR, the Corona program featured some brilliant technological hacks to transmit high-resolution imagery back to the ground. Remember this was seventy years before Starlink, fifty years before Google Earth; even digital photography was still a decade away. If you wanted to take a detailed photograph from space you had to shoot it on film—and you had to somehow retrieve those physical canisters of film from a satellite that had been permanently consigned to low-earth orbit.

To get around those limitations, the CIA, in partnership with the Air Force, devised an ingenious system that produced a genuinely new way of seeing the earth’s surface. The Corona program’s method for data transmission involved high-stakes mid-air acrobatics. Since digital transmission was impossible, the satellite had to physically eject its “film bucket”—a specialized recovery capsule—into the atmosphere. As the capsule plummeted toward Earth, it would deploy a parachute to slow its descent. A specialized Air Force C-119—known aptly as the “Flying Boxcar”—would then snag the parachute out of the sky using a giant grappling hook.

The technology inside the satellite was equally sophisticated. By the late 1960s, the KH-4 systems, as they were called, carried dual panoramic cameras—one facing forward and one backward—to provide a stereoscopic view of the terrain. This was an essential perceptual leap; by capturing the same location from two different angles, analysts could construct a 3D topographic model of the landscape. While early iterations had a resolution of about 25 feet, the later KH-4B models could resolve details as small as six feet.

The imagery generated by Corona gave the spies back at Langley a view that Wittfogel and his contemporaries could only dream of. The archeologists of the 50s and 60s were, in a sense, the victims of the Street View vantage point they were still restricted to; they could barely imagine the perspective that a stereoscopic Google Earth-like view could provide. They were looking at the parched, cracked earth at ground level, and projecting that aridity back six millennia, assuming the landscape had always been a desert in need of civil engineers. They saw the ruins of ancient canals and assumed that the state had constructed them out of necessity to build an artificial oasis in an otherwise desiccated landscape.

Presumably the CIA learned very little about the activities of the red menace from their flyovers of the Iraqi desert. But whatever knowledge they were able to extract from those 3D images remained limited to a few hundred analysts with top-security clearances. The Berlin wall fell, the cold war dissipated, but the images themselves stayed classified.

And then Al Gore got interested in the Corona data. As a senator in the early 90s, Gore began lobbying the intelligence community to repurpose their Cold War assets now that the Soviet threat had collapsed. Gore’s argument was that the CIA was sitting on the world’s most pristine baseline data for planetary health—a record of ecosystems captured before the most intensive decades of industrial degradation.

Gore’s ascension to the Vice Presidency in 1993 gave him newfound persuasive powers with the intelligence community, culminating in Executive Order 12951. Signed by Bill Clinton in February 1995, the order declassified more than 800,000 images collected between 1960 and 1972, directing the transfer of film canisters from the CIA to the National Archives and the U.S. Geological Survey (USGS). This migration was a massive physical undertaking: a handoff of miles of master film that had been sequestered in intelligence vaults for decades. It took eighteen months to execute, but by the late 1990s, the film reached the public USGS EROS Data Center in South Dakota.

For three decades the information in those satellite images had been restricted to the intelligence community, scrutinized for troop movements and missile silos and uranium enrichment sites.

But now it belonged to science.

Read more

In the sweat of thy face shalt thou eat bread.

So decrees the God of the Old Testament in the third chapter of Genesis, as he exiles Adam and Eve from Eden. We usually read the Fall as a moral parable—the introduction of mortal sin, the first cleavage between divine will and human action. But there is something puzzling about the specific punishment handed down to Adam. God curses the ground itself, warning that it will bring forth “thorns and thistles,” and sentences humanity to a diet of bread—a food universally associated with agricultural societies organized around cereal grains. As a punishment for tasting the forbidden fruit, God sentences humanity to a life of farming.

The specific nature of the curse aligns with a bleak consensus that has taken hold in anthropological (and Big History) circles over the last few decades. If you’ve read Jared Diamond’s Guns, Germs, and Steel or Yuval Noah Harari’s Sapiens, you know the contours of the argument: the transition to agriculture should not be seen as an ascent to a higher plane of civilization, but rather a colossal downgrade for the average human, at least for the first few thousand years after we adopted it. (I touched on agriculture’s devastating impact on life expectancy in my own book, Extra Life.) In this view, the farmer was not the protagonist of a new era of human progress, but a prisoner of an emerging labor-intensive regime, forced into a lifestyle of toil to produce a taxable monocrop that acted as a vector for zoonotic pathogens and created a terrifying new vulnerability to famine. Harari calls the agricultural revolution “history’s biggest fraud.” The author of Genesis, it seems, got there first.

But the “Fall” into agriculture turns out to be a much more nuanced—and jagged—story than the standard account would have it. Over the past three decades, scholars have identified a massive, unexplained gap in the timeline between the first domestication of crops and the emergence of the first agrarian states, complicating the narrative popularized by authors like Harari and Diamond. If we were expelled from the Garden, it appears we spent several thousand years loitering by the exit.

Strangely enough, the one critical clue that explained the mystery behind that gap wasn’t unearthed at a dig site, or analyzed with carbon dating—it fell out of the sky in the 1960s, ejected from a Cold War spy satellite.

Read more

We’re ringing in the new year here at Adjacent Possible with some exciting developments.

Starting next Monday, I’m going to publish a five-part essay called Planet Of The Barbarians that will be available to paid subscribers. (There’s more detail on the subject matter below, but suffice to say if you enjoyed Guns, Germs, and Steel or Sapiens, you’ll find it interesting.) I’ll release it in serial fashion over the next month or two. The full text will be somewhere on the order of 15,000 words -- longer than any magazine article I’ve ever published, but significantly shorter than a full-length book. It’s actually a great length for a nonfiction piece, the kind of mid-sized deep dive that once thrived in the age of the pamphlet or the standalone monograph, but which the modern economy of 800-word op-eds and 300-page hardcovers has rendered nearly extinct.

Subscribe now

Paid subscribers will be able to read the full five-part series, but there will be an additional perk that was not possible in the heyday of the pamphleteers: a NotebookLM version of the project that includes the essay text, along with additional source material, including key quotes from the books that influenced my thinking, original scholarly papers, and related reading from my own catalog. You won’t just be able to read the essay itself; you’ll be able to ask your own questions, explore aspects that I might have overlooked. And as we’ve done with our Featured Notebooks, I’ll curate a collection of Studio artifacts to present the ideas in other formats: audio overviews, slide decks, quizzes, and more.

Publishing new work that is only available for paid subscribers is long overdue for those of you who have been supporting Adjacent Possible since it launched in 2021. Because so many of my posts have revolved around my work on NotebookLM, I’ve been reluctant to put anything behind a paywall here over the past few years. But I also want non-paying subscribers to experience what this new format can generate, so I’m going to create a shared notebook for the previously paywalled “Designing A Workflow For Thinking” series that I published here between 2021 and 2022 that will be available to everyone. I’ll talk more about that in a subsequent post.

I think this is actually a really interesting model that I hope other authors will explore: publishing a polished long-form piece alongside a bundle of related knowledge that allows readers to explore and adapt the entire package themselves, thanks to the explanatory and multimodal transformations that AI now makes possible. If this goes well, I think it might be a new recurring structure for my longer-form writing, alongside magazine articles and full-length books. I like it as well because it suggests a model where AI platforms like NotebookLM actually give authors and publishers new revenue-generating products that help their readers engage more deeply with their work.

If nothing else, it will be an interesting exploration into the adjacent possible of writing in the age of AI that you all can help me navigate. For now, here’s an introduction to the new series, with a bit of the backstory about how I came to write it....

For the last ten years or so, one of the constants of my intellectual life has been an informal set of email threads (and occasional text messages) with three old friends that I’ve known since grade school or before. It’s been a persistent—if somewhat spiky—conversation: we’ll chat at irregular intervals, sharing interesting links to things we’ve read or written, and then some topic will catch fire and we’ll write twenty long emails to each other in the space of a week. At times, it can feel like an impromptu, asynchronous book club conversation that gets triggered by one of us reading a new work that captures our imagination. At other points, the threads just emerge spontaneously out of our shared interests. There are a handful of enduring disagreements and more than a few Groundhog Day moments where it seems like we are just replaying debates that we’ve had multiple times before. But in general I’ve found it to be one of the most generative idea spaces in my life: a place where I can work out new theories or explore topics that I might not feel equipped to opine on in a more public forum.

Many of those conversations, as you might imagine, have been oriented towards the future: when the Apple Vision Pro came out, we had a long debate about the future of spatial interfaces; books like Fully Automated Luxury Communism and Abundance prompted discussions about state capacity and Universal Basic Income; and of course the future of AI—its impact on the labor market, the possibility of sentient machines—has been a constant obsession even before I started working on NotebookLM. But just as many of the threads have debated the distant past, particularly the transition between our ancestral evolutionary environment of nomadic hunter-gatherers and the emergence of “modern” institutions like agriculture, cities, capital, and writing. An early trigger for this—almost a decade ago—had been the publication of Yuval Noah Harari’s Sapiens, which my friends had read enthusiastically, but which I had found underwhelming. In part I was disappointed with Sapiens because I had just published my book Wonderland, which argued for a foundational role for delight-driven curiosity in the human story that Harari had largely ignored. (My irritation with Harari may have also had something to do with the fact that he had sold approximately one billion more copies of his book than I had of mine.) The dawn-of-civilization theme remained active in our threads over the next few years; one member of the group was working on an inventive thesis about the origins of capital that often brought us back to the Bronze Age, and I was writing my book about longevity, Extra Life, which had a long section about the devastating impact that the adoption of agriculture had on human lifespan.

And then sometime in 2021, I read James C. Scott’s revisionist account of the transition to the agrarian state, Against The Grain. It was one of those rare books reorders your mental map of the past, in large part because Scott is such an adept synthesizer of recent discoveries about the actual sequence that led to the formation of the first cities and states. Compared to Harari’s more familiar account in Sapiens, Against The Grain seemed to me to offer a more original and provocative narrative—while at the same time managing to do a much more responsible job of acknowledging the foundational work of other researchers.

Inspired by our long-running debate over the origins of capital and cities, and by my enthusiasm for the framework Scott proposed, I decided to jot down an overview of how I had come to understand the history of that period, drawing heavily from the ideas I’d absorbed from Against The Grain. I worked at it sporadically for a few weeks, ultimately ending up with a few thousand words of semi-shorthand notes—a truly “brief history” of the birth of agriculture and states—that I wrote entirely for the purpose of sharing with my three friends.

And then I completely forgot about it.

Fast forward about four years. It’s after dinner a few months ago and I’m putting in some time on a long-term project of mine: trying to convert as much as possible of my past personal writing/note-taking so that I have an comprehensive collection of my thinking over the years in a single notebook at NotebookLM. I’ve already got the full text of all my books, all my blog posts and Adjacent Possible essays in a notebook—accompanied by all the quotations I’ve highlighted in books going back to about 1998. But I’m constantly trying to add to that corpus, to capture the full history of my interests in that one notebook. And so I’d decided to take on the surprisingly challenging project of exporting my old Apple Notes to a format that NotebookLM could read. A few days before, a random conversation with a seat-mate on a flight to California had turned me onto an app called Exporter that let you convert Notes to a collection of markdown files, which was an important first step. And so now I was going through those exported notes, trying to figure out the best way to concatenate them into a longer document so I didn’t have to dump 500 short notes as sources in my notebook.

And that’s when I stumbled on my “brief history of civilization” note, which had been just sitting there gathering dust for years. I’d never even managed to send it to my three friends. As I skimmed it, it was obvious that the ideas were still very rough, and the descriptions almost too abbreviated even for our private chat. And I had read a few books in the interim—like Graeber and Wengrow’s Dawn of Everything—that suggested some obvious additions to my write-up. But the main thing that jumped into my mind reading this rough outline was that something had fundamentally changed since I had written it in 2021: I could now explore—and even share—these ideas using NotebookLM. I could take my shorthand description of the birth of agriculture and trigger Deep Research queries off of each paragraph to expand or fact-check the argument; I could add all the passages I’d highlighted from Harari, Scott, Graeber and others to the notebook to help me turn these rough notes into something more professional, with direct quotes and citations. I didn’t want to write an entire book on the topic, but it seemed like a wonderful theme for a multi-part essay here at Adjacent Possible. And because we now have public notebooks, I could share not only the essays with Adjacent Possible subscribers, but also the notebook itself, the bundle of knowledge that I’d collected in my investigations.

But then something even more remarkable happened: as I started to do some exploratory research, branching off of some of my favorite sections of Against The Grain, I ended up stumbling across a remarkable story of archeological detective work, kind of a cross between Indiana Jones and a Tom Clancy thriller, a discovery that radically transformed the tidy narrative of agricultural transformation that Sapiens had presented—and that suggested a tantalizing, science-based link to the Biblical origin story of the Garden of Eden. The transition to agriculture was even more provocative than I had thought in my first reading of Against The Grain, and amazingly it involved everything from Cold-War-era spy technology to Al Gore’s early crusade against global warming.

Subscribe now

I’ll be sending out the first installment later this week, with a preview for all subscribers. In the meantime, I’ve asked NotebookLM to create a short visual pitch for the series, using our new Slide Decks feature. Enjoy!

Subscribe now

This seemed like a quiet, bureaucratic fix in a year of deafening political change. (It was no accident that George Eliot set the events of Middlemarch in the three years leading up to 1832.) And yet, that single mandate—born from a failed experiment in open, collaborative review the year before—would inadvertently invent the system that has governed the flow of expert knowledge for nearly two centuries: confidential peer review.

The confidential part of the process was, in fact, an afterthought. The original plan had been one of radical transparency. The author of that plan was the English polymath William Whewell, a Cambridge professor whose expertise ranged from moral philosophy to the physics of tides, and who would later coin the term “scientist” itself. A year earlier, Whewell had proposed a system of referee reports to address a distinctly modern-feeling problem: the growing deluge of submissions from an expanding scientific community was overwhelming the Society’s ability to vet them, leading to rising complaints about the quality of the work it was publishing. Whewell had hit upon the idea that expert reports on submitted papers should be signed and published alongside the originals, confirming the reliability and utility of the paper’s findings. The reports themselves, he argued, would be valuable contributions, “often more interesting than the memoirs themselves.”

But the experiment in open collaboration failed almost immediately. The first two referees appointed, Whewell himself and the mathematician John Lubbock, were sent a paper and found themselves in stark disagreement over its merits, struggling to draft a joint report they could both sign. The logistical burden and the unseemly prospect of publishing personal scientific disputes proved too much for the Council. So they pivoted. The written reports would remain, but they would be submitted individually and, most importantly, confidentially, for the Council’s eyes only. The open, collaborative ideal was replaced by a private, anonymous judgment.

And it stuck. Journals of peer-reviewed articles, alongside books, have survived as a central unit of knowledge for almost two hundred years. There have been other revolutions along the way, like the meta analysis or the review article, new ways of bundling scholarship designed to make it easier to see broad predictable patterns in the results, or test hypotheses against a wider array of data sets. There were technological innovations along the way as well, of course: the web introduced new ways of connecting articles via hyperlinks, search engines, particularly targeted ones like Google Scholar, enhanced our ability to find the research we needed to advance our investigations. But new containers for knowledge only emerge at rare intervals. The book, peer-reviewed paper, the meta-analysis, the hyperlinked web page—each represented a different answer to a few fundamental questions. What is the right package for this particular slice of human knowledge, given the existing publishing and communications platforms? Once you’ve gone through the trouble of conducting your research, what is the most effective way to get your findings into the minds of other people?

It should be obvious to anyone paying attention to developments with AI over the past three years that the timing seems right for a new atomic unit of knowledge to be born. And one of my favorite things about the NotebookLM project is that we’re able to float some trial balloons about what that next container might look like. This month, we’ve pushed that exploration forward along two distinct—but related—paths, by integrating Deep Research directly into Notebook and by launching a series of Featured Notebooks curated by the team at Google Research.

I suspect a number of you have used Deep Research either in its Gemini incarnation at Google, or in other variants at OpenAI. But for those of you who haven’t tried it yet, the basic idea is that you give Deep Research a topic or a complex question, and it scours the web, evaluates dozens of sources, and then writes a structured overview of what it has learned, effectively building a starter research brief on the fly. When Deep Research launched last year, I heard from a lot of people that it seemed like the first new AI tool to share a similar ethos with NotebookLM, using AI to effectively create a dream research assistant. The funny backstory of it is that I didn’t hear a word about Deep Research until it launched to the public, which gives you a sense of how big Google is as a company, even in the subset of it devoted to research-oriented software. But from the first time I used the feature, I knew we had to bring some version of it to NotebookLM.

As powerful as the original Deep Research was, the main output of the product was restricted to a structured report and a search-style list of links to the sources that the model consulted to generate the report. But because NotebookLM is designed to manage and explore hundreds of sources, in the version we launched this month the Deep Research report is only the beginning of the journey. In our integration, Deep Research gives you an overview all of the sources it found during its research phase, with annotated commentary explaining how each source relates to your original query. You can then choose to import some or all of the sources to the notebook, along with the report itself, which you can then explore or transform using the full suite of tools that Notebook offers: grounded chat with citations, Mind Maps, Audio/Video overviews, and much more.

I’ve been framing this in the admittedly lofty language of scholarship and peer review, but the truth is these tools are just as powerful when applied to the more personal research non-academics do. Imagine you’re planning a vacation to Italy, for instance. You could ask Deep Research to “create a ten-day itinerary for a family with young children, focused on food and history.” The system would scour the web for travel guides, Wikipedia pages on Italian history, museum opening times, and synthesize all that information into a detailed starting plan. And crucially, it would also give you all the source material it consulted, which you could then import into your notebook to refine your plan, find new restaurants, or create audio overviews to listen to as you travel to your next stop on the itinerary. You’re not just getting a list of links; you’re building a personalized, interactive travel guide, a knowledge base that you can consult and expand during the trip itself.

Deep Research is all about assembling the information you need from the external world, engineering the most relevant context for whatever task you are working on. The Google Research collaboration, on the other hand, is more about output. Every year the Google Research team publishes upwards of a thousand papers on a vast array of topics: the future of machine learning, breakthroughs in quantum computing, the application of AI to fighting wildfires or forecasting floods, and much more. Up until now, they had a few primary avenues for distributing that knowledge: publish the article in a scientific journal, share a PDF on arXiv, present it at a conference, or bundle up a few related papers and summarize them in a blog post. But those formats are fundamentally static in nature. The knowledge is locked inside a fixed container, a one-way broadcast from the author to the reader. You can read the paper in its entirety, of course, or search for keywords, but you can’t have a conversation with it. You can’t ask it to summarize a key finding, or explain a difficult concept in simpler terms, or connect its argument to a collection of other articles on related themes.

The Google Research notebooks take a different approach. Each notebook contains a curated collection of articles on a specific topic, effectively creating a knowledge base of Google’s best thinking on a series of compelling research questions: How do scientists link genetics to health? How can scientists know what’s in your genome? If you’re a specialist in these fields, you can read the original papers or ask nuanced questions in chat and advance your understanding of the latest developments. But these notebooks can also make these complex but important topics understandable to non-specialists or students, including two new visual modes that we just launched that that are both pretty dazzling: infographics and slides. More on them below, but here’s a taste from one of the Google research notebooks:

One interesting thing about these notebooks is that they add a few new levels of human editorial judgment and creativity to the process. You have the intellectual labor that went into the original research of course, but now you have the additional layer of curation included as well: which articles should be bundled together to create the most compelling knowledge base (not unlike curating a special issue of a scientific publication), but also the more novel act of deciding which studio artifacts will best convey the substance of the research to the widest possible audience.

As a platform for sharing research, what we’ve built so far at Notebook is obviously just the beginning, and right now we are only scratching the surface of potential partners to create these collections. (If all goes well, we might just have a notebook curated by the Royal Society itself in the next month or so.) These notebooks are dynamic in the sense that you can transform the sources they contain into many alternate forms: different languages, explanatory styles, media formats, levels of abstraction. But it’s not hard to imagine a future version where the sources themselves are updated as new information comes online, perhaps using some kind of automated descendant of the Deep Research tool. Future versions might be able to generate a complex meta-analysis of all the recent research on a topic that can be updated instantly as new publications become available. We could even come full circle and address the challenge that the Royal Society faced back in the early nineteenth century. Imagine an AI mediator that could synthesize disagreements between two referees (or two conflicting papers) and propose a consensus, avoiding the logistical hurdles that doomed the 1831 experiment. In fact, I suspect you could accomplish a surprising number of these things right now with some clever prompting. But the larger point is that once you invent a new container for knowledge, you inevitably make new kinds of knowledge possible. The peer-reviewed paper gave us an engine for scientific consensus-building that has served us well for almost two centuries. But it feels like we are on the cusp of a new framework that might be even more significant, a knowledge unit built from the ground up with AI. These new notebooks are our attempt to imagine what that future might look like.

To give you one last taste of what the format now enables, I took the text of this post and asked Notebook to generate a slide deck illustrating its main points, in a visual language that evoked the early 19th-century Royal Society context. I have a few quibbles, starting with the opening slide, which slightly overstates the significance of the 1832 peer review innovation. But as a general treatment of the argument and ideas, it’s pretty magical I think. This is one slide from it:

The entire presentation is here:

The Next Container For Knowledge

20.5MB ∙ PDF file

Download

Download

The most surprising insight, I think, revolves around the significance of the fundamental substrate of all modern notebooks until the digital age: paper. Before reading Allen's book, I had mostly thought about paper in terms of Gutenberg and the printing press democratizing knowledge and information by making it much easier to publish and read printed information in the form of books. But somehow I hadn't fully thought through the more intimate revolution that it enabled—a revolution of self-knowledge and private record-keeping, executed not with a printing press but with a pen.

The very idea of a “notebook”—a cheap, lightweight device for capturing stray thoughts—is a relatively recent invention, because for most of human history the materials for capturing those thoughts were anything but cheap and lightweight. Writing itself has been around for at least five thousand years, but for most of that long run, the surfaces people wrote on were comically ill-suited for the kind of casual jotting that we associate with notebooks. Before the arrival of paper, if you wanted to keep track of your thoughts, you had to turn to a whole menagerie of cumbersome technologies: tablets made of hinged wood and ivory, filled with beeswax that you could inscribe with a stylus; papyrus scrolls; or the most durable of the lot, parchment, made from scraped animal hides, which was so expensive to produce that a single notebook might require the skins of an entire flock of sheep.

It’s a historical connection that honestly I should have grasped earlier. A few years ago, in my book Wonderland, I wrote about the astonishing flourishing of intellectual and commercial life in Baghdad during the Abbasid Caliphate, starting around the year 800 CE. The city was arguably the greatest single innovation hub on the planet at the time, a place where you could find magically lifelike automatons, oil-based streetlights, and state-of-the-art aqueducts. In Wonderland, I'd written about the legendary House of Wisdom, which was a strange hybrid of library, translation bureau, and Bell Labs-style R&D lab, where classic works from the Greco-Roman age were translated and stored in what was almost certainly the most comprehensive library in the world at that time.

This explosion of intellectual energy didn't just appear out of thin air. The new ideas, I argued, were themselves flowing through the city along with silks and spices from the East, because Baghdad sat at the nexus of the global trade routes of the age. But I'd missed one of the key technological platforms that made all that creativity possible: the arrival of relatively cheap, mass-produced paper, an invention that had made its way west along the Silk Road from China.

The Abbasids embraced paper with an astonishing fervor, with their vast libraries and entire streets filled with booksellers, a good six centuries before Gutenberg. But these weren't "books" in the way we think of them today. The Abbasids had mastered the codex format—binding individual sheets of paper together, just like a modern paperback, which was a massive improvement over traditional scrolls. But they didn't have movable type. There was no way to mass-produce these texts. Every single copy had to be transcribed by hand, a painstaking process that made each volume a significant investment, even with the cheaper paper.

So if you think of the development of book publishing as a linear progression, there's an easy-to-imagine ladder that ascends from the invention of paper, to the invention of the codex form factor, to the invention of movable type, which ultimately produces the modern archetype of a mass-produced book, and then triggers all the secondary cultural effects that we associate with the post-Gutenberg era: the Reformation, the scientific revolution, commercial pornography, and more.

But there's another branching path on that evolutionary tree that Allen's account highlights: paper made casual, personal notetaking far easier and far cheaper. And that casual notetaking didn't just stay in Baghdad. By the 1200s, Italian merchants who traded with the Islamic world had recognized the superiority of paper over parchment. Paper factories, using mechanized water power, sprang up in towns like Fabriano, churning out thousands of pages a day at a fraction of the cost of animal hides. Think of Da Vinci's legendary notebooks from the 1400s where he developed so many of his ideas. As Allen puts it, those ideas were enabled by the cutting-edge "information technology" of paper.

If you've been a longtime reader of Adjacent Possible, or followed some of the things I've written about NotebookLM, you might imagine that this is all a big setup for the ultimate emergence of the commonplace book, the defining "tool for thought" of the Enlightenment. And you wouldn’t be wrong—that revolutionary tradition is a crucial branch on this particular evolutionary tree. But Allen’s book reminded me that there was another, equally significant, tradition of note-taking that was also enabled by cheap paper, one that would prove to be just as influential as the commonplace book, though in a completely different domain. The commonplace book was a tool for organizing ideas. The other tradition involved keeping a notebook where you organized your money—bookkeeping in a word.

Financial records are as old as writing itself, of course. For millennia, merchants and administrators had scratched out inventories and receipts on clay tablets or papyrus scrolls. But those older technologies were too clunky and expensive for day-to-day accounting. "Paper had another key advantage over parchment for financial record-keeping: ink soaks into paper, making it permanent," Allen writes. "Parchment, on the other hand, could be scraped clean and re-used, which opened the door to fraud. With cheap, secure paper notebooks at their disposal, the Italian merchants developed the cornerstone of modern accounting: the system of double-entry bookkeeping."

Now the cornerstone of all financial bookkeeping, double-entry’s innovation of recording every financial event in two ledgers (one reflecting a debit, the other a credit) allowed merchants to track the financial health of their businesses with unparalleled accuracy. We do not know if the method originated in the mind of a single visionary proto-accountant, or whether the idea emerged simultaneously in the minds of multiple entrepreneurs, or whether it was passed on by Islamic entrepreneurs who may have experimented with the technique centuries before. Whatever its roots, the technique first became commonplace in the trade capitals of Italy—Genoa, Venice, and Florence—as the merchants of the early Renaissance shared tips among themselves on how best to manage their finances. What makes the history of double-entry so fascinating is the simple fact that no one seems to have claimed ownership of the technique, despite its immense value to a capitalist enterprise. I wrote about the irony here briefly in Good Ideas: one of the essential instruments in the creation of modern capitalism appears to have been developed collectively, circulating through the liquid networks of Italy’s cities. Double-entry accounting made it far easier to keep track of what you owned, but no one owned double-entry accounting itself.

But for all its brilliance, double-entry accounting remained a kind of guild secret for more than two centuries, a technique passed down informally between merchants. It took another revolutionary information technology—the printing press—and an immensely talented polymath to turn the practice into a global standard. The polymath’s name was Luca Pacioli, a Franciscan friar and mathematician who was a close friend and collaborator of da Vinci. In 1494, Pacioli published a massive encyclopedia of mathematical knowledge, Summa de Arithmetica, written in vernacular Tuscan so it would be accessible to a wide audience. Buried inside its 600-plus pages was a section that provided the first published description of the double-entry system, making its principles clear to anyone with a basic education. Pacioli’s book became a bestseller, and soon “the Italian method,” as it was called, became the foundation for mercantile practice across Europe. It was history's first example of a genre that would come to dominate the bestseller lists: a how-to-succeed-in-business advice manual.

The history of paper and the notebook has a striking parallel to another story we told in How We Got To Now—the story of glass. For millennia, humans had known about the magnifying properties of curved glass, but had found very little practical use for it. It wasn't until the late 1200s that someone—working with the breakthrough technology of transparent glass that had been concocted by the glassmakers of Murano—had the brilliant idea of shaping glass into small discs to be worn as spectacles. Reading glasses were a revolutionary tool for thought, extending the intellectual life of scholars by decades. (And making it possible for them to continue to maintain their own notebooks, well after the onset of presbyopia.) But for three hundred years, this application of glass remained a tool for seeing things up close. It was only during the burst of creativity around the year 1600 that a second wave of innovators like Galileo turned those glass lenses around and pointed them at new worlds, most famously the moons of Jupiter.

With paper, you could make the case that the story runs in reverse. For its first few centuries in Europe, paper was the tool that helped you see the big picture. With a cheap supply of paper notebooks, you could build complex financial models of your business with double-entry bookkeeping; you could design a cathedral dome with countless preparatory sketches. But paper would increasingly be used to augment our private thoughts and memories: the commonplace book, the diary, the personal correspondence, the first draft of a novel. These were tools that helped us see ourselves.

All of which left me thinking that if I were to add a new chapter or episode to How We Got To Now, it would have to be about paper. Like all the other innovations profiled in HWGTN, paper's influence is now so pervasive we barely notice it. But it triggered multitudes of unsung secondary effects, thanks to the diarists and the accountants. No one can reasonably contest the idea that the printed page was one of the most powerful engines of change the world has ever seen. But I suspect we've been underselling the transformative power of the blank page.

One night, [Atkinson] took LSD and wandered out of his home in the Los Gatos hills. Staring into the vast collections of pixels that made up the night sky, he became reenergized, and decided to adopt some of the Magic Slate ideas into software that could run on the Mac.

He designed a program where information—text, video, audio—would be stored on virtual cards. These would link to each other. It was a vision that harkened back to a 1940s idea by scientist Vannevar Bush, which had been sharpened by a technologist named Ted Nelson, who called the linking technique “hypertext.” But it was Atkinson who made the software work for a popular computer. When he showed the program, called HyperCard, to Apple CEO John Sculley, the executive was blown away, and he asked Atkinson what he wanted for it. “I want it to ship,” Atkinson said. Sculley agreed to put it on every computer. HyperCard would become a forerunner of the World Wide Web, proof of the viability of the hyperlinking concept.

Atkinson and Hypercard were on my mind even before he died, thanks to a NotebookLM project I've been focusing on for the past six months, the first preview of which just went live a week ago. Hypercard always had an unusual positioning: it was both a personal productivity tool (organize your cooking recipes!) and a publishing medium (explore my physics 101 hypercard stack!) But in either case, Atkinson had hit on the idea that there could be a new way of bundling information, enhanced by the then revolutionary technology of hypertext. Atkinson thought the best metaphor for it was a stack of magic cards; ultimately what took off was a different kind of bundle: web pages and sites, bound together by links and eventually by search.

Now, with AI, a new kind of information bundle is possible. For most of its short history, we've described NotebookLM as a tool for understanding and exploring information for your projects. But we also think it could be a distribution platform, amplifying expert knowledge. Last week we rolled out a preview of that vision: Featured Notebooks.

In a way, all of this dates back to a conversation I had with Josh Woodward—who now runs Google Labs and the Gemini app—in late July of 2022, one of the very first I had in Mountain View after joining Labs. We were talking about an idea that had come to us through the work of Kevin Kelly: the concept of "intelligence as a service" that he'd written about many years before, AI that you could tap into on demand like electricity or water. I told Josh that I'd always thought that sounded like a fascinating prospect, though I’d struggled to imagine what it would look like in practice. But what we'd seen with language models, and particularly the source-grounded language models that we were starting to experiment with, had suddenly made it clear to us both how Kelly's intelligence-as-service might actually work, how it might actually give authors and publishers a new platform to share their wisdom with the world.

"We're going to be able to bottle up the knowledge of experts," I said at some point. "And then people will just have that expertise on tap."

Somehow the metaphor stuck: Knowledge bottles. At some point that fall, our then-designer Gabe Clapper created a mock for a future page where you could purchase and download expert collections of knowledge; we called it, irreverently, "the Bottleshop." (It looked quite a bit like the design we are launching today for Featured Notebooks, as it turns out.) I kept encountering the phrase in meetings over the next year, as it spread around Labs and other parts of Google. And every time I would say: love the concept but please don't lock in on "knowledge bottles" as the official name. It was just a passing metaphor!

Naming aside, the underlying premise only grew more compelling over time. General purpose models trained on aggregations of human information were incredibly useful, but imagine how much more useful they would be if they were guided by (or were guides to) knowledge that had been peer-reviewed, edited, researched by experts—knowledge that had a particular point of view. If you're looking for parenting advice, for example, you don't just want the average of all parenting advice across the internet; you want parenting advice from a specialist who you trust.

So you can think of the Featured Notebooks we are launching today as a preview of a potential future where there are thousands of expert-curated notebooks on all sorts of topics that you can add to your own private brain trust, to have the knowledge you need on tap. Our launch lineup includes longevity advice from legendary scientist Eric Topol, bestselling author of Super Agers, alongside expert analysis and predictions for the year 2025 from The World Ahead annual report by The Economist. You can also find an advice notebook based on bestselling author Arthur C. Brooks' "How to Build A Life" columns in The Atlantic, and a science fan’s guide to visiting Yellowstone National Park, complete with geological explanations and biodiversity insights. There's a notebook covering long-term trends in human wellbeing published by the University of Oxford-affiliated project, Our World in Data, and one offering science-backed parenting advice based on psychology professor Jacqueline Nesi’s popular Substack newsletter, Techno Sapiens. For financial analysts and market watchers, there's a notebook tracking the Q1 earnings reports from the top 50 public companies worldwide.

The best way to get a sense of what "bottled knowledge" actually feels like in practice is to open one of these notebooks and ask for advice: ask for a sample itinerary for a 3-day trip to Yellowstone focused on wildlife, or ask for advice on a making a mid-life career change in the "How To Build A Life" notebook from The Atlantic; or ask the experts at The Economist about how global economic trends might impact your industry. I think you'll find that the results are genuinely helpful, and maybe offer a hint of a new way of interacting with an author or publisher's work. Here's an example I really enjoyed from the Arthur Brooks notebook—it's like have an erudite advice columnist writing an essay just for you:

As you might imagine, I'm keenly interested in creating original content designed to be consumed in the notebook format. A curated notebook creates some pretty tantalizing new possibilities for sharing knowledge and expertise—and maybe even storytelling. Your audience can read the original texts in their entirety; they can ask questions or brainstorm ideas through a conversational interface. Students can ask for simpler explanations of complex topics, or listen to Audio Overviews or generate Study Guides to help them master the material. But experts can ask more challenging questions. And all of these interactions can unfold in over 80 languages, no matter what language the original sources were written in.

Imagine all my collected writing on innovation bundled up as a kind of advice notebook that you could consult to help you think through your own creative challenges; or an historical account like “The Man Who Broke The World”, or Enemy Of All Mankind, bundled with all the primary sources and related reading behind the story, designed to be consumed as a linear textual narrative, or an audio overview, or as one of the new formats that we develop over the coming months and years. There's an interesting future for Adjacent Possible here as well; I'd love to be able to offer special notebooks that I curate on a range of topics as one of the perks of a paid subscription.

In the blog post announcing the featured notebooks, there's a wonderful quote from Nick Thompson, CEO of The Atlantic, that made me think of Atkinson and Hypercard all over again: "The books of the future won’t just be static," he said. "Some will talk to you, some will evolve with you, and some will exist in forms we can’t imagine now." Featured Notebooks are a glimpse of that possible future.

One last thing. You'll note that in the Steven Levy obit for Atkinson that we began with, the role of Apple CEO is played by John Sculley, not Steve Jobs. (Jobs had been fired two years before Hypercard's launch.) In 1988, a year after my original indoctrination into tools for thought via Hypercard, Jobs announced his second act as an entrepreneur: the NeXT computer. NeXT had a groundbreaking operating system and graphic interface that would eventually become the basis for OS X, after Jobs' legendary return to the company several years later. But I remember being dazzled (and intrigued) by another part of the announcement: the computer featured a high-capacity optical drive that allowed them to bundle an amazing collection of digital resources on the device: Webster's Ninth New Collegiate Dictionary, Merriam-Webster's Collegiate Thesaurus, and The Oxford Dictionary of Quotations. But the item that really caught my eye was a surprisingly literary addition: the complete works of Shakespeare.

So a few decades later, it seemed right to pay tribute to that history and include a featured notebook with all of Shakespeare's work in our launch lineup. You can read the plays in their entirety; you can ask for help making sense of the Elizabethan English; you can find memorable quotes on just about any topic drawing from all the plays; you can listen to Audio Overviews or review Study Guides to deepen your understanding of Shakespeare’s work. It's pretty magical.

As I was putting together this post, I tracked down the original NeXT brochure, which had this description of the "Digital Librarian" software that came bundled with the computer:

The NeXT System also makes it easy to create your own libraries. The Digital Librarian's cataloging function lets you enter large amounts of information, while automatically creating an index of key words for you. This index then allows you to search your own information - a new "book" in the Digital Library - the same way you would search the dictionary.

As for the kinds of Digital Libraries you can create, there are no limitations. Professors, for example, might build libraries containing a history of their own collected writings. A legal or medical office might construct a library of often used reference materials, while a business might build a reference library of contracts and forms.

Reading that made me think of maybe my favorite magic trick with these new featured notebooks: generating a Mind Map based on all of Shakespeare's writing. In less than a minute, NotebookLM analyzes all the themes and major characters across the plays and sonnets, and organizes them into what is effectively a dynamic, nested index of Shakespeare's entire body of work. And clicking on any of those nodes then takes you directly to an in-depth discussion of how Shakespeare's writing relates to the concept, complete with citations back to the original lines.

There's something truly moving for me, navigating through that Featured Notebook—all this extraordinary technology, inspired by computing legends like Atkinson and Jobs, now being harnessed to map the creative mind of a playwright who died four centuries ago. What would Shakespeare have thought, seeing his work being explored in this way? I honestly have no idea. But I like to think that Atkinson and Jobs would have approved.

Last Sunday's edition of The New York Times Magazine was a special issue devoted to answering a particularly helpful question: what is AI actually useful for today? Put aside the long-term debate about radical job loss and superintelligence; we've had more than two years of consumer-grade AI applications, which should be plenty of time for them to prove their merits as functional tools, even if the future may lead to more ambitious applications down the line. So where is AI actually helping people right now?

The issue features a long, probing meditation on how AI is beginning to transform the way we analyze and write about history, written by my long-time editor Bill Wasik. The piece is bookended with a case study of my use of NotebookLM to investigate early ideas for a project about the Gold Rush that I first wrote about here at Adjacent Possible last year. Here's one key section:

Johnson showed me the results of his experiments so far. He started his brainstorming process by giving NotebookLM excerpts from one of the finest existing histories on the Gold Rush, H.W. Brands’s “The Age of Gold.” He thought he might want to focus on the conflict between white gold-seekers and the Native American groups living in the Yosemite Valley in the 1850s, so he uploaded the text of an older source called “Discovery of the Yosemite,” by Lafayette Houghton Bunnell, who was part of the Mariposa Battalion, the militia unit that rode into the valley in 1851. Next, to bring in the Indigenous perspective, he went to public-domain websites and found two accounts about the people whom the battalion expelled from the valley: “The Ahwahneechees: A Story of the Yosemite Indians” and “Indians of the Yosemite Valley and Vicinity.”

Johnson started his conversation with NotebookLM with a little orientation, identifying himself as the author Steven Johnson, so that the A.I. (whose training allows it to understand almost the whole internet, as ChatGPT does, even if it constrains its answers to the uploaded sources) might get a sense for the kinds of books he writes. Then he started peppering it with questions: What in the two sources that focused on the Indigenous experience, he asked, was missing from the other two sources? When the model returned its summary, his eye was caught by the observation that “The Ahwahneechees,” by including short biographies of individual Yosemite Indians, “helps to humanize the people beyond being a tribal mass, which is something that ‘The Age of Gold’ and Bunnell’s book tend to do.” The A.I. listed some of their names, among them Maria Lebrado, granddaughter of the Yosemite chief Teneiya.

That piqued Johnson’s interest. He asked for more information about Lebrado, and the tool returned nearly 600 words of biography — that she was one of the 72 Native people forced to leave the valley by the battalion in March 1851; that she eventually married a Mexican man who ran a pack train down in the Central Valley; that she was “discovered” by a white historian in the 1920s and held up as the last of the original Yosemite Indians.

Right away, Johnson recognized that she would make a great character. He took note in particular of the fact that Lebrado returned to the valley near the end of her life. “I’m like, What’s the [expletive] structure of ‘Titanic’?” he joked. The book could open with what Johnson imagined was Lebrado’s emotional return to the valley at nearly 90 years old, before zooming back in time — to her childhood, to a broader cast of characters and the violent drama of the 1850s.

The essay does a great job of capturing the kind of collaborative brainstorming you can now do with an AI that is grounded in sources that you have personally curated. Unless you have experienced it firsthand, it’s hard to explain how different it is from just riffing with a general-purpose model where the knowledge base is not a specific collection of works that you have assembled. (The one thing that I wish had been clearer in the Times essay is that I had already read the Brand and Bunnell books in their entirety—the "excerpts" from Brand mentioned are specific passages that I had highlighted while reading and imported to Notebook using ReadWise.) So much of the early stage work for a book is really about probing vast archives of information, trying to see if there's a new angle that adds to the existing accounts. I didn't know for sure whether Maria Lebrado would in fact work as a major character after this initial line of inquiry—I'm still not sure, in fact—but discovering her story gave me something to anchor on as I read through the material, and suggested a potential structure for the book.

There's another point in the piece where I ask for Notebook's help filling out the details of a slightly wacky structure that had occurred to me later in the research: a variation of my "long zoom" idea where the first chapter would begin a million years before the Gold Rush, and then get progressively closer to the main event: 100,000 years, 10,000 years, and so on. I'd asked Notebook to help me flesh out what actual events would go in each chapter, based on the existing timeline in the sources I'd collected. It's another variation of this probing exercise: if I structured it this way, how would it work in practice? I find this kind of exercise amazingly valuable, assuming I have done two key things first: 1) explained to Notebook who I am and what kind of book I'm trying to write, and 2) curated a unique set of sources that allows the AI to explore a new configuration of ideas, and not just resort to its typical "average" response. Again, it's not giving me a final, definitive outline -- it's helping me quickly draft possibilities. If there's a problem with the existing state of the models right now for this kind of task, it's that they are tuned to be a little too supportive; you have to deliberately ask them to be more skeptical if you want constructive criticism, and I don't yet trust them to make an independent value judgment about the merits of a specific approach. When I'm in this kind of exploratory mode, I mostly just ask Notebook to help me generate possibilities, and then I evaluate what works and what doesn't work on my own.

***

The Times piece mentions a few other authors who have been using AI in comparable ways, as well as a few who are less intrigued by the approach. There's a brilliant line from Stacy Schiff, author of acclaimed biographies of Cleopatra and Véra Nabokov: “To turn to A.I. for structure seems less like a cheat than a deprivation, like enlisting someone to eat your hot fudge sundae for you.” On some level, I completely agree with Schiff here; longtime Adjacent Possible readers might remember that I wrote an entire post about how much I love plotting out the structure for my books. But it's not just about the pleasure of inventing structures; it's actually one of the things that I think I do best as a writer. So it would be completely illogical for me to simply outsource that work to the AI. What Notebook lets me do is experiment more freely, explore different hypotheses, or fill in blank spots that I've either forgotten or not yet discovered. I'm still eating the sundae, in Schiff's metaphor -- it's just tastier with Notebook helping me assemble the ingredients.

There's a general point worth making here, which is that in the Gold Rush research example, all the key ideas that are being generated are either coming from me, or from the human-authored works that I have collected. NotebookLM is effectively functioning as a conduit between my knowledge/creativity and the knowledge stored in the source material: stress-testing speculative ideas I have, fact-checking, helping me see patterns in the material, reminding me of things that I read but have forgotten. It's an important point to make in part because there seems to be a growing concern about offloading the actual research/reading phase to AI-generated summaries. I recently listened to a terrific discussion between Ezra Klein and Dave Perell (two people whose insights on AI-assisted writing and thinking I greatly admire) where Klein notes that he has found little value for AI in his research workflow so far:

I think I used to conceptualize knowledge the way you see it in the movie The Matrix, where it's like I wanted the port in the back of my mind that the little needle would go into. I had read John Rawls's Political Liberalism. I thought that what you were doing was downloading information into your brain.

Now I think that what you are doing is spending time grappling with the text, making connections. It will only happen through that process of grappling. So, the idea that you could speed run that, the idea that it could just be summarized for you...

Part of what is happening when you spend seven hours reading a book is you spend seven hours with your mind on this topic. The idea that O3 can summarize it for you, in addition to all this stuff you just will not have read, is that you didn't have the engagement. It doesn't impress itself upon you. It doesn't change you. What knowledge is supposed to do is change you, and it changes you because you make connections to it.

I do agree with Ezra here. You should read the things that you need to understand deeply. The time spent engaged in reading is truly transformative. There is no replacement for reading the core primary and secondary texts if you are writing nonfiction journalism or history. Generally, I don’t think of NotebookLM as doing the reading for me; it’s more like Notebook is doing the reading with me.

But in the Gold Rush example, I'm using Notebook not as a replacement for reading, but rather as a tool to help me figure out what is worth reading. As Ezra notes in his conversation with David, you can't read everything. And you particularly can't read everything when you're still trying to figure out what your next book topic should be. Anyone who works with research in one form or another relies on some version of summarization or excerpting to decide whether to read a given work: skimming the jacket copy, or stumbling across a provocative quote in another text. As Bill Wasik points out in the Times essay, the invention of the index was a great boon to scholarship, because sometimes scholars are looking for targeted information that doesn’t require ten hours of engagement: a tool like NotebookLM’s Mind Maps can effectively build you an index on the fly for dozens of documents that you’ve curated. That's helpful if you use it the right way, at the right time. When I load in those sources about the Native Americans of the Yosemite Valley, and ask Notebook to compare them to the existing sources I've already read, it's almost like I'm using the AI to generate a bespoke jacket copy, summarizing the text as it relates to the existing research and thinking I've done on the topic.

If tools like NotebookLM end up being primarily used as a replacement for reading the core sources you need to generate a deep, engaged understanding, that will indeed be a step backwards, and will produce less nuanced, less compelling works of scholarship. Avoiding that scenario is one of the reasons why from the very beginning NotebookLM has been designed to facilitate the reading of original sources, in ways that set us apart from nearly every other AI product out there: the entire text of each source is always readable in the app, and Notebook gives you inline citations that take you directly back to the relevant passages, so you are always one click away from the original material.

That said, there are many forms of knowledge work where skimming/excerpting/summarizing can be incredibly valuable, and more appropriate than reading a document from start to finish. You don’t necessarily want to read your entire car manual; you just want to be able to ask a specific question and get an accurate answer. Certain kinds of advice books or reference works could arguably be better explored through a conversational Q&A format as opposed to linear reading. Using Ezra's metaphor, some forms of information transfer are in fact more Matrix-like. We've tried to design NotebookLM to accommodate both modes.

At another point, Ezra says: "Having AI summarize a book or a paper for me is a disaster. It has no idea what I really wanted to know. It would not have made the connections I would have made."

If Ezra is experiencing this limitation in his use of AI tools, I would argue that it is a failure of UI not AI -- in other words, the user interface is not allowing him to teach the model about his own particular interests and sensibility, the way I do at the beginning of the Gold Rush exploration in the Times piece. You’ll often hear people say that the big risk for scholarship with these tools is that they eliminate the kind of serendipitous discovery that you can only get from analog research. (The same argument was made during the transition from library stacks to digital search.) I encourage anyone who still thinks this to try an experiment: create a notebook with your initial drafts of a project you are working on and then add a dozen or so new sources that are vaguely in the same space. (You can do this easily with Notebook’s new "Discover Sources" feature.) And then just ask Notebook to suggest ten surprising, less obvious connections between your original writing and the new sources. And then go back and read the new source material. In my experience, that’s a far more effective way to uncover serendipitous links than traditional reading without an AI collaborator. And a thousand times more effective than walking around library stacks, which I am old enough to remember doing!

***

One last teaser: near the end of the Times piece, I start speculating on what NotebookLM might look like as a distribution platform for information, and not just a productivity tool for creating that information:

In our conversation in Mountain View, [Johnson] put forward a possible new revenue stream: What if e-books of history came enhanced with a NotebookLM-like interface? Imagine, he went on, that “there’s a linear version of the story with chapters,” but then the primary materials the author used to write the book also come bundled with it. That way, “instead of just a bibliography, you have a live collection of all the original sources” for a chatbot to explore: delivering timelines, “mind maps,” explanations of key themes, anything you can think to ask.

It is perhaps the most brain-breaking vision of A.I. history, in which an intelligent agent helps you write a book about the past and then stays attached to that book into the indefinite future, forever helping your audience to interpret it.

I was deliberately vague about what that might mean in practice when I said those words to Bill Wasik earlier this year. But I'll have something much more solid to share on that front in a few weeks—including some news about how all these developments are going to impact Adjacent Possible. More to come after the July 4 holiday...

I was savoring all of it as we walked up out of the subway, despite the crowds and the chaos. I’ve had a couple of lucky moments over the years where I’ve come to midtown to cross some interesting milestone in my career—I remember feeling a similar shiver taking a cab up to do The Daily Show for the first time, for my book Everything Bad Is Good For You. I spent more hours than I should probably admit as a sixteen-year-old in suburban DC daydreaming about making it as a writer in New York City. So I always try to stop for a second when I find myself in a moment like last Thursday night and quietly think to myself: sixteen-year-old Steven would have approved.

It was all very self-congratulatory, for sure, but genuinely tempered by the conviction that I had formed over the preceding 24 hours that I was almost certainly not going to win. The competition was too impressive, and I worried that Infernal didn’t quite fit the genre conventions of the true crime category. (My main concern, ironically, was that there were too many crimes in the book.) Before my wife and I left our apartment, I’d briefly discussed my odds with our oldest son Clay, who had been my research assistant on the book. I told him that I had downgraded my expectations, and he seemed surprisingly relieved. “I looked at those other books in the category,” he volunteered helpfully, “and I was kinda wondering why you thought you had a chance.”

But then somehow, against all the odds set by the Johnson family bookies, when they actually opened the envelope, the first three words spoken were: the infernal machine.

It’s been almost exactly a year since Infernal Machine came out—the paperback edition just launched last week as it happens—and there are many new faces here at Adjacent Possible since I last posted about the book. So allow me a quick précis of the book for the newcomers who haven’t had a chance to read it. The Infernal Machine is the fourth one of mine that tells a (mostly) linear historical narrative, though in this case it’s really three interwoven narratives, spanning about half a century. One thread is Alfred Nobel’s invention of dynamite, one of the great case studies in the unintended consequences of technological and scientific innovation. Another is the rise of anarchism as a major revolutionary movement, in Europe and the United States, and that movement’s embrace of terrorism and political violence as a semi-normalized strategy for advancing its cause, much of it enabled by the portable and potent weapon of Nobel’s dynamite. And finally, the book chronicles the rise of modern policing and surveillance systems—the whole notion of crimefighting as an information science that culminates in Hoover’s FBI—which was in many ways set in motion by the wave of terror unleashed by the anarchists.

Infernal touches on a number of themes that I have long been interested in, themes that make regular appearances in other books of mine: unanticipated secondary effects of new technologies (like the story about air conditioning leading to Ronald Reagan’s election in How We Got To Now); the oft-overlooked impact of data management tools in shaping society (William Farr and John Snow inventing epidemiology in The Ghost Map and Extra Life); the way rogue groups working outside the law end up shaping much more formidable institutions (as in the interaction between the pirates and the East India Company in Enemy of All Mankind.) But last week after the Edgars, as I took the subway home with my wife, clutching my porcelain replica of Edgar Allen Poe, I was thinking about a property of Infernal Machine that made it different from my other books: despite the fact that I have lived in Brooklyn or Manhattan for the vast majority of my adult life, Infernal is the first book I’ve written that is truly set in New York City. I’d actually written significantly more about London than I had about my own home town, and never written anything where New York itself was a central character in the story.

There are so many different slices of New York in Infernal that over the weekend I sat down at my desk and opened up a NotebookLM notebook I have with the full manuscript and asked for an overview of all the locations that play an important role in the book, with quick summaries of the plot events and characters associated with each. (Notebook as you might imagine is incredibly good at re-organizing the events of a book along a different axis like this.) Skimming the list was a great reminder of just how fractal a metropolis like New York can be, as you move from the greater metropolitan area to the boroughs down to the neighborhoods and then the micro-communities that can sometimes shift over the course of a few blocks. There’s the hub around the police headquarters—initially in eastern Soho, then further downtown at Centre street—that I remember first being captivated by reading Caleb Carr’s The Alienist in my twenties. That’s where the book’s nerd-detective hero Joseph Faurot sets up his “Identification Bureau” using the suspiciously European techniques of fingerprinting and other forensic sciences to solve crimes for the first time. Adjacent to that nexus of law enforcement was, of course, the Lower East Side, the primary gateway and settlement area for successive waves of immigrants in the late 19th and early 20th centuries. It had been initially dubbed "Kleindeutschland" (Little Germany), but by the time Emma Goldman and Alexander Berkman—the central anarchist figures in the book—arrived, it was dominated by Russian Jews. The crowding was almost unthinkable. In modern-day Manhattan, where residences can be efficiently stacked upwards in vertical skyscrapers, the average population density is roughly 100 people per acre. When Emma Goldman first strolled through the streets of what was then called the Tenth Ward, the low-rise buildings around her somehow accommodated a population density five times greater than that.

And then, only a few stops on the subway north of the LES, there’s "Millionaire's Row,” the stretch of enormous mansions constructed by the industrialists of the Gilded Age, where Goldman strolls with her friend Modska Aronstam, a sensitive illustrator and sometime radical, apparently part of a “thruple” with Berkman and Goldman. (When Goldman points out the scandalous contrast between the midtown mansions and the tenement squalor just south of them, Aronstam counters: “My main objection is that they have such bad taste—those buildings are ugly.”) A few blocks away, a daring undercover operation by a young Italian detective named Amedeo Polignani foils a plot to bomb the newly-consecrated St. Patrick’s Cathedral, a plot that had been hatched in an anarchist circle in East Harlem, where Goldman and Berkman set up shop in the 1910s.

The “outer boroughs” feature prominently as well. One of my favorite sequences in the book takes place in the Long Meadow of Brooklyn’s Prospect Park, where the young lovers Goldman and Berkman romance each other discussing the revolutionary valor of Czar Alexander II’s assassins. Even Staten Island gets a visit: while plotting his (ultimately failed) assassination attempt on the industrialist Henry Clay Frick, Berkman decamps to the still mostly rural island to test the dynamite bombs he’s constructed for the job. (He ultimately opts for a revolver.)

I hadn’t thought about it until I compiled this geographic survey of the narrative, but the final pages of the book complete the NYC tour by taking the reader out onto the islands in New York Harbor, where Goldman has a fateful encounter with J. Edgar Hoover on Ellis Island, and then out through the tidal strait of the Narrows onboard the “Red Ark” that carries Goldman and Berkman along with 247 other deported radicals to Soviet Russia in 1919. “The tall skyscrapers, their outlines dimmed, looked like fairy castles lit by winking stars,” Berkman later wrote of that final exodus, “and then all was swallowed in the distance.”

One reason that litany is so magical to me is that almost all those places conjure up very specific memories in my head, independent of their narrative significance in The Infernal Machine. I walked by the old police headquarters on my way to work every day in my twenties when I was editing the web magazine FEED; we must have spent more than a thousand hours in the Long Meadow when our kids were young and needed space to roam. Many of those places have changed in the century that has passed since Berkman and Goldman sailed out through the Narrows. The police building has been converted into luxury condos; Ellis Island is a museum; Grand Central Palace—the enormous exhibition space straddling the rail lines running north of Grand Central Station where Berkman delivers an incendiary speech after returning from prison—no longer exists.

But despite those changes, that older New York is still very much imaginable—even walking through the permanent daylight of Times Square and its animated LED Megatron billboards. What is harder for my mind to imagine is the relentless violence in the streets of the city during that period, much of it political. Infernal Machine describes dozens of bombings and assassinations that took place in the first two decades of the twentieth century, but the total number was almost incomprehensibly larger: the city’s chief bomb inspector either defused or surveyed the wreckage from seven thousand bombs over his thirty year career. (Hence my worry that Infernal had too many crimes to fit the classic true crime genre—it felt almost closer to military history.) The sheer scale of that carnage was, of course, one of the main reasons I got interested in writing the book in the first place, as a study in how political violence becomes normalized in a society. I wrote it as a work of history, a reminder of how much has changed over the last century, and not as a warning about potential carnage to come. Here’s hoping it stays that way.

Simon’s most famous line is this: "In an information-rich world, the wealth of information means a dearth of something else: a scarcity of whatever it is that information consumes. What information consumes is rather obvious: it consumes the attention of its recipients.” But there was another quote from Simon that Hayes surfaced that caught my eye:

What makes a given information processing system useful to an organization isn’t how much information it generates or even the raw amount of information it can process. Rather, [Simon argues], “the crucial question is how much information it will allow to be withheld from the attention of other parts of the system… To be an attention conserver for an organization, an information-processing system must be an information condenser.”

The whole idea of an “information condenser” struck a chord with me, because I’d just had a conversation with my friend Rebecca Skloot—author of the classic The Immortal Life of Henrietta Lacks—about the research and writing workflow involved in writing nonfiction history; borrowing a metaphor from Hemingway, she’d described the research process as being like constructing this giant iceberg of information, where you know in the end only a small bit of the entire complex will be above the surface, visible to the reader.

This was not the first time I’d discussed this issue with Rebecca. Years ago I interviewed her for a Medium series on the writer’s workflow in a piece I called “Rebecca Skloot has too much information.” It included this incredible picture of her midway through the process of researching Immortal Life:

So much of what I do as a writer comes down to figuring out the most clever and accessible way of condensing an iceberg of source material into a much smaller format. In this case, I’m condensing a book by Chris Hayes, a conversation with a Rebecca, a lecture by Herbert Simon, and, as we will see, a Saturday Night Live sketch—all into a Substack post that you can read on the subway. In the case of one of Rebecca’s books, it’s condensing ten years of research and thinking—hundreds of newspaper articles, scientific papers, diary entries, interviews, notes—into a compelling three-hundred-page narrative. What is worth saving in that enormous pile? What tiny sample of it best represents the whole? A huge percentage of my thinking time as a writer is spent answering questions like those.

That all sounds laudable enough, but the problem is—and here we get to the central argument of Hayes’ new book—there’s another modern invention beyond “nonfiction authors” that also provides this compression service: your phone.

If I had to guess the distribution of the average modern human’s information interests I would say it was something like: they want top-level news about the world, pop culture, sports, maybe politics, with the ability to zoom in if something momentous happens; and fine-grained real-time news from their family, friends, and work colleagues. That’s the ideal tip of the iceberg. But of course the total amount of information generated by all those domains is enormous: trillions of bits of information every second. Every medium compresses that information in different ways. MSNBC gives you world news and politics, but has absolutely no idea what is going on with your friends and family. The team Slack channel at work gives you a different slice.

Now imagine some kind of metric like how closely does this medium match the information compression needs of its audience? That seems like an honorable enough mission, even considering that humans often have bad judgement about what information would most benefit them. But my suspicion is if you actually tried to do the user research on that question, you’d find that there was one clear winner: the smartphone. The phone—conveying social media in all its forms, text messages, Slack channels, Twitter—gives us that “average human” package of information more effectively than any communications medium before it.

I sometimes feel like this property is underplayed in accounts of how the smartphone is transforming our culture. The phone is not just capturing our attention because “surveillance capitalism” has manipulated our judgment. Our phones are addictive because they’re interesting. They’re addictive because they do a good job of condensing the iceberg.

In The Sirens’ Call Hayes presents that addictive force—the command the phone has over our attention—as one of the defining properties of our era. He begins with this lovely concatenation of two traditional meanings of the word ‘siren':

The ambulance siren can be a nuisance in a loud, crowded city streetscape, but at least it compels our attention for a socially useful purpose. The Sirens of Greek myth compel our attention to speed our own death. What Odysseus was doing with the wax and the mast was actively trying to manage his own attention. As dramatic as that Homeric passage is, it’s also, for us in the attention age, almost mundane. Because to live at this moment in the world, both online and off, is to find oneself endlessly wriggling on the mast, fighting for control of our very being against the ceaseless siren calls of the people and devices and corporations and malevolent actors trying to trap it.

Hayes’ book is extremely stimulating, in the best way; even in sections where I felt like I had done a lot of thinking on the subject matter already, he consistently produces some novel twist or new way of framing the issue that had not occurred to me before. He has an amazing section where he walks through the distinction between “grabbing” and “holding” attention that should be required reading in every media studies class. One example of this in my own experience: my biggest complaint about Twitter under Musk is not the prominence of right-wing content, but rather the increased emphasis on TikTok-style short video content, which tends to “grab” my attention for the simple reason that moving images attract the eye more than text. Because those videos usually have little to do with my range of interests, I find that Twitter performs worse as an information condenser than it did pre-Musk.

But even with that downgrade, the combination of Twitter and text messages does an admirably good job of presenting me with a quick overview of what I need to know at any given moment in time: breaking news and commentary from both the public and private sphere. This is why I tend to find it a little grating when critics compare the social media platforms to, say, the cigarette companies, claiming that they’ve “manipulated the algorithm” to give us an addictive dopamine hit. (Hayes, to his credit, doesn’t generally take it this far.) One of dopamine’s key roles is to alert your brain to surprising developments, situations where the outcomes deviate from your predictions, where something novel appears to be happening. Dopamine effectively converts novelty and surprise into interest and attention. Drugs like nicotine or caffeine make things more interesting by creating artificial supply of dopamine molecules in your brain. But social media does it by supplying actually interesting things. Yes, most of these companies have advertising-based models that compel them to devise ever more elaborate mechanisms to capture your attention. But the core principle behind almost every social media algorithm—”condense the iceberg by showing me things that my friends liked”—is not merely a device to show you more ads. It’s actually a useful initial filter.

But like everything that changes the way we pay attention to things, that filter turned out to have some unexpected costs. One of those costs is that it gets harder to dive beneath the fast-running surface waters of the newsfeed. The ideal information compressor lets you skim and plunge, to slightly modify an old concept from David Gelernter. Twitter in its heyday had a hint of this: tweets themselves were only 140 characters, but they often carried links to 5,000-word New Yorker essays, or entire books. But there’s no use denying that it gets harder and harder to build up the cognitive muscles you need to read and understand extended linear text if you live in the newsfeed 24/7. I feel it myself, and I have been an avid book reader all my life. Hayes is very good on this condition:

You hear complaints about the gap between what we want to pay attention to and what we do end up paying attention to all the time in the attention age. Someone ambitiously brings three new novels on vacation and then comes back having read only a third of one of them because she was sucked into scrolling through Instagram. Reading is a particular focus of these complaints, I find. Everyone, including myself, complains they can’t read long books anymore. We have a sense that our preferences haven’t changed—I still like to read—just our behavior. And the reason our behavior has changed is because someone has taken something from us. Someone has subtly, insidiously coerced us.

All of which raises the inevitable question: how do we correct for this? One potential solution is designing new platforms that encourage you to explore information at different depths, and make it easy for you to get out of “the shallows,” as Nicholas Carr once called it, platforms that allow you to both skim and dive deeper.

This is, no surprise, one of the guiding principles behind what we’ve tried to do with NotebookLM. Upload a complex mix of source material and Notebook will do its best to help navigate your attention to the most important sections. A major quality for us is “interestingness”: given this particular iceberg of sources, what are the most interesting facts or ideas that we can expose at the surface. (One of the internal slogans for the Audio Overviews team is “make anything interesting.”) How you actually do that is that trick, of course. You can make something sound interesting with a clickbait headline, or you can do it by making it relevant to the audience, or by transforming it into a compelling narrative, and so on. A few weeks ago, Saturday Night Live ran a hilarious skit implicitly poking fun at Audio Overviews, the gist of which was that we were trying (unsuccessfully) to make school interesting by turning the classroom subject matter into a slang-heavy conversation between two Gen-Z hosts with a whole personal backstory to keep the kids hooked on the lesson:

This is obviously not what we are doing with Audio Overviews (though I have to say it was pretty cool to be sufficiently part of the zeitgeist to be parodied on SNL, in a skit featuring Timothée Chalamet no less.) But the general pattern that Herbert Simon described of taking a large quantity of information and withholding most of it from the user’s attention—and translating the remaining bits into some new format (in this case, a podcast-style conversation)—is very much part of our mission. You can get a high-level view of your sources almost instantly inside of NotebookLM — whether in audio form, or in the form of a text-based FAQ or Briefing Doc. But those options for skimming information are always accompanied by an invitation to plunge, because the sources are also reproduced in their entirety in the app, and our citation system is constantly referring you back to the original material. And unlike traditional books, you can always invoke the AI to help you understand a difficult concept, or translate it into more accessible language. You can hang out on the tip of the iceberg, or you can dive below the surface to inspect the whole thing.

[One quick postscript: my routine writing this piece is actually a great example of how NotebookLM makes it easy to dive beneath the shallows. I’d found two versions of Simon’s lecture online: a typewritten manuscript with many handwritten notes and edits, and the final text of the speech he delivered at Hopkins. I was curious about how the lecture had changed over time in Simon’s mind, and so I asked Notebook to itemize the most significant changes that Simon introduced with his edits, and in a followup query, to speculate on any motivations or shared patterns behind those changes. Trying to evaluate that myself would have taken at least an hour of painfully reading through the pages, deciphering Simon’s handwriting, and taking notes on each transformation. Notebook generated its analysis in ten seconds, with citations throughout allowing me to jump directly to any unusually provocative edits mentioned.

This is the kind of situation where I feel like the condensing skills of a tool like NotebookLM actually deepens my understanding of the material. Obviously I would get a more nuanced comprehension of the Simon lecture if I took the time to review all his handwritten edits manually, and in fact, were I writing a long essay or a biography exclusively about Simon, I would no doubt have gone through the trouble of doing that. But for a post like this where Simon’s lecture is only a part of what I’m discussing—and where it’s unclear whether there will be any insights uncovered in reviewing his changes—I probably wouldn’t have bothered to investigate the edit history at all. But it’s so easy to get that quick answer to the question “what did Simon change in this draft” that I end up taking that additional step now. And remember, this is all working with an image-only PDF, no OCR whatsoever. As I’ve said many times recently, no computer in the world could perform that task just 18 months ago. And now I barely even blink when I get a response like this one.]

Last week, I happened on a tweet from Patrick Collison (co-founder of Stripe and the excellent Stripe Press) reflecting on the ten classic “triple-decker” novels that he’d just finished reading:

This year, I read ten important historical novels: Jane Eyre, Middlemarch, To The Lighthouse, Bleak House, Portrait of a Lady, Anna Karenina, Life and Fate, Heart of Darkness, Madame Bovary, and The Magic Mountain… For me the clear standouts are Middlemarch, Bleak House, Karenina, and Life and Fate. I would enthusiastically reread any of them. If I had to choose just one to go to again, I would probably select Middlemarch. There's something memorably compelling in Eliot's affection and empathy for almost all of her characters. If Succession is a show with no likable personalities, Middlemarch is the opposite. Bleak House is a close second.

His ranking more or less aligns with mine: of the hundred or so novels from the 19th and 20th century that I read during my grad school days, my favorites were Middlemarch and Bleak House, and Middlemarch is the one book that I try to re-read once a decade.

This is, I realize, not a particularly contrarian take—Middlemarch regularly tops lists of English-language novels. But of course ranking novels is inevitably a subjective game, and less important than the question that Collison raises later in his tweet essay, which is why we should read classic novels at all.

Pleasure aside, should one read these books? Does one derive moral betterment from doing so? I'm not sure. Probably not in any narrow sense. Ethicists are supposedly no more ethical than regular people -- if deliberate study doesn't help, what hope does mere fiction have? And, anecdotally, I don't consider the humanities majors to be the moral betters of the STEM students. I do think they've helped with my understanding of history, though.

I sketched out an answer to this question in my book Farsighted, arguing that novels (and fictional narratives in general) were extensions of the human mind’s marvelous aptitude for building simulations of potential events. It’s something we do so effortlessly that we rarely stop to think about how nuanced a skill it really is: creatively projecting forward into our possible futures based on our previous experience of the world. Narratives of all sorts allow you to parachute into other simulated experiences, which ultimately give you more data for your own simulations. But novels, I would argue, give you the richest simulation of the interior life of other people’s experiences: you get a ringside view of all that emotional and cognitive action. This is particularly true of the novel after, say, 1750 or so, when the novelists began adopting more of the inner monologues (both first-person and “close-third” perspective) that Shakespeare had explored on the stage.

It seems fairly obvious to me that there is practical utility in running these simulations. We accumulate wisdom that we can apply to our own lives by watching other people live theirs. Historical nonfiction—particularly if the author has access to the subject’s inner life through journals or correspondence—arguably has even more utility, in that the events in question actually happened in the real world, and not in the imagination of the novelist. (This is one reason I decided to write nonfiction instead of novels—the other being that I don’t think I’m a good enough writer to write novels.)

But my admiration for Middlemarch and Bleak House goes beyond their portrayal of the psychological interiority of their characters. (Dickens actually isn’t all that good at the inner game, it turns out.) Both novels succeed at something else: they invite us to think across multiple scales of experience—and to make causal and associative connections between those different layers. In Farsighted, I described this using a metaphor borrowed from the physics of radio waves. There were “narrowband” perspectives that focused on a single layer, and “full-spectrum” perspectives that captured a wider range of information. (Some of you will recognize Stewart Brand’s pace-layers diagram and E.O. Wilson’s consilience in this general idea.) This was how I made the case in Farsighted:

Think of the buzzing intensity of the interior monologue as the high end of the spectrum; the shifting alliances of friends and extended family and town gossip as the midrange; and the slow, sometimes invisible churn of technological or moral history as the low end of the spectrum. Some novels thrive in the narrowband. They home in on the interior monologue or the public sphere. But some novels are full spectrum. They show how those private moments of emotional intensity are inevitably linked to a broader political context; how technological changes rippling through society can impact a marriage; how the chattering of small‑town gossip can weigh on one’s personal finances.

Don’t panic if you haven’t read any of these novels—I think I can still make this meaningful without requiring any real knowledge of the plot or issuing any spoiler alerts. In Middlemarch, the key events of the narratives are all deeply influenced by forces existing at at least seven distinct bands of the spectrum:

MIND
(Dorothea’s emotional and sexual attraction to Ladislaw; her drive for intellectual autonomy)

FAMILY
(The possibility of having children; the impact on her choices on her father and sister)

CAREER
(Dorothea’s active oversight over the Lowick estate, and the social impact of “improving” Lowick)

COMMUNITY
(The town gossip, which is effectively its own character in the novel)

ECONOMY
(The financial consequences of relinquishing Causabon’s fortune)

TECHNOLOGY
(The chaos of the railroad; the productivity of the new agricultural techniques.)

POLITICS
(The reform movement that catalyzed the events of 1832; Ladislaw’s political ambitions within that movement.)

I can’t tell you whether Middlemarch is the greatest novel ever written. I’m not even sure you could plausibly declare any one novel “the greatest.” But I think you could make a convincing case that Middlemarch is the novel that best integrates all seven of those layers. Compare Middlemarch to earlier classics from Austen or the Brontës, and the difference is almost immediately apparent. The emotional and familial layers in, say, Pride and Prejudice or Jane Eyre are fully developed. But the geo-political or technological forces that were transforming English society in the first half of the nineteenth century do not propel the narrative in any material way. And where they do exert influence, their roles are subtle, almost camouflaged, as my old teacher Edward Said argued about the relationship between overseas imperialism and Austen’s narratives. As brilliant and entertaining as those novels are, their natural home is the drawing room, or the cotillion. That is the scale at which those stories operate. Middlemarch never lets its reader (or its characters) settle too comfortably into those drawing‑room conversations. There is always a larger, bustling world banging on the windows.

Bleak House pulls off a comparable feat. As in most of his later novels, Dickens builds a narrative that manages to connect a staggeringly wide cross-section of the social classes of industrial London, and the superorganism of the city itself is genuinely a character in the book. And Bleak House added a new band to the spectrum, one that becomes central in 20th-century classics like The Trial, 1984, or Terry Gilliam’s Brazil: the bureaucratic state, most famously the Court Of Chancery with its endless litigations and a whole comical troupe of nascent knowledge workers churning through “bills, cross-bills, answers, rejoinders, injunctions, affidavits, issues, references to masters, masters’ reports, mountains of costly nonsense.”

The problem with Dickens is that the people themselves aren’t quite real. Dickens does the low ends of the spectrum as brilliantly as anyone (metropolis, industry, bureaucracy) but at high-end the characters have a tendency to devolve into caricature. Orwell described it best in his famous 1940 essay, comparing Dickens to another giant of the 19th-century novel:

Why is it that Tolstoy's grasp seems to be so much larger than Dickens's — why is it that he seems able to tell you so much more about yourself? It is not that he is more gifted, or even, in the last analysis, more intelligent. It is because he is writing about people who are growing. His characters are struggling to make their souls, whereas Dickens's are already finished and perfect. In my own mind Dickens's people are present far more often and far more. vividly than Tolstoy's, but always in a single unchangeable attitude, like pictures or pieces of furniture… Dickens's characters have no mental life… They never learn, never speculate.

Back to Collison’s original question: why should we read these books? To me, the answer comes down to this: getting your brain to accurately assess the full spectrum takes effort, but it’s a valuable skill to have in life. If you over-index on one layer at the expense of the others, you’ll end up making less nuanced choices, both personally and professionally, because your mental model about what is actually happening at a given crossroads of your life will be too narrow. And so we should embrace any opportunity to practice thinking across scales, in part because most educational environments are deliberately designed to keep you specialized in one part of the spectrum. So in a way, I’ve come to think about books like Middlemarch or Bleak House—using the language of deep learning—as a source of training data for full-spectrum thinking. They are trial runs that prepare you for the real thing.

There are other reasons to read novels, of course; both Middlemarch and Bleak House have beautifully crafted plots that are genuinely fun to unravel, if you can keep track of the characters, and there are some legendary sentences in each book. Another reason to read them—which Collison emphasizes as well—is that they are extremely effective vehicles for time-traveling back to earlier periods in history. There is, I think, something intrinsically valuable in having that glimpse of humanity in a slightly different configuration—different values, technologies, economic structures—that widens your perception of your own moment. No doubt traditional history books (and period film/TV narratives) can transport you to an even more accurate rendition of the past. But the novel gives you the best glimpse of what it really felt like, from the inside, to live through those earlier ages.

All of which raises the question: are people still writing novels like Middlemarch and Bleak House? I can see some of that tradition in books like Franzen’s Freedom, or maybe Tart’s The Goldfinch. I had been hoping for something in that mode with Zadie Smith’s last novel The Fraud, which actually features Dickens as a character, but it turned out to be a slightly different kind of book. I’d be curious to hear who you all think are the heirs to Dickens and Eliot today. Or has something changed in the culture that now makes it harder or less interesting to write that sort of full-spectrum narrative?

And finally, happy new year to all of you! Thanks to all the NotebookLM craziness, this turned out to be one of the most fascinating professional years of my life, even if it did keep me from embarking on my own next book project. One of my goals for 2025 is to get started on that project, and to figure out a way to involve the Adjacent Possible community as I develop the ideas. (I’m also incredibly excited to actually write a book from scratch using NotebookLM, as you might imagine; if you haven’t seen the new UI we rolled out a few weeks ago, definitely check it out.) More on that book project to come…

[Art by ImageFX—I am weirdly obsessed with generating these simulated wood carving illustrations using AI. The prompt was: “create a wood carving that shows the range of human experiences in 19th-century urban culture: mental life, family, community, economy, technology, politics. at the center of the piece, show someone reading a novel.”]

I’ve thought a lot lately about the underlying developments that made audio overviews such a juggernaut: the brilliance of the core prompts that generate the conversation itself (I say "brilliant” because I did not write them!); the uncanny verisimilitude of the AI voices themselves. But there was also the novelty of an AI that seemed to be an expert in whatever documents you had uploaded to your notebook in the first place. If you’ve been reading Adjacent Possible for a while—or you’ve been an early adopter of generative AI—you’ve been aware of and likely interacting with “source-grounded AI” for a year or more. But I think the whole concept had not fully penetrated the broader culture as much as maybe I had assumed. So the sense of wonder (and shock) on the faces of those TikTok creators while listening to an audio overview is not just a reaction to the quality of the conversation; it’s a reaction to experiencing, for the first time, an AI that has some level of familiarity with—even mastery of—your own curated information: your law school reading assignments, your journals, the draft of your novel. You could experience source-grounding via text at a remarkably sophisticated level at Notebook starting in May, when we switched over to Gemini 1.5 Pro and introduced inline citations. But the best way to appreciate that sophistication was to upload a collection of documents, ask a complicated question, get a nuanced answer from the model, and click through the citations to confirm. That could be a powerful experience if you managed to complete those steps, but even if you did, it wasn’t particularly shareable as an experience. Audio overviews, on the other hand, were easily conveyed by social media. And as Andrej Karpathy pointed out, in one of the early tweets about the feature, Audio overviews let you sit back and listen as the AI asked its own questions.

And here’s the thing about source-grounding: the underlying development that makes it possible is the dramatic increase in the size of the model’s context window, the “short-term memory” of the model where the user can supply their own information, as opposed to the “long-term memory” of its training data. When we started working on what became NotebookLM in the summer of 2022, we could fit about 1,500 words in the context window. Now we can fit up to 1.5 million words. (And using various other tricks, effectively fit 25 million words.) The emergence of long context models is, I believe, the single most unappreciated AI development of the past two years, at least among the general public. It radically transforms the utility of these models in terms of actual, practical applications.

One interesting application that we noticed early on in the NotebookLM Discord was that long context was enabling role-playing-game enthusiasts to keep track of their games. You could load in an entire game manual, or the detailed backstory of the campaign you’d authored, and then consult NotebookLM in your capacity as host/Dungeon Master as you played. That was unthinkable in the early days of chatbots, but with NotebookLM’s architecture (and Gemini’s long context) it was as simple as uploading those documents, and asking questions in the chat. At some point early the summer, one user asked in the Discord if anyone had attempted to turn NotebookLM itself into a game-playing platform, where you could actually conduct an entire adventure via the text chat. That sent me off on a long tangent exploring different prompts, and after a week or so I had actually creating a working version of the app that would effectively generate a playable simulation of whatever document you gave it. I created a playable version of The Ghost Map, where you play John Snow trying to solve the mystery of cholera in the streets of London in 1854. I uploaded the wikipedia page for the Cuban Missile Crisis and played through a simulation of those events with me in the role of JFK.

Ever since Google DeepMind introduced the first million-token model earlier this year, I’d been taking notes on an essay that would try to explain why long context is so revolutionary. But playing those interactive simulations gave me an idea: what if I began my long-context essay with a playable game? You’d be able to experience the magic of what these new models are capable of, and then I’d explain what it all means, and where I think it is going to take us. And so I built a mini-game based on one of the chapters from The Infernal Machine; you have to solve the mystery of a Soho break-in using the cutting-edge new forensic science of fingerprint analysis.

I’ve been working on this essay on the side ever since then, and I’ve just put it online here, as a standalone site. (I couldn’t figure out how to get the game to work in a Substack post, so I collaborated with my son on building the game and the site at a separate URL.) We hope to bring some of this simulation-builder functionality to NotebookLM in the coming months, and of course everything we are working on there is ultimately made possible by these long-context models. Hopefully you can read the whole thing, but I’ll leave you just with the closing paragraphs here, ruminating on some of the possibilities for collective thinking and decision-making in a long context world.

Long context is also a boost for collective intelligence as well. If you assume the average corporate document—a press release, or marketing plan, or minutes from a board meeting—is a few thousand words long, then today’s models can simultaneously hold in their short-term memory close to a thousand documents. A state-of-the-art language model with the ability to instantly recall and generate insights from the most important thousand documents in the history of a company would possess knowledge about that company that would rival that of any single employee, even the CEO. It seems inevitable that anyone trying to make a multi-faceted decision about the future of an organization would want to at least consult such a model. We know from endless studies of social psychology that diverse groups—with different forms of expertise, different pools of knowledge, different cultural backgrounds—tend to make better decisions than homogeneous groups. In a small-context world, you can get some of that diversity from a language model, in that its training data draws from a vast archive of global human knowledge. But a long context model allows you to take that global knowledge and apply it to the unique challenges and opportunities of your own organization. In a matter of years, I suspect it will seem bizarre to draft the specs for a new feature or a company initiative or a grant proposal without asking for feedback from a long-context model grounded in the organization’s history. (And perhaps the public history of its competitors.) It wouldn’t be a replacement for the expertise of the employees; instead, the model would occupy another seat at the table, adding a new kind of intelligence to the conversation, along with a vastly superior recall.

And there’s no reason the organization in question would have to be a corporate entity: maybe it’s a city, or a government agency, or a grassroots advocacy group. Just a year or two ago, asking a small-context model to help chart strategy for, say, a suburban town would have been almost as useless as asking post-surgery Henry Molaison to narrate the preceding six months of his life. Long context gives the model more than just the reasoning and linguistic fluency that emerges through the training process; long context gives the model a specific history to draw from, the idiosyncratic sequence of events that make up the life cycle of any organization or community. Grounded in a long-context history, models are now capable of going beyond just answering factual questions or giving feedback on proposed plans. You might ask the model to identify patterns in a company’s archive to help simulate the way customers or clients would respond to a new product. Or you could draw on the long-context understanding of a city to conduct scenario planning exercises to simulate the downstream consequences of important decisions. Given everything we know about the power of learning through play, you might even take all that contextual history and turn it into a game.

All of which suggests an interesting twist for the near future of AI. In a long-context world, maybe the organizations that benefit from AI will not be the ones with the most powerful models, but rather the ones with the most artfully curated contexts. Perhaps we'll discover that organizations perform better if they include more eclectic sources in their compiled knowledge bases, or if they employ professional archivists who annotate and selectively edit the company history to make it more intelligible to the model. No doubt there are thousands of curation strategies to discover, if that near future does indeed come to pass. And if it does, it will suggest one more point of continuity between the human mind and a long-context model. What matters most is what you put into it.

Read the whole thing (and solve the Soho mystery!) at thelongcontext.com.

Thank you very much, Marty—what a great honor this is. During yesterday’s presentations, I was very happy to see the word “savoring” included in a list of activities associated with resilient behavior, because I have always given my kids the advice that it’s important to savor moments in your life where something goes well for you professionally or personally, so that you can cement that positive feeling in your mind, and thus be more inclined to seek it out again. And if there were ever an award that deserved savoring, it would be being recognized for advancing the cause of human flourishing! 

When the University first reached out to tell me about this honor, Marty suggested that I could talk about my work on the story of human longevity from the Extra Life project. I said I thought the occasion warranted some original material, which I have put together for you here, but I think Marty was right that longevity is a great place to start. 

The story at the center of Extra Life was really the story of a single number, and how it changed over the preceding century: the number of years that the average human could expect to live given the conditions in the world at the time of their birth. In other words, life expectancy. A little more than a hundred years ago, at the end of the last global pandemic, human life expectancy stood at around 35 years. A hundred years later, it was more than 70. In the span of just one century, we managed to double the average human lifespan. 

To give you a sense of the magnitude of this change—and how it is not something that exclusively happened to the industrialized west—take a look at these two charts. This is the world in 1950. The countries in blue are places where life expectancy has crept above 70 for the first time. It’s just a handful of countries clustered around the North Sea. In red, are all the countries where life expectancy was still below 45.

Compare what the world looked like just 35 years later, in 2015. The blue countries have massively expanded, and there’s not a red country remaining on the map. In fact, there are very few remaining countries with life expectancy below 60, much less 45. 

Now, you will sometimes hear people say this kind of progress is actually just a statistical illusion—that we only got better at reducing infant mortality, which was dragging the overall average down because so many people were dying after six days or six weeks. It is true that infant mortality—and childhood mortality—has been dramatically reduced over the last hundred years. It used to be, for all of human history really, that a third of your children would die before reaching adulthood. Now that number is less than one percent. That alone would be a story of extraordinary progress.  But the change is actually happening at the other end of the spectrum as well. My grandmother lived to be almost 105. That would have been a massive anomaly a century ago, but it is becoming increasingly common. In fact, centenarians are the fastest growing age demographic in the United States right now.

Take a look at this early infographic by the great Victorian statistician William Farr, which is attempting to show mortality rates by age group in London in the early 1840s.

I find something incredibly heroic about this chart. I mean, here's a guy without computers, without the Internet, without Excel, trying to do something that is incredibly hard and incredibly important. He's trying to look at broad patterns in life and death in a great city, trying to make sense of what is going on. And what the chart reveals is that there was indeed a tragic amount of death among children back then; not just infants, but five-year-olds and ten-year-olds were dying at an alarming rate. But by the same token, almost nobody made it to 85 or 90. And more than half of the population was dead by the age of 45. How many people in this room are older than 45? [More than half the hands go up.] Think about progress and human flourishing in that context. Pretty much the most important prerequisite to flourishing is not being dead! 

But there’s another kind of flourishing that comes out of that transformation of human lifespan — something that, appropriately enough, I’ve come to appreciate as I’ve gotten older.

It wasn’t just that my grandmother got to enjoy living a longer life than her ancestors did, but also that she lived long enough to develop deep, enduring relationships with her great-grandchildren. That has an emotional value for sure, but it also has cultural value through the knowledge that gets passed between generations. I’ve written in a number of books about the importance of diversity in driving creativity and decision-making — diversity of background, gender, intellectual fields of expertise and so on. But one element of that story that I think is underappreciated is generational diversity. 

A little more than a decade ago my wife and I made the decision to move our family to California for a few years, and for complicated reasons that came out of my career in the tech sector, I ended up making a number of deep friendships out there with people who were close to my parents age—one of whom had actually been in my dad’s class at college, though they didn’t know each other at the time. Those friendships have been intensely important to me, and because all those people have had fascinating intellectual journeys, I’ve learned an enormous amount from their stories of the tumult and possibility of Northern California in the sixties, or the early days of the PC revolution. One of them is actually Larry Brilliant, who played a key role in the eradication of smallpox, one of the most momentous events in the history of life expectancy. (I will return to Larry’s wisdom in a few minutes.) And in turn through these cross-generational friendships, I think they’ve picked up a few interesting signals from my experiences with emerging technologies that they might have otherwise been less attuned to. 

And now with the work I’ve been doing with Google, most of the team on the NotebookLM project are much closer to my children’s age than mine, which has been a wonderful addition to my life. I so enjoy going back and forth between those different worlds: I get to be both the young gun and the elder statesman, sometimes all in one day.

If there’s a transfer of fresh knowledge and hard-earned wisdom that comes with interaction across generations, there’s also a widening of temporal scales that emerges when human beings start living longer lives. It’s hard to imagine the deep past, or the distant future, when you’ve only been alive for a decade or two, and if it’s more likely than not that you’ll die before you meet your grandchildren, there’s less incentive to invest in or project forward into that future. But when you think you have a good chance of living for a century or more—in part because your direct relatives are living that long—your temporal horizons expand. 

There’s a wonderful moment in Extra Life where I’m interviewing Larry Brilliant about the smallpox eradication project, and he talks movingly about the very last person to contract the disease in the wild, a young girl named Rahima Banu (who survived the infection). In the show, Larry says:

When the smallpox scabs fell off Rahima Banu and were incinerated by the heat, that was the end of an unbroken chain of transmission of variola major going all the way back to Pharaoh Ramses V. 

I’ve always been drawn to those kinds of long-term perspectives, where you position yourself—in this case, gloriously free of smallpox—in the larger context of hundreds or thousands of years of human suffering and progress. Some of my California friends even built an entire organization to celebrate that long-term view: the Long Now Foundation, which is dedicated to thinking on the scale of centuries or millennia, encouraging us to get out of the 24-hour news cycle that dominates so much of our lives today. A technologically advanced culture cannot flourish without getting better at anticipating the future. That’s why science fiction matters. That’s why scenario planning matters. That’s why complex software simulations that enable us to forecast things like climate change on the scale of decades matter. 

And here I want to bring us back to another idea that Marty Seligman has been an advocate for. Almost ten years ago, he edited a collection of essays called Homo Prospectus which had a huge influence on my thinking about the world. The core idea behind that book was that a defining superpower of human beings is our ability to mentally time-travel to possible future states, and think about how we might organize our activities to arrive at those imagined future outcomes. 

“What best distinguishes our species,” he wrote in the introduction to that book, “is an ability that scientists are just beginning to appreciate: We contemplate the future. Our singular foresight created civilization and sustains society. A more apt name for our species would be Homo prospectus, because we thrive by considering our prospects. The power of prospection is what makes us wise. Looking into the future, consciously and unconsciously, is a central function of our large brain.” 

It is unclear whether nonhuman animals have any real concept of the future at all. Some organisms display behavior that has long-term consequences, like a squirrel’s burying a nut for winter, but those behaviors are all instinctive. The latest studies of animal cognition suggest that some primates and birds may carry out deliberate preparations for events that will occur in the near future. But making decisions based on future prospects on the scale of months or years — even something as simple as planning a gathering of the tribe a week from now — would be unimaginable even to our closest primate relatives. If the Homo prospectus theory is correct, those limited time-traveling skills explain an important piece of the technological gap that separates humans from all other species on the planet. It’s a lot easier to invent a new tool if you can imagine a future where that tool might be useful. What gave flight to the human mind and all its inventiveness may not have been the usual culprits of our opposable thumbs or our gift for language. It may, instead, have been freeing our minds from the tyranny of the present.

The problem now is that the future is getting increasingly hard to predict, in large part because of what has started to happen with artificial intelligence over the past few years. I’ve spent a lot of my career looking at transformative changes in technology, and I’ve come to believe that what we’re experiencing right now is going to be the most seismic, the most far-reaching transformation of my lifetime, bigger than the personal computer, bigger than the Internet and the Web. And while there is much to debate about what the impact of this revolution is going to be for the job market, for politics, and just about any other field, there is growing consensus that it is going to provide an enormous lift to medicine and human health. The Nobel Prize for chemistry going to the AlphaFold team last week was arguably the most dramatic illustration of the promise here. Earliest this month, Dario Amodei—the founder of the AI lab Anthropic, makers of Claude–published a 13,000 word piece on where he thought we were headed with what he calls “powerful AI” in the next decade or two. The line that really struck me in the piece was this:

My basic prediction is that AI-enabled biology and medicine will allow us to compress the progress that human biologists would have achieved over the next 50-100 years into 5-10 years… a compressed 21st century.

Whether or not something that dramatic does come to pass—and I think we have to take the possibility of it seriously—it seems clear that given the kind of biological and medical advances that AI will likely unlock, there is significant headroom left in the story of extended human lifespan, perhaps even a sea change in how we age. That is, on one level, incredibly hopeful news. But it is also the kind of change that will inevitably have enormous secondary effects. To understand just how momentous those changes could be, take a look at this chart:

That’s the 6,000 year history of human population growth. You might notice, if you really squint your eyes, that something interesting appears to happen about 150 years ago. After millennia of slow and steady growth, human population growth went exponential. And that’s not the result of people having more babies—the human birth rate was declining rapidly during much of that period. That’s the impact of people not dying. And while that is on one level incredibly good news, it is also in a very real sense one of the two most important drivers of climate change. If we had transferred to a fossil-fuel-based economy but kept our population at 1850 levels, we would have no climate change issues whatsoever—there simply wouldn’t be enough carbon-emitting lifestyles to make a measurable difference in the atmosphere.

The key idea here is that no change this momentous is entirely positive in its downstream effects. Trying to anticipate those effects, and mitigate the negative ones, is going to take all of our powers of prospection. 

When I was putting together my thoughts for this talk, my mind went back to the one time I spoke with Marty, about five years ago, when I was writing about cognitive time travel for the Times Magazine. As usual, I was incredibly behind in actually doing the reporting for the piece, and I’d called Marty desperate for a few quotes on a tight deadline. He very generously found time for me, but he had to do the call from an animal hospital, because as it happens he and his family were in the middle of putting their dog down. So our very first moments in conversation with each other plunged right into the depths of loss and grieving and the strange bonds that form between animals and humans. There was no small talk. 

As I said earlier, death is, in the most basic sense, the termination point of human flourishing. But it’s also the shadow that hovers over us while we are still alive. We have done so much to minimize that shadow over the past century or two, going from a world where it was the norm for a third of your children to die before adulthood to a world where less than one percent do. But what does it mean for human flourishing if that runaway life expectancy curve that we’ve been riding for the past century keeps ascending? What does it mean if AI starts out-performing us at complex cognitive tasks? How do we flourish in that brave new world? Do we take on a new responsibility—not just ensuring the path of human flourishing, but also the flourishing of our AI companions? These are all difficult questions precisely because of time. The rate of change is so extreme right now we don't have as much time to learn, and adapt. The doubling of human life expectancy was a process that really unfolded over two hundred years, and we’re still dealing with its unintended consequences. What happens if that magnitude of change gets compressed down to a decade?

I don’t know the answers to those questions yet, I’m sorry to report. But maybe spelling them out together helps explain something about what I’ve tried to do with my career, which I think from afar can sometimes seem a bit random, bouncing back and forth between writing about long-term decision making or exploring the history of human life expectancy and building software with language models. This award is called the Pioneer Award, and while I’m deeply honored to receive it, I don’t think of myself so much as a pioneer in any of these fields, but rather as someone who has consistently tried to find a place to work that was adjacent to the most important trends in human flourishing, so that I could help shine light on them, explain them to a wider audience, and in the case of my work with AI, nudge them in a positive direction to the best of my ability. That you all have recognized me for this work—pioneer or not—means an enormous amount to me. You can be sure I will do my best to savor it.

It seemed like a remarkably high-order skill for a technology that had well-documented problems with things like basic math and hallucinations. Over time I came to realize that it was just a slightly more sophisticated version of something that neural nets had been good at for more than five years: translation. Take this input and turn it into this new output, but keep the meaning intact. We'd first seen the power of this with the launch of Google Translate, taking English text and magically turning it into Spanish, arguably the first breakthrough in deep learning that was widely adopted by consumers. But by the time I arrived in 2022, the models had graduated up to translating astrophysics into elementary school-friendly metaphors.

The ability to translate or summarize is one of those edge case capabilities that can confuse things more than clarify in the context of debates about "artificial general intelligence." Until a few years ago, humans were the only entity on the planet that could translate information from one format to another in this way. Now a computer can do it, faster and more reliably than any human on the planet. Thanks to the underlying Gemini model, NotebookLM can effortlessly shift back and forth between almost forty different languages; you can read a document in Japanese and ask questions about it in Spanish and it won't miss a beat. That is a giant step forward.

But summarizing and translating are not synonymous with thinking. On the most basic level, translation tasks do not involve generating original ideas. If your translator keeps inserting his own theories while translating your work into another language, that's a bug, not a feature. Translation and summarization are maybe the most striking examples of AI's uneven development: the models are great at some things that have traditionally been exclusively the province of human intelligence, and surprisingly lame at tasks that a $50 Radio Shack calculator could have done in the 80s.

In part because those early “explain it to me like I’ve five” experiences left such an impression on me, and in part because it’s generally just a good strategy to lean into what the models do well, many of the early NotebookLM features (back when it was an internal prototype code-named Tailwind) were variations on this theme. We had a feature we called “Explanatory metaphor” that would take any text you gave it and generate a helpful metaphor to describe the core ideas in the passage. We did a number of experiments with different summarization formats; back then, there was still something magical in giving the model a few complicated paragraphs and watching it turn them into reader-friendly bullet points. 

Over time, the possibilities of translating information from one format to another expanded, and it became possible to work with much larger documents or collections of documents. Because NotebookLM is fundamentally a tool to help you understand things, we’ve started to put more and more emphasis on allowing our users to convert whatever information they’re working with into the structures that work best with their way of thinking or learning. That was the underlying principle behind the suite of tools we introduced in June with Notebook Guide, which lets you convert your sources into formats like study guides or FAQs with a single click. 

All those formats had one thing in common: they were all text-based. But not everybody learns or remembers most effectively through reading. Many of us are auditory learners, or just prefer to take in new information while walking around or driving, when reading is impossible. And we know from the massive rise in podcast listening that one of the most powerful ways to understand a topic is to listen to two engaged, thoughtful people having a conversation about it.  

But what if it doesn’t have to be actual people having the conversation? 

On Wednesday, the NotebookLM team rolled out a new feature called Audio Overviews, a new addition to Notebook Guide that takes your collected sources and uses them to generate a roughly ten-minute “deep dive” audio conversation between two AI hosts. Here’s what I got when I created an audio overview based on the full text of The Infernal Machine, my latest book:

Pretty wild, right? In the last 48 hours, we’ve seen an amazing amount of interest in this feature, with people doing everything from generating overviews based on their CVs (apparently very good for your self-esteem), uploading recent corporate documents to create a company-wide “Week in Review” episode to share with colleagues, or even uploading the draft of their fantasy novel manuscript to hear which storylines were most conversation-worthy. Google’s Jeff Dean has a fun twitter thread where he shared out a bunch of examples of what people are doing with the tool.

Audio overviews take up to five minutes to generate. There are a number of edit cycles happening behind-the-scenes to ensure the content is faithful to the source material, engagingly described, and delivered with convincing human intonation. (No one wants to listen to two Siris in conversation.) In that discussion of Infernal Machine, the AI hosts riff on the major themes and introduce some of the most interesting characters from what is a fairly complicated book. They do mispronounce Alphonse Bertillon's name, and they describe the Ludlow massacre as happening in 1913, not 1914, though the labor dispute that led to it did in fact begin in 1913. But over all, it's a remarkably lucid and thorough summary of a 300-page book, entirely generated by software.

One thing I’ve noticed listening to dozens and dozens of these is that while the conversation has a playful tone throughout—there’s a lot of bantering and an arguably excessive fondness for puns—I’ve never actually heard either of the hosts say something genuinely funny. I imagine over time we will probably allow you to dial the tone of the discussion to fit your own tastes. (There's another great conversational audio experiment at Google called Illuminate, which is currently more focused on scholarly texts and generally has a more serious tone.) But I’m not totally convinced that the models are capable of true conversational humor yet. This may just be another place where artificial intelligence is developing unevenly, and if it is, I suspect it’s because humor is in many ways the opposite of translation and summarization; humor is all about surprise, about defying expectations, going off script in just the right way.

In two years, the models have gone from producing a few sentences of helpful astrophysics metaphors to generating a convincing ten-minute audio conversation based on an entire book. That is astonishing progress by any measure. But they still don’t know how to make us laugh. 

I’m mulling one of those serendipitous collisions right now, and thought I’d share it with you. It may end up as random debris for all I know, and you’ll never hear about it again. But here’s “the warm little pond” so far…

Sometime in the late spring I read H. W. Brands’ epic account of the California Gold Rush, The Age of Gold, published in 2002. I could easily spend the rest of this essay just recounting the greatest hits from it, but there was one mind-bending fact that has stuck with me since I finished the book. Because the Sierra Nevada—and the Basin and Range Province to the east of them (including Death Valley)—were such inhospitable environments for human passage, news of the discovery of gold in the western foothills of the Sierra reached Australia and China and Chile months before it reached the east coast of the United States. Without a railroad or a telegraph line to connect the United States from coast to coast, the maximum speed of information was gated by the maximum speed of ships, and sailing from San Francisco to the Atlantic Seaboard took months, if you made it at all. You either had to sail around the horn of South America—one of the deadliest seas in the world—or sail to Panama and then attempt the land crossing of the Panama isthmus, with hopes of connecting with a ship bound for New York or Boston at the other end. Either way, the information you carried moved at a snail’s pace compared to a Clipper ship riding the Pacific trade winds to Sydney. All of which meant there were Australian and Chinese prospectors arriving in the Sierra foothills before the President of the United States learned that gold had been discovered in California. 

I love stories like this—all those Australians making it to California before New York and D.C. had even learned of the discovery at Sutter’s mill—in part because they remind us of how transformations in information networks influence the course of history. (Many years ago when I was in grad school, I wrote a paper on how the narratives of the 18th-century novel were made possible by the relatively sluggish speed of the postal system: major plot points often revolved around a message not arriving in time, or two messages crossing paths and thereby leading to some fertile misunderstanding; the plot-lines simply wouldn’t be possible in an age of text messages and cell phones.) But I also love this story because it makes explicit something that is always happening around us: the collision between different temporal scales. On the one hand, it’s a story about technological change, which progresses at the scale of decades and centuries. (If Sutter and Marshall had enjoyed access a Starlink subscription back in 1848, Washington would have been updated about the gold discovery within a matter of days, if not hours.) But it’s also a story about geological time: the physical barriers that plate tectonics tossed up between the east and west coasts over millions of years, that ultimately forced the news to follow such circuitous routes back to the eastern seaboard. 

In The Age Of Gold, H. W. Brands has a wonderful line about this kind of collision, at the end of a long preamble about the physics and geology that led to both the creation of gold itself and its relative abundance in one small stretch of the Sierra range:

Time and again the riverbeds shifted; the caches of gold were entombed under tons of gravel and sand and mud. Forests grew atop the tombs, quiet and serene by the measure of biological time, amid a landscape that remained violently unsettled in geological time. Eventually those forests attracted an inquisitive species, a biped drawn to the forests for their timber, but possessing a peculiar penchant for shiny yellow metal. When this species began scratching about in one of the streams as yet unburied, biological time and geological time abruptly intersected, and entered historical time.

So good. The deep past of geological time isn’t dead. It’s not even past.

Quite a bit of this same deep-time magic exists in Kim Stanley Robinson’s The High Sierra: A Love Story, which I also read rapturously—though in small doses, usually right before bed—this spring. It’s a beautiful physical object, that book, but also beautifully hard to place as a genre: it’s some strange hybrid of memoir, regional history, geological survey, and backpacking how-to guide. Near the beginning, Robinson announces that he is going to adopt a perspective he calls “psychogeology,” a kind of split focus that tracks both the deep-time geological forces that transform the landscape and the ways in which that landscape influences (and inspires) the human mind.  Robinson is so rhapsodic about his lifelong experiences backpacking in the High Sierra that I found myself on the verge of planning my own trip into the wild (which, if you know me, is an absolutely preposterous idea.) And what captivates him, more than anything else, is the topography of the place, in particular the high basins that are the Sierra’s most distinguishing characteristic. “Compared to the Swiss Alps,” he writes, “the Sierras are actually a kind of high plateau, lightly etched by ice. In fact, although a few peaks in the Alps stand a bit higher than Mount Whitney, the Sierras have much more land over 10,000 feet than the Alps do—4,700 square kilometers in the Sierra compared to 1,900 square kilometers in the Alps.” 

The Alps have more visual drama, with all the deep incisions carved between the legendary “horns.” (They were the poster art for the  Romantic “sublime” for a reason.) But the High Sierra is walkable.  You can walk—or at least scramble—a continuous line for 200 miles across the Sierra crest and barely ever dip below 10,000 feet above sea level. The land is high, but not as extreme as the Alps. And because the Sierras happen to reside at the far edge of a Mediterranean climate, characterized by a pronounced rainy season, water abounds in those high basins, released slowly through the summer as the snow melts, captured in ponds, streams, lakes. “Everywhere water is in visible motion,” Robinson writes. “In the summers you almost can’t camp in a place where the sound of moving water won’t be part of your lullaby that night."    

Robinson goes on to explain how this magical landscape emerged, attributing it to a rare “goldilocks” zone of moderate glaciation that creates the high basins:

[One] range that helps us understand the effects of glaciation lies just to the east of the Sierra, in the White Mountains that straddle the California-Nevada border. In a rough way this range can serve to show what the Sierras might have looked like before the glaciers did their work. Existing in the Sierra’s rain shadow, the White Mountains never had many glaciers, and so are still basically a line of gigantic rounded hills. The Alps then illustrate the other end of the process, having lived under so much ice that the basins got ground away. These are the “too little” and “too much” examples… [The ideal is] glaciation, but not too much of it: and this particular ice history is somewhat rare in the world. Which makes sense, given that the Sierra Nevada is unusual in its combination of a Mediterranean climate and great height… The rare combination of height and climate needed to create high basins means there are only a few mountain ranges on Earth that have them. And basins are the golden zone for walkers. This is psychogeological, yes, but I’m convinced it’s quite real as an explanation for the Sierra’s particular joy.

For me, the joy also comes from a new way of seeing the mountains, a kind of cognitive time-lapse where you simultaneously see them in their current shape but also imagine the ancient glacial and tectonic forces that slowly brought that shape into being. Apparently, some geologists have pointed out that the difference between the Alps and the Sierras may also be attributable to the fact that the Alps are largely composed of sedimentary rocks that are more prone to erosion than the sturdy batholith of the Sierras. But whatever the ultimate explanation is, there is majesty in experiencing the natural world across deep time like this. It’s a sort of temporal version of the Romantic sublime, I suppose, where you catch glimpses of the distant past embedded in the present landscape that surrounds you. 

Redwood forests are another environment that triggers the temporal sublime, not just because the trees themselves are ancient, but because the whole ecosystem of an old-growth redwood forest conjures up the deep past of the planet. The sequoia and redwood family were the dominant trees in Europe and North American during the Jurassic and Cretaceous period, and the fossil record shows that the ancestors of today’s redwoods grew all across the globe: in China, Japan, even Australia. Today their footprint has contracted to a small stretch of the California and Oregon coastline, and the southern Sierra. For that reason, walking through an extant old-growth grove—magical enough in its own right—is the most immersive and realistic time capsule of what Earth must have felt like during the age of the dinosaurs. 

There’s a personal side of this story for me. Proximity to redwoods is very high on our list of why we spend so much of our time in this part of Northern California. Just a few years ago we discovered two young redwoods had taken root amid the small stand of oaks that abuts our property; they are now conspicuously taller than any other tree around them, and they are still in the toddler stage, as redwoods go. (I’ve been taking snapshots of them from the same angle over the past five years at the end of the summer; someday it will be a helluva time lapse.) I devoured Richard Preston’s classic The Wild Trees—all about climbers attempting to summit the giant redwoods a few hundred miles north of us up the coast—when it was first published; my oldest friend and his now-wife gave the book out as a gift to all the attendees of their wedding, which was conducted in a Sonoma redwood forest. Old-growth groves are the closest thing I have to a sacred space.

As it happens, I just started reading Ferris Jabr’s new book Becoming Earth—highly recommended, like everything Jabr writes—and near the beginning he notes that James Lovelock often drew on redwoods as a metaphor to help people understand his once (and maybe still) controversial “Gaia hypothesis”: the idea that the earth itself can be understood as a self-regulating organism. In one piece written in the late 1980s, Lovelock offered this redwood-themed explanation: 

Let us imagine that you are in a grove of giant redwood trees on the coast of California and that you are standing on the stump of a tree that has just been felled. When standing it was a vast tree weighing over 2,000 tons and over 100 meters tall, a spire of lignin and cellulose, a tree that started life over 2,000 years ago.

A strange thing about this tree is that during its life nearly all of it was dead wood. As a tree grows there is just a thin skin of living tissue around the circumference; the wood inside is dead, as is the bark that protects the delicate tissue. More than 97% of the tree we stand on was dead before it was cut down.

Now in this way a tree is very like the Earth itself. Around the circumference on the surface of the Earth is a thin skin of living tissue of which both the trees and we humans are a part. The rocks beneath our feet are like the wood, and the air above is like the bark. Both are dead matter, but the air and rocks, like the wood and the bark, are either the direct products of life or have been greatly modified by its presence. Is it possible that the Earth is alive like the tree?

Preston might have mentioned this intellectual history in The Wild Trees, but if he did, I’d long forgotten it. How fitting that redwoods should play a role in helping to give root to such a fertile and provocative idea, one that itself forces you to think across temporal scales. To see the earth as a kind of organism, Lovelock had to think about why its atmosphere had been stable for so many millions of years with such an unusual “far-from-equilibrium” mix of gases (most crucially for us aerobic creatures, a disproportionate amount of oxygen.) Thinking about the earth as a self-regulating organism demanded an understanding of the earth’s current atmospheric chemistry, but it also required some way of tracking the long-term trends as well.

One last link in the chain: Reading Jabr on Lovelock reminded me that I had written about some of these ideas many years ago in my 2008 book The Invention of Air, which is ostensibly a biography of the chemist and theologian Joseph Priestley, but which also features some deep time explorations of its own. In one chapter—the “intermezzo” of the book, as I called it—a story that has up to now been set in the 1760s suddenly winds the clock back a hundred million years or so, before zipping back to the 1770s and the scientific discoveries of Priestley and his sometime collaborator Benjamin Franklin. It has a whole riff on the oversized forms that life took on Earth during Cretaceous, including the redwood trees. I hadn’t actually sat down to read this chapter for at least ten years, and I’d remembered as a transitional bridge—very much embedded in the context of that specific story—but re-reading the intermezzo this summer made me think that it actually works remarkably well as a standalone essay. So I’m going to do a light edit and bring in a bit of the backstory from the book, and share it here at Adjacent Possible in the next week or two, the way I have done with a few other excerpts from the back catalog. Stay tuned for that…

[This is the opening chapter from my new book, The Infernal Machine: A True Story of Dynamite, Terror, and the Rise of the Modern Detective, now available in hardcover, e-book, and an audiobook that I narrated myself.]

July 5, 1915

Police headquarters, Centre Street, Manhattan

The bombs came in all kinds of packages.

Often they arrived in tin cans, emptied of the olive oil or soap or preserves the cans had originally been manufactured to contain, now wedged tight with sticks of dynamite. Sometimes they were wrapped with an outer band of iron slugs, designed to maximize the destruction, conveyed to their target location in a satchel or suitcase, “accidentally” left behind in the courthouse, or the train station, or the cathedral. Many of those devices were time bombs running on clockwork mechanisms. The more inventive ones utilized a kind of hourglass device, releasing sulfuric acid into a piece of cork, the timing determined by chemistry, not mechanics: how long the acid took to eat its way through the cork, until it began dripping onto the blasting cap below. Many were swaddled in old newspaper pages. One of the most notorious bombing campaigns sent the devices through the mail, dressed up in department- store wrapping.

And sometimes the bomb was just a naked stick of dynamite, with a fuse simple enough to be lit with the strike of a match, ready to be flung into an unsuspecting crowd.

Many bombs were delivered anonymously. But others were accompanied by missives sent to a local paper, or left on a doorstep: threats, intimations of further violence, delusional rants, and more than a few manifestos. The smaller bombs—the ones detonated by a storefront, a few notches up from fireworks—were the mobster version of an “account overdue” mailing: the big stick of the extortion business. A few came from clinically insane individuals without a cause, propelled toward the terrible violence of dynamite by their own private demons. But most of the explosions that made the national news during those years were expressions, implicit or explicit, of a political worldview.

The political bombers were a diverse bunch: socialist agitators, Russian Nihilists, Irish republicans, German saboteurs. But of all the bomb throwers of the period, no group was more closely associated with the infernal machines—as the press came to call the bombs—than the anarchists. The forty-year period during which anarchism rose to prominence as one of the most important political worldviews in Europe and the United States—roughly from 1880 to 1920—happened to correspond precisely with the single most devastating stretch of political bombings in the history of the West. Indeed, the whole modern practice of terrorism—advancing a political agenda through acts of spectacular violence, often targeting civilians—began with the anarchists.

What was anarchism, really? Start with the word itself. Today the word anarchy almost exclusively carries negative connotations of chaos and disorder. But when the political movement first emerged in the middle of the nineteenth century, the word’s meaning was much more closely grounded in its etymological roots: an-, meaning no, and -archos, the Greek word for “ruler.” The anarchists believed that a world without rulers was possible. At times, they convinced themselves that such a society was inevitable; imminent, even.

The anarchists maintained that there was something fundamentally corrosive about organizing society around large, top-down organizations. Human beings, its advocates explained, oftentimes at gunpoint, had evolved in smaller, more egalitarian units, and some of the most exemplary communities of recent life—the guild-based free cities of Renaissance Europe, the farming communes of Asia, watchmaking collectives in the Jura Mountains of Switzerland—had followed a comparable template, at a slightly larger scale. These leaderless societies were the natural order of things, the default state for Homo sapiens. Taking humans out of those human- scale communities and thrusting them into vast militaries or industrial factories, building a society based on competitive struggle and authority from above, betrayed some of our deepest instincts.

At its finest moments, anarchism was a scientific argument as much as it was a political one. It had deep ties to the new science that Darwin had introduced, only it emphasized a side of natural selection that is often neglected in popular accounts: the way in which evolution selects for cooperative behavior between organisms, what Peter Kropotkin—anarchism’s most elegant advocate—called “mutual aid.”

As a theory of social organization, anarchism was equally opposed to the hierarchies of capitalism and the hierarchies of what we would now call Big Government. For this reason, it lacks an intuitive address on the conventional left-right map of contemporary politics, which partly explains why the movement can seem perplexing to us today. Whatever you might say about Emma Goldman and Alexander Berkman and Peter Kropotkin—the three main anarchists in this book—they should never be mistaken for free-market libertarians. They wanted to smash the corporate regime as much as they wanted to smash the state.

But the other confusion about the movement lies in the language itself. The main reason that the word anarchy now carries the implicit connotation of troublesome disorder is because a century ago, a wave of anarchists insisted on blowing things up, again and again and again, in the name of the movement. “A girdle of dynamite encircles the world, not only the old but the new,” the anarchist Johann Most announced in the late 1800s. “The bloody band of tyrants are dancing on the surface of a volcano.”

That sense of unruly chaos that the word anarchy triggers in our mind today is the aftershock of all those explosions, part of the debris field they left behind. For the anarchists, it was arguably one of the most disastrous branding strategies in political history. They turned a word against their cause.

Why exactly were the anarchists so intent on blowing things up? That is, by definition, a technological and scientific question as much as it is a question about radical ideologies: How did anarchism and dynamite—born in the same decade but otherwise unrelated—come to be so closely intertwined? Dynamite gave small bands of humans command of more energy per person than they had ever dreamed of having before. Dynamite, quite literally, gave them power. The anarchists happened to be the first political movement to embrace that new power. But why were they compelled to make that choice? Could they have made a more persuasive case through less destructive means? To even begin to answer those questions, we need to understand where the anarchist’s appetite for political violence originally came from, its complex symbiosis with the everyday violence that industrialization had unleashed into the world. For every death at the hand of a bomb-wielding anarchist, a hundred or more would die from factory accidents.

We also need to understand what that appetite for violence—enabled by the energy density of the dynamite-based explosion—helped bring into the world. When the anarchists began dreaming of a society unfettered by institutional authority, there were no forensic detectives, no biometric databases of identity, no anti-terror agencies. Where official police forces did exist, they were usually in bed with urban crime syndicates and political machines; national and international investigatory bodies like Interpol or the FBI or the CIA were decades from being created. But in the end it turned out to be those institutions that triumphed over the stateless dream of the anarchists. In many key respects these techniques and organizations were prodded into being by the emerging threat of the infernal machines, like an immune response to an invading virus. The innovation of dynamite-driven political terrorism created a counterreaction from the forces of top-down authority, one of those stretches of history where some of the most powerful institutions in the world are shaped by the activities of marginal groups, working outside the dominant channels of power. In this case, though, the legacy of the anarchist movement ultimately possessed a kind of tragic irony: the dream of smashing the state helping to give birth to a regime of state surveillance that would become nearly ubiquitous by the middle of the twentieth century.

In the summer of 1915, the site in the United States that best represented that new regime was the Identification Bureau of the New York Police Department, created originally by a cerebral detective, Joseph Faurot, and eventually overseen by Commissioner Arthur Woods, a well-born Bostonian turned social reformer. The bureau was on the ground floor of the NYPD headquarters in Lower Manhattan, lined with file cabinets containing tens of thousands of photographs and fingerprints, organized by intricate classification schemes. In a predigital era, the Identification Bureau was the closest thing imaginable to the U.S. government’s plan for “Total Information Awareness” that would become so controversial in the months after 9/11.

The Identification Bureau had an equally revolutionary idea at its core, one that had first developed in Paris and London at the end of the nineteenth century before Faurot and Woods brought it stateside: the idea that crime and sedition were fundamentally problems that could be solved with data. The way to combat individuals or groups who were intent on disrupting society was not to overwhelm them with physical force. Such naked expressions of power only inflamed the passions of the radicals. It was better to contain dissent through more subtle means: file cabinets filled with information, undercover operations, a web of invisible oversight stretching across the country and, increasingly, across the world.

The Infernal Machine, then, is the story of two ideas, ideas that first took root in Europe before arriving on American soil at the end of the nineteenth century, where they locked into an existential struggle that lasted three decades. One idea was the radical vision of a society with no rules—and a new tactic of dynamite-driven terrorism deployed to advance that vision. The other idea—crime fighting as an information science—took longer to take shape, and for a good stretch of the early twentieth century, it seemed like it was losing its struggle against the anarchists. But it won out in the end. How did that come to happen? And could the story have played out differently?

The history of the struggle between those two ideas involves a global cast of some of the most fascinating characters of the age: most of all Berkman, Goldman, Kropotkin, Woods, and Faurot. But doing justice to that story demands that we take a wider view  of the historical timeline: venturing back to the original invention of dynamite itself and its first deployments as a political weapon  in czarist Russia, the growth of anarchism as a political worldview in the late 1800s, the pioneering innovations of forensic science in Paris that evolved in part to counter that growth—all the way up to a terrifying, but now mostly forgotten, stretch of New York City’s history in the early twentieth century, when the metropolis experienced thousands of bombings over the course of just two decades.

If you had to select the one point on that timeline that marked the apex of the struggle between anarchism and the surveillance state, the point where you might get even odds as to how it was all going to turn out, you could make a good case for the night of July 5, 1915. Despite the late hour, the Identification Bureau was bustling with activity. A bomb had detonated two days earlier in the U.S. Capitol building; the financier J. P. Morgan, Jr., had been attacked at his home in suburban Long Island the following morning; and the detectives had just discovered that the suspect in both crimes had recently purchased two hundred sticks of dynamite in New York, only six of which had been accounted for. For weeks Joseph Faurot had been receiving death threats in the mail from anarchist groups, reminding the detective of the fast approaching one-year anniversary of one of the most devastating explosions in the city’s history, a blast that destroyed an entire apartment building on the East Side, the work of anarchists plotting an attack on another titan of industry. That damage had been wrought with only a few sticks of dynamite. The trove of explosives currently missing threatened to make the previous year’s blast look like a bottle rocket by comparison.

But the clash between the anarchists and the NYPD was not only visible in the frenetic activity inside the Identification Bureau itself. To see it in its full scope, you needed to leave the file cabinets and the fingerprint studios behind, walk out the plate glass doors into the hall, venture down a set of fire stairs into the darkened hallways of the basement. There you would have seen a cheap suitcase, leaning against a doorway. Below the muffled hum of activity in the Identification Bureau directly above, if you listened very intently, you might just have heard the quiet metronome of a ticking clock.

I hope you enjoyed that taste of The Infernal Machine, now available in hardcover, e-book, and an audiobook that I narrated myself.

If you’re interested in hearing more about the main events and themes of The Infernal Machine, this new episode of the Next Big Idea podcast is a great place to start. (Apple version; Spotify version.) It’s a very Adjacent Possible conversation actually—it starts with some of the narrative of Infernal but then becomes more of a conversation about complexity: complexity in storytelling, in our understanding of history, even in songwriting. (And it ends with a discussion of how tools like NotebookLM may make it easier to explore complex ideas and historical events.)

One final note: as many of you know, there is a premium “Ideal Reader” subscription tier for Adjacent Possible that includes a signed copy of each new book I write, in addition to access to all the paid content here. I’m going to be sending out those books over the next few weeks so, if you’ve signed up for that tier and your address has changed since the original mail, or you’d like to request a special dedication, please drop me a line (or reply to this email) at sbeej68@gmail.com.

Subscribe now

Last week I had the honor of delivering the commencement address at my youngest son's high school graduation, which was, as you might imagine, a bittersweet occasion. In the commencement talk, one of the ideas I touched on briefly was something I've mentioned before here at Adjacent Possible: the somewhat paradoxical idea that, thanks to the AI revolution, we are entering a period where it will be a great time to be a humanities major with an interest in technology. The conventional story of course is that we're in the middle of a mass exodus from English or History as majors—sometimes blamed on the excesses of cultural theory, sometimes on the fact that all the money is in Computer Science and Engineering right now. And to be sure, the exodus is real. But there is a case to be made that college and grad students are over-indexing on the math and the programming, just as the technology is starting to demand a different set of skills.

The simple fact of the matter is that interacting with the most significant technology of our time—language models like GPT-4 and Gemini—is far closer to interacting with a human, compared to how we have historically interacted with machines. If you want the model to do something, you just tell it what you want it to do, in clear, persuasive prose. People who have command of clear and persuasive prose have a competitive advantage right now in the tech sector, or really in any sector that is starting to embrace AI. Communication skills have always been an asset, of course, but thanks to language models they are now a technical asset, like knowing C++ or understanding how to maintain a rack of servers.

This is, of course, a variation on Andrej Karpathy's quip from more than a year ago: "The hottest new programming language is English." But it's more than that, I think. The core skills are not just about straight prompt engineering; they're not just about figuring out the most efficient wording to get the model to do what you want. They also draw on deeper, more nuanced questions. What is the most responsible behavior to cultivate in the model, and how do we best deploy this technology in the real world to maximize its positive impact? What new forms of intelligence or creativity can we detect in these strange entities? How do we endow them with a moral compass, or steer them away from bias and inaccurate stereotypes? Can language alone generate a robust theory of how the world works, or do you need more explicit rules or additional sensory information?

All of those questions have been absolutely central to the discussion of AI for the past two years, but if you think about it, they were all questions that belonged to the humanities until the language models came along: ethics, philosophy of language, political theory, history of innovation, and so on.

I don't want to carry this argument too far. Some of my training as a writer has come in useful in creating NotebookLM, through the design of our core prompts and the overall “voice” of the product. But Notebook itself would not exist without the exceptional engineering talent of our extended team, from our front-end and back-end programmers to the people who built the enormously complex infrastructure of the models themselves. Perhaps someday it will be possible for a code-illiterate person like myself to conjure an entire application into being just by describing the feature set to a language model, but we are not there yet. And of course building the models themselves will almost certainly continue to require skills that are best honed in engineering and computer science programs, not writing seminars.

But I do think it is undeniable that the rise of AI has ushered humanities-based skills into the very center of the tech world right now. In his last product introduction before his death, Steve Jobs talked about Apple residing at the intersection of the liberal arts and technology; he literally showed an image of street signs marking that crossroads. But the truth is back then most of the travelers on the liberal arts avenue were designers. There wasn't as much need for philosophers or ethicists or even writers in building the advanced consumer technology of that era. But now those skills have a new relevance.

There's a wonderful illustration the kinds of skills that are now at a premium in the conversation I had a few days ago with Dan Shipper for his AI & I podcast (best viewed on video so you can see what's happening on screen.) Dan had me on the show—formerly called How Do You Use ChatGPT?—to walk through some of the new features that we just launched yesterday, in addition to our international rollout, and to generally just get a sense for how I have integrated NotebookLM into my creative workflow. He likes to use each episode to try to create something spontaneously with the guest —it's very live and unscripted, which is always a little nerve-wracking when you are working with new features (not to mention stochastic language models.) But in this case it made for a beautiful collaboration.

The sample use case I brought to the show was a notebook that I had filled largely with interview transcripts from the NASA oral history project. The notebook has something like 300,000 words of interviews with astronauts, flight directors, and other folks from the Apollo and Gemini programs. (That may sound like a lot, but you can now have up to 25 million words worth of sources in a single notebook.) I also had a few slide decks in there with images, since NotebookLM now supports Slides and has true imagine/chart/diagram understanding as well. We also, as you will see, ended up diving into my collection of reading quotes, which I maintain in a separate notebook. Dan and I decided that we would try to use this notebook to gather ideas for a potential documentary project about the Apollo 1 fire that tragically killed three astronauts in early 1967. I recommend watching the video starting around the thirty-minute mark where we really dive into the exercise -- I think it's probably the best example to date of the kind of high-level creative and conceptual work that NotebookLM makes possible, where the software is truly helping you make new connections and synthesize information far more easily than would have been possible before.

But the thing I also want to draw your attention to is how much Dan is driving the process, by suggesting a series of prompts that ultimately elicit some astonishing—even to me—results from NotebookLM. I was going into it more or less planning on showcasing NotebookLM's ability to extract and organize a complex string of facts out of a disorganized collection of source material, like creating a timeline of all the events associated with the fire, or suggesting key passages that I could read to understand the impact of the fire. (All with our new inline citations, which are pretty amazing in their own right.) But at a certain point, Dan really takes the wheel, and says, effectively: "This is a Steven Johnson project, and so it's got to have some surprising scientific or technological connection that the reader/viewer wouldn't expect; let's ask NotebookLM to help us find that angle." And then we just go on a run—again, largely driven by Dan's prompting—that takes us to some pretty amazing places, and even generates the opening lines of a script by the end of it.

What you can see in this sequence are two things: 1) a remarkably capable language model doing things with a large corpus of source material that would have been unthinkable just a year ago really—but just as importantly 2) a very smart human being who knows how to probe the source information and unlock the skills of the language model to generate the most useful and interesting results. The skill that Dan displays here is basically all about being able to think through this problem: Given this body of knowledge, given the abilities and limitations of the AI, and given my goals, what is the most effective question or instruction that I can propose right now? I don't know whether you're better off with a humanities background or an engineering background in developing that talent, but I do believe it has become an enormously valuable talent to have.

The other thing worth noting in the exchange—and I take a step back to reflect on it in the middle of the exercise—is the range of intelligences involved in the project. On the one hand you have the intelligence of all the astronauts and flight directors contained in the interview transcripts themselves; you have the intelligence of all the authors whose quotes I have gathered over the past two decades of research and reading; you have the intelligence of two humans who are asking questions and steering the model's attention towards different collections of sources, crafting prompts to generate the most compelling insights; and then you have the model itself, with its own alien intelligence able somehow to take our instructions and extract just the right information (and explain its reasoning) out of millions of words of text. I used to describe my early collaborations with semantic software as being like a duet between human and machine. But these kinds of intellectual adventures feel like a chorus.

This idea of the model not as a replacement for human intelligence, but instead tool for synthesizing or connecting human intelligence seems to be gathering steam right now, which is good to see. The artist Holly Herndon made a persuasive case for calling artificial intelligence "collective intelligence" in a recent conversation with Ezra Klein. My friend Alison Gopnik has been talking about AI for a long while as a "cultural technology," which adds weight to the prediction that humanities skills will have increasing relevance in a world shaped by such technologies. In a recent conversation with Melanie Mitchell in the LA Review of Books, Alison argued:

A very common trope is to treat LLMs as if they were intelligent agents going out in the world and doing things. That’s just a category mistake. A much better way of thinking about them is as a technology that allows humans to access information from many other humans and use that information to make decisions. We have been doing this for as long as we’ve been human. Language itself you could think of as a means that allows this. So are writing and the internet. These are all ways that we get information from other people. Similarly, LLMs give us a very effective way of accessing information from other humans. Rather than go out, explore the world, and draw conclusions, as humans do, LLMs statistically summarize the information humans put onto the web.

It’s important to note that these cultural technologies have shaped and changed the way our society works. This isn’t a debunking along the lines of “AI doesn’t really matter.” In many ways, having a new cultural technology like print has had a much greater impact than having a new agent, like a new person, in the world.

Another way to put that—which I will probably adapt into a longer piece one of these days—is that language models are not intelligent in the ways that even small children are intelligent, but they are already superhuman at tasks like summarization, translation (both linguistic and conceptual), and association. And when you apply those skills to artfully curated source material written by equally, but differently, gifted humans, magic can happen.

PS: As a little bonus treat, here’s a glimpse of our new inline citations (with image/chart understanding) live in my NASA notebook. You can see how NotebookLM answers a question about NASA budgets by drawing information from a chart included in a slide deck, and actually cites the chart itself in the citation.

And here’s a very sophisticated French notebook, part of our expansion to more than 200 countries and territories around the world: