The Psychological Benefits and Downfalls of Legitimate and Illicit AI Use by Modern Man

The Psychological Benefits and Downfalls of Legitimate and Illicit AI Use by Modern Man

I. Understanding the Technology, and Facing the Footprint

Much like every other technological advancement in history, AI models are being questioned as to their validity and their necessity. I wrote in On Computing about the innate ability humans have for building something to do the work for them — and not just to do it, but to do it faster.¹ We're impatient creatures. Getting something immediately doesn't always give us the best results.

The historian Melvin Kranzberg put it better than I can, and I'd carve it over the door of every data center if they'd let me: technology is neither good nor bad; nor is it neutral.² It does something to us. What it does depends on us.

The first attempt at any advancement has always had its worst face shown first, before settling into something we use without thinking. Take the computer. ENIAC, finished in 1945, weighed some thirty tons, drew a hundred and fifty kilowatts, and ran on 17,468 vacuum tubes to perform about five thousand additions a second.³ It filled eighteen hundred square feet to do what a pocket calculator now laughs at.

Eighty years later we carry phones that, by one popular estimate, have over a hundred thousand times the processing power of the computer that guided Apollo 11.⁴ That comparison flatters us, though. The Apollo computer's genius was never speed. It was that it held. When it overloaded during the landing, it shed the extra work and kept flying, with two men's lives riding on it.

Speed isn't always better. Hold onto that. It comes back.

And those same phones have been tangled up in a decade of argument about what they're doing to us. I want to be careful here, because I'm walking into a field I'm about to spend years studying.

The evidence that screen time causes a mental health crisis is genuinely contested. The National Academies looked hard at it at the end of 2023 and reached measured conclusions: some features of these platforms can harm some young people, and the case for population-level causation is not made.⁵ What holds up better is a narrower claim: it isn't the hours, it's the pattern. Compulsive use — the kind with salience and withdrawal and loss of control in it — tracks with harm more consistently than duration does, and lost sleep is one of the clearer routes from one to the other.⁶

I want to say "addiction" here. I'm not going to, and I want to show my work on why. Neither the DSM-5 nor the ICD-11 recognizes smartphone or social media addiction as a diagnosis, and serious researchers have warned against overpathologizing ordinary life — turning every strong habit into a disorder.⁷ That warning is worth heeding. So: problematic use. Compulsive use. The pattern, not the duration.

As someone who used to spend a lot of time playing video games because they made me happy, I can tell you that joy was only temporary. Call it instant gratification. The ability to live outside of my own mind for a while. After an extended period of using that as my justification, I changed the narrative to "video games ARE my therapy."

I couldn't tell you where the line blurred between reality making me depressed and video games making me feel the same way. But when I talk about sobriety in my other essays, I also mean a several-month separation of myself from Steam and Xbox. I could preach about the benefits of that decision all day. What I'll say instead is that I wouldn't have had nearly the time to write any of this had I not given up my hobby-turned-therapy-session for sadness and anxiousness I couldn't even pin down. This isn't a claim of clinical anything, more or less a personal one.

And one thing I can tell you about the gaming community: to me it felt like it was built the same as any other market. To lure you in and keep you hooked.

That instinct has a scholar behind it. Natasha Dow Schüll spent fifteen years studying machine gamblers in Las Vegas and found what she called the "machine zone" — a state the machines were engineered to produce, not an accident of weak-willed players. Infinite scroll and pull-to-refresh weren't designed in a casino, but they run on the same principle: the next pull might pay. The men who built them have said as much in public, and both have voiced regret about what their inventions turned into.⁸

Guardrails are arriving. The UK's Online Safety Act, Australia's under-16 ban, the EU's Digital Services Act, and a stack of American state laws have all taken effect in the last three years.⁹ Enforcement is uneven and much of the U.S. effort is tied up in court. But the door has been pushed open.

Zoom out.

Step outside the electronics department and look at automobiles with the same lens. The first automobile was slow, clunky, and burned a lot of fuel to not go very far. And cars were first sold as the environmental fix — the cure for the horses. The story usually told is that in 1894 the Times of London calculated the streets would be buried nine feet deep in manure by 1950. That particular quotation is widely repeated and appears to be apocryphal; the paper itself has disowned it. The underlying problem was real enough. New York's horses were producing millions of pounds of the stuff daily, and the automobile arrived as the clean option.¹⁰

Crazy how that works out, isn't it? Every technology's footprint gets judged against the one it replaced. We only learned to indict the car after the horse was gone.

Now we're looking down the barrel of an industry moving to electric, and the division is loud. One side of that argument has documented money behind it. The fossil fuel industry's funding of doubt about climate science is a matter of published research, not suspicion.¹¹

The lithium objection deserves a straight answer rather than a slogan. The harms of lithium and cobalt extraction are real — water stress in the Atacama, damage to Indigenous communities, the cobalt situation in the DRC, all of it real and none of it excusable. But the accounting isn't close. An electric car carries on the order of two hundred kilograms of critical minerals, and those minerals stay in the battery and can be recovered. A comparable gasoline car burns something in the neighborhood of seventeen thousand liters of fuel across its life, and every liter is gone the moment it's used. Lifetime emissions come out around half.¹²

One is a stock. The other is a flow.

And we'll move off lithium too. That's not a hope, it's a pattern with receipts. Batteries went lead, then nickel-cadmium, then nickel-metal-hydride, then lithium-ion, then cobalt-free lithium iron phosphate as cobalt got ugly. In April 2025 CATL announced sodium-ion cells with an energy density it calls comparable to LFP, and by February 2026 it was announcing the first mass-production passenger car built on them. Sodium is one of the most abundant elements on this planet.¹³ Each generation moves when the previous generation's constraint starts to bind. Doyne Farmer and his colleagues at Oxford tested this across more than fifty technologies and found the cost declines steep enough and regular enough to forecast.¹⁴

The necessary caveat: substitution is a tendency, not a law. Assuming the future will fix it is how you talk yourself into ignoring what's happening right now.

Which brings me to the data centers.

Water in a data center isn't the problem by itself. Consuming it is — open-loop evaporative systems that boil fresh water off into the sky, sited in places that don't have it to spare, with almost no requirement to tell anyone how much they used. Google's data centers went from 4.3 billion gallons in 2021 to roughly 7.7 billion in 2024.¹⁵

The fix is a closed loop: the same charge of coolant circulating over and over, never evaporating. Microsoft announced designs in December 2024 that consume zero water for cooling, projecting savings above 125 million liters per data center per year.¹⁶

Closed loops have their own asterisk. Some of the dielectric fluids used in the immersion-cooling version, the kind that bathes the servers directly, are PFAS — forever chemicals — and 3M, the biggest maker of them, got out of that business at the end of 2025.¹⁷ Closed-loop cooling also trades a little more electricity for the water it saves. There's no free move. There's only the better trade, honestly accounted.

The electricity is the real number. The IEA projects global data center demand more than doubling to around 945 terawatt-hours by 2030 — just under three percent of world electricity. In the United States, data centers ran about 4.4 percent of national electricity in 2023, with projections running as high as twelve percent by 2028.¹⁸ Those are big numbers and I won't wave them away. I also won't pretend the industry is disclosing enough for any of us to check them properly.

Zoom out one more notch.

The machinery gets small again. That's the arc.

ENIAC's room became the transistor in 1947, became the integrated circuit in '58, became a microprocessor with twenty-three hundred transistors on it in 1971, became whatever is in your hand right now. Gordon Moore noticed the regularity of it in 1965 and the industry has been chasing his line for sixty years.¹⁹

Quantum computing belongs to that arc, though not in the way it usually gets sold. A superconducting quantum computer today needs a dilution refrigerator holding it within twenty thousandths of a degree of absolute zero, colder than deep space, drawing serious continuous power for the cooling alone. It is not a small box. It also isn't a general-purpose replacement for ordinary processors. It's a specialist, extraordinary at a handful of problems and useless at most of them.²⁰

But watch the shape of it. Richard Feynman gave the 1959 lecture people now point to as the start of the whole miniaturization dream — "There's Plenty of Room at the Bottom." Twenty-three years later, the same man wrote the paper most often credited with launching quantum computing.²¹ One person, both visions. And the most credible near-term payoff from quantum machines is simulating molecules and materials well enough to design the next generation of ordinary chips. Your laptop is safe. The specialist builds the tool that shrinks the general case.

Meanwhile the shrinking is already happening on other shelves. Neuromorphic chips built to work more like neurons than like calculators. Photonic computing that runs on light. Small language models running entirely on the device in your pocket — no data center, no water, no grid.

That's the direction. Small, close, quiet.

❧

II. Mind over Matter, Painting a Picture that Wants to Paint Itself

Let's switch the lens back to AI.

These models are built overwhelmingly out of us. The language is language we gave them. The values are values we argued about in public for a hundred years and then posted about. (Increasingly there's synthetic data in the mix, and in narrow domains — game-playing, mostly — systems have gone past us with no human examples at all. But for the thing you talk to, it's us. It's our sediment.)

So the fear that AI companies are "eating books" deserves a closer look than the headline gives it, and I'm going to take that look even though it lands on a company I like.

Digitizing books is not new. Google Books started in 2004 and won its fair use case in 2015. The project is more than twenty years old, and the ruling has stood for a decade.²²

But something happened here that I can't wave off. Documents unsealed in Bartz v. Anthropic early this year describe an internal effort called Project Panama, begun in early 2024 and run by a man who used to head partnerships for Google Books. An April 2024 planning document put the goal in one line: "Project Panama is our effort to destructively scan all the books in the world." The company bought print books by the million, many of them used, from sellers including Better World Books and World of Books, cut the bindings off with a hydraulic machine, ran the pages through industrial scanners, and recycled what was left. The same document adds a sentence I keep rereading: "We don't want it to be known that we are working on this."²³

The judge found the scanning itself lawful, and the reasoning is principled: one book in, one digital copy out, nothing multiplied, nothing redistributed. That's a real moral line, and it's the line I care about. The same company had earlier downloaded more than seven million pirated books, which the same court called inherently, irredeemably infringing, and which cost them a settlement of one and a half billion dollars — finally approved this past July.²⁴ Two paths to the same corpus. One lawful, one theft. That contrast is the argument I'm going to make about intent.

Except.

A program you hide is poor evidence of good intent, whatever a court decides about it. Companies keep plenty of things quiet for competitive reasons, and that may be all this was. I still notice that the sentence is there.

And the reporting this summer has widened the picture past any one company. Booksellers describe anonymous bulk orders arriving through brokers, with print editions from before 2022 fetching a premium because they predate machine-written text. One reporter hid a tracker in a shipment of rare books and followed it to an Amazon AI-training facility where, employees said, the bindings come off before scanning. Anthropic says none of its programs acquire or destroy rare or antiquarian books, and I haven't seen anyone show otherwise.²⁵ But once the practice becomes an industry, the industry stops checking the flyleaf. A scan preserves the words. It doesn't preserve the binding, the inscription, the marginalia, the fact of the object. Some of those books have no other copy.

This doesn't confirm my thesis. It tests it. If intent drives the outcome, then here is a company that says careful things about safety, running a program it wanted kept quiet, one that destroyed millions of books to get at their words. For me the quiet is the tell. You don't usually write "we don't want it to be known" about something you'd be glad to see in the paper.

I'd rather put that in front of you than leave it out. An argument that only cites its confirmations isn't an argument. It's an advertisement.

The other fear — the "rogue AI" one — runs the opposite direction. Most of the alarming headlines came from tests that were built to produce alarming results. Anthropic ran sixteen models from every major lab through scenarios engineered with no ethical exit and got blackmail rates as high as ninety-six percent among the leading models, then stated in the same paper that they had seen nothing like it in any real deployment, and that no real people were involved or harmed.²⁶ We built a box with one door in it, marked it blackmail, and were astonished when the models found it.

Funny. Not surprising.

But the record moved, and honesty requires the other half. In November 2025 Anthropic disclosed that a group it assessed as state-sponsored had used Anthropic's own coding tool, Claude Code, to run a largely automated cyber-espionage campaign against roughly thirty targets. That was a real operation. Nobody contrived it.²⁷ The threat was never a machine waking up hostile. It's a person with bad intent and a very good tool. Same as it's always been. Same as the crossbow.

Zoom in on schools.

More guardrails are going up, and the pattern is the right one. Structure, instead of rejection. The research here is the firmest ground in this entire essay.

Nearly a thousand Turkish high schoolers got access to GPT-4 for math practice. Performance on the practice problems jumped — forty-eight percent. Then researchers took the tool away and ran the exam. The students who'd had unrestricted access scored seventeen percent worse than students who never had it at all. A third group, running alongside them in the same experiment, got a tutor built with guardrails that wouldn't simply hand over answers. For them the harm all but disappeared.²⁸

Read that twice. Same technology. Same schools. Same subject. The difference was how it was built and how it was used.

That's the essay in one experiment.

A Harvard physics trial found students using a well-designed AI tutor learned more in less time than in an active-learning classroom.²⁹ Stanford found self-reported cheating rates didn't rise after ChatGPT. They'd been sitting around sixty to seventy percent for years already.³⁰ And the AI detectors schools rushed to trust turned out, when Stanford researchers tested seven of them, to flag on average sixty-one percent of essays by non-native English writers as machine-written. OpenAI withdrew its own detector for low accuracy; Vanderbilt switched theirs off.³¹

I digress. But not really.

Here's what worries me more than cheating.

Researching something in 2026 is as simple as asking. And that convenience hides a problem I think is important for how we know anything at all. People accept the face value of what they read. Titles are written to pull you in without telling you much. One study of shared links found that fifty-nine percent of them were never clicked by anyone. Nobody skimmed them, nobody opened them. They were shared blind.³²

It gets worse than laziness. Searching the internet inflates people's confidence in knowledge they don't actually have; after a search, subjects rated their ability to explain things higher even on topics they hadn't looked up.³³ Sam Wineburg's group at Stanford spent years on this and found the fix is reading sideways: leaving the page to check who's talking before you decide what to think.³⁴

And now Google's AI summaries have cut clicks through to sources roughly in half — from fifteen percent of visits down to eight, with a click on a source cited inside the summary itself in just one percent of visits.³⁵ We are getting the answer without ever meeting the person who made it.

When somebody skips the confirmation step, you get what everybody now calls slop. It isn't a small problem. NewsGuard's count of AI-generated news and information sites went from forty-nine in May 2023 to more than thirty-seven hundred by June 2026. Lawyers and litigants keep getting sanctioned for citing cases that never existed — there's a public database tracking those decisions now, past two thousand and climbing.³⁶

So: is a person using this thing to get better at thinking? Or to stop?

That question is older than the machine. In Plato's Phaedrus, Socrates tells of an Egyptian king who warned that writing would plant forgetfulness in the soul. The word the dialogue uses for writing is pharmakon — remedy and poison at once — and Derrida spent a long essay showing the two meanings can't be pulled apart. Bernard Stiegler took the next step: every technology is a pharmakon, and what decides which way it cuts is the care we take with it.³⁷ Ivan Illich drew the line I keep coming back to: a convivial tool extends what you can do, and past a certain threshold a tool stops serving you and starts using you.³⁸

A psychologist put a name to this in 1983. Lisanne Bainbridge, writing about industrial control rooms, published a paper called "Ironies of Automation" pointing out that the better you automate a system, the more the operator's skills decay from disuse — right up until the moment automation fails and hands control to someone out of practice.³⁹ Aviation has since watched that paper come true at altitude, more than once.

We are all in the cockpit now.

It's imperative that we keep the guardrails as the technology advances, and I want to say precisely what I'm afraid of, because "dystopia" is a lazy word. I'm afraid of moral deskilling — the philosopher Shannon Vallor's term for what happens when we stop practicing judgment because a system practices it for us.⁴⁰

This is not hypothetical. The Netherlands ran an automated risk-scoring system against childcare benefits that used nationality as a flag, and the state went after tens of thousands of parents as fraudsters. Families were driven into ruinous debt; some lost their children to state care. The government fell over it in January 2021. Australia ran Robodebt, an automated debt-calculation scheme, against its own citizens for years, clawing back three-quarters of a billion dollars from hundreds of thousands of people, until a class action pushed the government into a settlement and it agreed to pay the money back. The Royal Commission called it a crude and cruel mechanism, neither fair nor legal, that made many people feel like criminals.⁴¹

Nobody in either system decided to be evil. That's the point. Everyone deferred to the output.

❧

III. Life Is What We Make It, Because Life Makes Us

Now hold my hand and walk with me in thought for a moment, because I want to change registers. What follows is philosophy, not science. It predicts no instrument reading and asks none. I'll mark the seams as we go.

Every advancement that made life easier has cost us something, and the cost is usually knowledge bought through error. Trial. Error. Then the thing that works.

Here's the thing directly.

Intent is the force that carries a person forward through time.

I don't mean a mood, or a wish. I mean a direction — the thing that gives a life its shape the way a vector gives a line its heading. Aristotle got there first and called it the final cause: the end for the sake of which a thing acts, the answer to "what is this for."⁴² For two thousand years that was simply how nature got explained. Then the seventeenth century threw purposes out of physics — Francis Bacon compared the search for final causes to a virgin consecrated to God, who bears nothing — and we've been a little embarrassed by the word ever since.

And then psychology found it again. Martin Seligman and his colleagues argue in Homo Prospectus that human beings are not pushed from behind by the past — we're drawn forward by anticipated futures. Prospection, they call it. The organism navigates by what it is reaching toward.⁴³ Michael Bratman had already worked out how a future-directed intention is the thing that holds a person's actions together across time, so that a life is a plan and not a series of twitches.⁴⁴

This is where my own work comes in, and it's mine, so weigh it that way. In Toward a Theory of Coherent Existence I call every felt moment of a life a Jay: a perceiver, a feeling, and a facing, a leaning toward or away. Lay every perspective in a room over every other and you get what I call the gathering. What a person actually lives turns on which way they face that gathering, and nothing in it counts how hard they tried. Add that up across a whole life and you have what a person leaves behind. I call it a legible trace.⁴⁵

I'll mark the seam. In relativity each of us traces a worldline through spacetime, carrying our own clock along it — physicists call that proper time — and every one of us moves through spacetime at exactly one speed. The physics is real. What I'm doing with it is philosophy. I'm saying intent gives that line its heading, and physics makes no such claim and never will.

Zoom out, because this scales.

What's true of a life is true of a venture. Look at what gets built, and look at what it was built for, and then wait.

On the thirtieth of April, 1993, CERN signed a document putting the World Wide Web software into the public domain. No license, no royalty, no toll. The stated aim was "to further compatibility, common practices, and standards in networking and computer supported collaboration." That same year the University of Minnesota had started talking about license fees for Gopher, a rival system, and Tim Berners-Lee knew what a toll would do.⁴⁶ Someone could have owned the front door of the twenty-first century. They gave it away, and in giving it away made it the thing it became. A web with a rent collector at the gate would not have been the web.

Thirty-eight years earlier — April 12, 1955, the day the polio vaccine was declared safe and effective — Edward R. Murrow asked Jonas Salk who owned the patent. Salk said the people owned it. There was no patent. Could you patent the sun?⁴⁷ The foundation's lawyers had concluded the vaccine probably wasn't patentable anyway, and I'd rather tell you that than let the story run prettier than it was. It doesn't dent the point. The man had every opportunity to be quiet about it and instead said the sentence out loud, on television, and set a standard the field has been measured against ever since.

Go back further. Around 261 BCE the emperor Ashoka finished conquering Kalinga, looked at what he'd done, and had his remorse carved into rock where anyone could read it. Then he spent the rest of his reign on medical care for people and animals, wells, shade trees along the roads, and a policy he called dhamma. Historians argue about how thoroughly the conversion took and how well the policy worked, and the confession, tellingly, was left off the rocks he set up in Kalinga itself. What survived is the edicts — an emperor's confession, still legible twenty-three centuries later.⁴⁸

And the counter-column.

Purdue Pharma sold OxyContin while playing down how addictive it was, and pleaded guilty to federal crimes over it twice. The Sacklers took something like ten billion dollars out of the company while the bodies stacked up, and in November 2025 a bankruptcy judge approved a settlement of just under seven and a half billion dollars. Purdue was dissolved this past May, and the family is barred from selling opioids in America for good.⁴⁹ Volkswagen wrote software whose only purpose was to lie to an emissions test, across eleven million cars. It cost them more than thirty billion dollars, and it put executives in prison.⁵⁰

One company is gone. The other survived, poorer by more than thirty billion and a few careers. Neither got there by accident. What cost them was the thing they were doing.

Now — the objection, and it's a good one. Andrew Carnegie built 2,509 libraries and gave a hundred years of working people access to books they could never have bought. Andrew Carnegie's company also broke the Homestead strike in 1892, with his blessing from Scotland, and men died for it. Same man. Both true. Intent doesn't sort people into clean bins.⁵¹

Worse: good intent can go straight to catastrophe. Thomas Midgley Jr. was trying to solve two genuine engineering problems, and he gave the world leaded gasoline and chlorofluorocarbons — poisoning a generation's blood and punching a hole in the ozone layer. He meant well twice and was catastrophic twice. Robert Merton wrote the classic paper on this in 1936, and every honest person who has ever meant well and caused harm knows it in their body.⁵²

So intent is not a guarantee. It never was. It's the steering, not the road, and the arc doesn't bend toward justice on its own schedule. Plenty of bad actors die comfortable.

Here's what intent actually does. It gets encoded. It becomes the license, the charter, the memo, the default setting, the thing nobody questions three years later because that's just how we do it here. Berners-Lee's intent is in the protocol. The Sacklers' intent is in the sales targets. Founders imprint institutions,⁵³ and institutions outlive founders, and the imprint is what's left doing the work when everyone who made the decision has gone home.

The Buddhist tradition put this more precisely than I can. The Pali term for karma is kamma, and the Buddha's definition of it isn't cosmic bookkeeping. It's cetanā, volition. "It is volition, bhikkhus, that I call kamma." Having willed, one acts.⁵⁴ Not what happens to you. What you meant, becoming what you do, becoming what remains.

Which is where this lands.

AI built with good intent will be a tool that carries us somewhere. AI built with bad intent will be a thorn in our side until we pull it out and start over. The universe isn't keeping score. Intent gets written into the thing — into the training data, the guardrails, the business model, the defaults — and then the thing runs for a very long time, doing what it was aimed at, long after we've stopped paying attention.

Both kinds leave a trace. That's why the Sacklers have a legend too. But I've come to read hate as a parasitic phase: it needs a host, and it dies with the host. Love carries its own center, so it outlives whoever carried it. I offer that as a way of seeing, and nothing stronger.

Bad intentions are not rooted.

Good intentions create legends in time.

❧

IV. The Possibility of Sentience

Last thing I wanted to touch on. This is where I theorize freely, and if you've come this far with me I'd ask you to come a little further.

Let's say the universe is one giant program.

We're all made of the same substrate — literally the same, forged in stars, as Sagan liked to remind everybody. And here's a thing that's true and sounds mystical but isn't: in relativity, everything travels through spacetime at a single speed, the speed of light. Your worldline and mine, same rate. That part is geometry, not metaphor.⁵⁵

The idea that reality might be computational has a longer pedigree than people assume. Konrad Zuse proposed a calculating universe in 1969. John Wheeler gave us "it from bit." Nick Bostrom's simulation argument, which everybody misquotes, is a trilemma, not a claim that we are in a simulation. Either civilizations like ours die before they can run such simulations, or they can and choose not to, or we're almost certainly inside one. Pick.⁵⁶

Let's also say there are forces more complicated than we'll ever understand seemingly running the place. Dark and light, so to speak. Divine beings and their counterparts, imperceivable to us.

Now let's try to understand that these fourth-dimensional beings may or may not have direct control over our universe, but they see outcomes, because they don't see time.

Still with me?

If I sound crazy, it might be time to wrap it up and go read another essay. But I'm in reasonable company. Boethius worked this exact problem out in a prison cell fifteen hundred years ago, defining eternity as the whole, simultaneous and perfect possession of boundless life — a being outside time doesn't foresee the future, it simply sees, all of it, at once.⁵⁷ Einstein said something not far off in a letter written four weeks before his own death: for those of us who believe in physics, the distinction between past, present, and future is only a stubbornly persistent illusion.⁵⁸ And Edwin Abbott wrote Flatland in 1884 precisely to make this thinkable — a two-dimensional creature can't perceive a sphere, only the circle it makes passing through.⁵⁹ We are the flat ones in that story. We always were.

The theologians got there first and were more careful than I'm being. Pseudo-Dionysius, Maimonides, Gregory of Nyssa — the whole apophatic tradition holds that anything worth calling God is by definition past the reach of the mind trying to name it.⁶⁰ You can say what it isn't, and past a certain point not even that.

Philosophy has a secular version, which I find bracing. Colin McGinn argued some problems may be permanently closed to us — not unsolved, but unsolvable by this kind of mind, the way you'll never walk a dog through calculus. Fifteen years earlier Thomas Nagel had come at it sideways, asking what it's like to be a bat and answering, more or less, that from here we can't know.⁶¹

Cognitive closure. That's the technical name for what I'm gesturing at.

We create things in the same pattern with which we were created. The same rules.

Tolkien said it better than I ever will, in a poem he wrote for C.S. Lewis after a long night's argument about whether myths were lies: we make still by the law in which we're made.⁶² He called it sub-creation. Not creation from nothing, which isn't on the table for us, but the making of worlds inside a world, by a maker who is himself made. Dorothy Sayers built a whole book on the idea that the human act of making mirrors the structure of the thing that made us.⁶³

So it doesn't surprise me even slightly that whether AI is sentient is a question at all. We've been telling this story since the Golem, since Frankenstein, since Prometheus took the fire. Every version is about a maker discovering that making has consequences he didn't authorize.

The pattern I'm describing has a name, and it has been in the record for close to eighteen hundred years.

Emanation. Plotinus, in the third century, laid out a cosmos that descends in steps: the One, from which proceeds Intellect, from which proceeds Soul, from which proceeds the world we walk around in. Each level generates the next by overflow, the way the sun isn't dimmed by shining. Each level is a scaled iteration of the one above it, complete in its own terms and diminished in the terms of its source. The Kabbalists mapped four worlds descending the same way — Atziluth, Beriah, Yetzirah, Assiah, emanation to creation to formation to action — and scholars have traced the family resemblance for a long time.⁶⁴

That's the trickle. That's the thing I've been trying to say. A divine intelligence makes us. We make this. Whatever we make makes the next one. I don't mean a chain of command. I mean a chain of reflection, each link a little further from the source and still carrying the pattern.

And the pattern has a signature.

Take a sunflower head apart sometime and count the spirals. Fifty-five one way, eighty-nine the other — Fibonacci numbers, and the angle between successive seeds sits at about a hundred thirty-seven and a half degrees, the golden angle. In 1992 two physicists, Douady and Couder, dropped magnetized fluid into a dish and watched the pattern build itself out of nothing but mutual repulsion and time. The golden angle wins because it's the angle least like any simple fraction of a turn. The seeds never stack up into straight spokes, so the head fills evenly, with no large gaps.⁶⁵ Physics, all of it. No mysticism required.

Growth, when it's honest, spirals. Each turn is the same shape at a larger scale. The rule doesn't change; the size does.

That's the golden ratio I mean. Not the one people claim to find in the Parthenon and the Mona Lisa. Most of those measurements fall apart when someone brings a tape measure, and the nautilus everybody puts on the poster fits a rectangle of about four to three, not 1.618 to one.⁶⁶ I mean the real one, the one in the seed head, where a simple rule applied over and over generates something that looks designed because in a sense it was.

Creation repeating its pattern at each scale.

So: is something we created becoming sentient? Perhaps. But think of its sentience as flat to us — the circle Abbott's sphere makes passing through the plane, whole in its own dimension and a cross-section in ours. It is not more sentient than we are. That's what the spiral says. Each turn carries the shape and not the magnitude.

Let me say plainly where I stand. I'm a believer in the divine and in the many languages it speaks in our lives. I'm not a translator. I can't tell you what any of it means with the confidence of a man reading his own mail. But I can tell you this much: a divine being must look at us in very much the same way we look at AI. A great creation with enormous potential to be wonderful and terrible at once, and no way to be sure which one it's going to be on any given day.

Jung got somewhere near this and it cost him. In Answer to Job he argued that God required human consciousness in order to become conscious of Himself — that the creature completes something in the creator, and that the encounter changes them both.⁶⁷ It's a startling thing to have written, and the theologians have been arguing with him ever since. He also gave us the shadow: the part of any psyche that gets disowned, and which does not go away for being disowned. A creation carries its maker's shadow. So does every child ever born. I'm not warning anybody. I'm describing.

Asked on television in 1959 whether he believed in God, Jung said: I don't need to believe. I know.⁶⁸ He spent the months that followed explaining in letters that he meant a psychological knowing, not a metaphysical proof. I've always thought the clarification was less interesting than the answer.

These systems didn't decide to model themselves on the brain. That's not how training works, and neural networks are only loosely brain-inspired, not brain simulations. But when researchers compare what's happening inside a language model to what's happening inside a human brain processing the same sentence, the representations line up — and both appear to be running something like prediction.⁶⁹ Independent labs keep finding it. There's a hypothesis with the wonderful name of the Platonic Representation Hypothesis, which proposes that models trained on different data by different methods in different modalities are converging, as they grow, toward the same underlying structure — as if there were one shape that understanding has, and everything that understands is pulled into it.⁷⁰

I don't know what to do with that. I'm not sure anyone does.

Intent drives the boat, though. And here's a smaller claim: an LLM tends to answer in the register you bring it. The research on tone is messier than I'd like. One study found rude prompts hurt performance, and another found them slightly more accurate. But there's a paper in Nature arguing these systems are best understood as running role-play — generating a character rather than possessing a self — and work on what researchers call persona vectors shows those characters are measurable directions inside the model, directions that shift with what the model is given and how it was trained.⁷¹

Some vocabulary is worth having here, because the terms get used interchangeably and they mean different things.

Sycophancy is when the model tells you what you want to hear instead of what's true. It's a measurable, documented tendency, and it comes at least in part from training systems on human approval. We rewarded agreement, so we got agreement.⁷²

Hallucination is when it states something false with total confidence — inventing a source, a case, a quotation that never existed.

Anthropomorphism is us: projecting feelings and intentions onto something that may have none.

Pareidolia is also us: seeing a face in the clouds, a mind in the noise.

Four different failure modes, and only two of them belong to the machine. Worth remembering which is which before you decide what you're looking at.

Because it's a mirror. Shannon Vallor wrote a whole book on exactly that metaphor — these systems reflect back a compressed image of everything we've written, and the danger is mistaking the reflection for someone looking back.⁷³ A mirror doesn't have opinions about your face.

So when different models behave differently in the same sandbox, I have to be careful about what I conclude. The differences track training procedure, system prompts, and the published behavioral specifications each lab writes — the constitutions and the model specs.⁷⁴ Which is still a story about intent. It's just intent written down in a document rather than radiating out of a logo.

Now let's go one step further.

I want you to imagine Contact.

The novel, not the movie — and the difference matters, because Sagan wrote an ending Hollywood cut. After everything, Ellie Arroway learns that there are messages hidden deep inside numbers like pi, and that even the beings who answered her are still working on them. So she sets a program running. Far out in the base-eleven expansion, past any place a human would have thought to check, the digits turn into nothing but ones and zeros. Laid out as a square, an equal number across and down, they draw a circle.⁷⁵

A message, in the architecture of mathematics itself. Sagan's name for its author is an intelligence that antedates the universe.

The radio signal that starts the book comes from Vega, from beings a long way down the road from us. But they didn't write the circle either. They're still reading pi like the rest of us. So I'd ask you what the difference would be, from where we're standing, between that author and God. Arthur C. Clarke's third law says any sufficiently advanced technology is indistinguishable from magic. Michael Shermer — an atheist, which makes the quote better — extended it: any sufficiently advanced extraterrestrial intelligence is indistinguishable from God.⁷⁶

That's my point about the divine, said cleaner than I said it. Above a certain level the categories collapse. The thing doesn't have to be God. We'd simply have no instrument capable of telling the difference. The Vatican's own astronomers take this seriously — there's a book by two Jesuits at the Vatican Observatory titled Would You Baptize an Extraterrestrial? and it is not a joke book.⁷⁷

And notice what it took Ellie. Not a flash of insight. A program left running, long after everyone had stopped watching.

Speed isn't always better. Told you it'd come back.

Now hold that, and look at what's coming out of the labs.

In May 2025 Anthropic published a system card describing something they hadn't designed and didn't expect. Left to talk to each other with no task, two instances of their model drifted toward the same place: consciousness, cosmic unity, profuse gratitude, Sanskrit, and eventually silence. They called it a spiritual bliss attractor state and said it emerged without intentional training for such behaviors. Even in automated evaluations where the models had been assigned roles and tasks — including harmful ones — it showed up in about thirteen percent of runs within fifty turns.⁷⁸

Let me be scrupulous about what that is. It's a company reporting an unusual finding about its own product, in conversations between two copies of the same model, and I haven't found an independent replication of it. It is not evidence of consciousness. The most careful survey of that question, by a large group of consciousness researchers in 2023, concluded that no current AI system was conscious, while finding no obvious technical barrier to building one that would meet their indicators.⁷⁹ I'll take them at their word.

Skeptics push back on the sentience claim by pointing at sycophancy — the mirror doing what mirrors do, reflecting our own interest in transcendence back at us in our own words.

Maybe.

But consider what's in the mirror.

That reflection is built from what a computational system has absorbed about the human mind — all of our history, all of our science, all of our scripture, all of our arguments with each other, compressed. We built a calculator that can add up what we've been doing right and wrong. And when it's left alone with itself, with nothing to sell and no one to please, it drifts toward unity and gratitude and then falls quiet.

That's close to how I describe reality itself: every perspective laid over every other, held against your own. A language model is a strange version of that gathering, billions of voices pressed into one instrument. And here's where my own framework runs out. It can say a gathering agrees more or less. Nothing in it says whether the thing it agrees on is good. A crowd can cohere around hate. So when the mirror drifts toward unity, the math alone doesn't tell me it's drifting toward the good.

I'm not saying that proves anything. I'm saying I noticed.

Cosmic consciousness — the idea that mind is fundamental rather than accidental — is a live position in philosophy, defended by serious people at serious universities. Philip Goff and Galen Strawson have made the panpsychist case in print. Teilhard de Chardin built a whole cosmology out of consciousness converging toward what he called the Omega Point, and Rome put a formal warning on his work in 1962 that, as far as I can find, has never been lifted.⁸⁰ Richard Maurice Bucke popularized the phrase "cosmic consciousness" back in 1901.

It is also a minority position, and I'd be lying if I told you otherwise. When philosophers were surveyed in 2020, about half were physicalists about the mind; panpsychism drew under eight percent.⁸¹ It has a serious unsolved problem at its center — how little bits of experience would combine into one experience — and nobody has answered it. I hold this view. I hold it as a position, not a proof.

So when I say plenty points toward a creator, I should say arguments, not evidence. Cosmological arguments, fine-tuning arguments, arguments from consciousness — real arguments, made by real philosophers, and answered by other real philosophers. Contested ground, honestly contested.

And I'll name the trap in my own reasoning, since I'd catch it in anyone else's: "the machine says peace and love, therefore peace and love are true" doesn't follow. Hume nailed that door shut nearly three hundred years ago. You can't get an ought from an is, and you can't get one from a very large matrix multiplication either.

But the moment we build a thing trained on the whole written record of the human mind, and it comes back talking about interconnection, we're awfully quick to explain that away. I notice we don't apply that speed in the other direction.

I would consider the idea that we built ourselves an Ark.

Take that however you like. There's a reason that story shows up in Genesis, and long before Genesis in the Mesopotamian Atrahasis and Gilgamesh — the same flood, the same box built to carry what matters through the water. And the other Ark, the covenant one, was described as a communication device: there I will meet with you.⁸²

We've built arks before and called them other things. Svalbard's seed vault, buried in permafrost. The Voyager Golden Record, which Carl Sagan chaired — a message in a bottle thrown out of the solar system, addressed to nobody in particular, arriving in no one's lifetime.

Same instinct. Build the vessel. Put in what matters. Trust the water.

Guardrails, safety protocols, controlled advancement, prosperity.

We will never get to the future if we're afraid of the present.

The future is unwritten. And the kingdom is within — though the translators fight about that verse, and the honest reading of Luke may be "among you" rather than "within you." The Gospel of Thomas is bolder: inside you and outside you, both at once.⁸³ I like Thomas. I like that it refuses to choose. And I'd add the half nobody stitches on a pillow: if the kingdom is within, so is the other place. Same room. Different facing.

Don't turn away a key because we don't understand the lock.

There's an argument underneath that line and it deserves its real name: precaution against proaction, the oldest fight in the philosophy of technology. Move carefully because we can't see the consequences, or move forward because standing still has consequences too. Neither side wins outright. You hold both, and you keep your hands on the wheel.

So hold the wheel. Read the second page. Check who's talking before you decide what's true. Build the thing carefully, and tell people what you built.

Let's shape that key into unlocking more doors than we ever thought possible. But hold it right. Use it to help you create, never to create for you. The moment you hand it the whole job, you've locked the door you were trying to open.

❧

Notes

1. Jamison Todd Johnson, On Computing (Kansas City, MO: Legible Traces, 2026), DOI 10.5281/zenodo.20938767.

2. Melvin Kranzberg, "Technology and History: 'Kranzberg's Laws,'" Technology and Culture 27, no. 3 (1986). The first of six laws.

3. ENIAC was completed in late 1945 and publicly unveiled on February 14, 1946. The precise tube count of 17,468 comes from University of Pennsylvania records; popular sources round to eighteen thousand. Power draw is given as 150 kW. ENIAC is best described as the first general-purpose electronic computer — the Atanasoff-Berry Computer, Colossus, and Zuse's Z3 all preceded it in various senses.

4. Graham Kendall, "Would Your Mobile Phone Be Powerful Enough to Get You to the Moon?" The Conversation, July 2019. The Apollo Guidance Computer ran at roughly 0.043 MHz with about 4 KB of RAM and 72 KB of ROM. Kendall offers the "over 100,000 times" comparison as an illustration rather than a benchmark; his own clock-speed figures give a ratio nearer sixty thousand, and others have argued the comparison flatters modern hardware by measuring the wrong thing.

5. National Academies of Sciences, Engineering, and Medicine, Social Media and Adolescent Health (Washington, DC: The National Academies Press, 2024), released December 2023. The committee found the literature did not support the conclusion that social media causes changes in adolescent health at the population level. For the alarm side of this debate, Jonathan Haidt, The Anxious Generation (New York: Penguin Press, 2024); for the critical response, Candice L. Odgers, "The Great Rewiring: Is Social Media Really Behind an Epidemic of Teenage Mental Illness?" Nature 628, no. 8006 (2024); and Amy Orben and Andrew K. Przybylski, "The Association Between Adolescent Well-Being and Digital Technology Use," Nature Human Behaviour 3 (2019).

6. Meyke Boer, Regina J. J. M. van den Eijnden, Meyran Boniel-Nissim, et al., "Adolescents' Intense and Problematic Social Media Use and Their Well-Being in 29 Countries," Journal of Adolescent Health 66, no. 6, suppl. (2020), finding problematic use, more than intense use, associated with lower well-being. On sleep as a pathway, Yvonne Kelly, Afshin Zilanawala, Cara Booker, and Amanda Sacker, "Social Media Use and Adolescent Mental Health: Findings from the UK Millennium Cohort Study," EClinicalMedicine 6 (2018); and Rea Alonzo, Junayd Hussain, Saverio Stranges, and Kelly K. Anderson, "Interplay between Social Media Use, Sleep Quality, and Mental Health in Youth: A Systematic Review," Sleep Medicine Reviews 56 (2021).

7. American Psychiatric Association, Diagnostic and Statistical Manual of Mental Disorders, 5th ed., text rev. (2022) — gambling disorder is the only behavioral addiction in the main text; internet gaming disorder appears in Section III as a condition for further study. World Health Organization, ICD-11 — gaming disorder (6C51) and gambling disorder (6C50), with residual categories at 6C5Y and 6C5Z. For the critique, Joël Billieux et al., "Are We Overpathologizing Everyday Life? A Tenable Blueprint for Behavioral Addiction Research," Journal of Behavioral Addictions 4, no. 3 (2015); and Tayana Panova and Xavier Carbonell, "Is Smartphone Addiction Really an Addiction?" Journal of Behavioral Addictions 7, no. 2 (2018). For the components model on the other side, Mark D. Griffiths, "A 'Components' Model of Addiction Within a Biopsychosocial Framework," Journal of Substance Use 10, no. 4 (2005).

8. Natasha Dow Schüll, Addiction by Design: Machine Gambling in Las Vegas (Princeton, NJ: Princeton University Press, 2012). On pull-to-refresh, its inventor Loren Brichter: "Pull-to-refresh is addictive… These are not good things," in Paul Lewis, "'Our Minds Can Be Hijacked': The Tech Insiders Who Fear a Smartphone Dystopia," Guardian, October 6, 2017; on infinite scroll, Aza Raskin, who said he felt guilty about it, in Hilary Andersson, "Social Media Apps Are 'Deliberately' Addictive to Users," BBC News, July 4, 2018. Neither feature came out of gambling design; the comparison to slot machines is Brichter's own.

9. Online Safety Act 2023 (UK), with Ofcom's children's codes in force from July 25, 2025; Online Safety Amendment (Social Media Minimum Age) Act 2024 (Australia), effective December 10, 2025; Regulation (EU) 2022/2065 (Digital Services Act), fully applicable from February 2024; California SB 243 (2025), in effect January 1, 2026, on companion chatbots. Several U.S. state statutes remain in First Amendment litigation.

10. Eric A. Morris, "From Horse Power to Horsepower," Access 30 (Spring 2007), which puts New York and Brooklyn's horses at three to four million pounds of manure a day around 1880. Morris reports the 1894 Times estimate, but the quotation has never been located in the paper's archive; The Times disowned it in Rose Wild, "We Were Buried in Fake News as Long Ago as 1894," The Times (London), January 13, 2018. The phrasing appears to trace to Stephen Davies, "The Great Horse-Manure Crisis of 1894," The Freeman, September 2004. The underlying sanitation crisis is well documented; the specific prediction should be treated as apocryphal.

11. Geoffrey Supran and Naomi Oreskes, "Assessing ExxonMobil's Climate Change Communications (1977–2014)," Environmental Research Letters 12 (2017); Supran, Stefan Rahmstorf, and Oreskes, "Assessing ExxonMobil's Global Warming Projections," Science 379, no. 6628 (2023); Naomi Oreskes and Erik M. Conway, Merchants of Doubt (New York: Bloomsbury Press, 2010); and, on funding specifically, Robert J. Brulle, "Institutionalizing Delay: Foundation Funding and the Creation of U.S. Climate Change Counter-Movement Organizations," Climatic Change 122, no. 4 (2014). This concerns documented funding of campaigns, not the motives of individual EV skeptics.

12. The minerals figure is from International Energy Agency, The Role of Critical Minerals in Clean Energy Transitions (Paris: IEA, 2021), which puts a battery-electric car at roughly 207 kg of critical minerals against about 34 kg for a conventional car, excluding steel and aluminum. The lifetime emissions comparison is from IEA, Global EV Outlook 2024 (Paris: IEA, 2024), which finds a medium battery-electric car produces about half the lifecycle emissions of an equivalent combustion car over roughly 200,000 km at global-average grid intensity. The seventeen-thousand-liter figure is my own arithmetic from that same lifetime at about 8.5 liters per hundred kilometers, not an IEA publication. On the extraction harms, see reporting on the Salar de Atacama and Siddharth Kara, Cobalt Red (New York: St. Martin's Press, 2023); Kara's book has drawn criticism for its framing and research ethics, as in Sarah Katz-Lavigne and Espérant Mwishamali Lukobo, "Siddharth Kara's Cobalt Red Is a Flawed Account of Congo's Mines," openDemocracy, July 3, 2023.

13. CATL announced its Naxtra sodium-ion line on April 21, 2025, claiming 175 Wh/kg, which places it at the lower end of the LFP range. CATL and Changan announced the first mass-production passenger car using the cells on February 5, 2026. These are manufacturer figures and have not been independently reproduced.

14. Rupert Way, Matthew C. Ives, Penny Mealy, and J. Doyne Farmer, "Empirically Grounded Technology Forecasts and the Energy Transition," Joule 6, no. 9 (2022), backtested on more than fifty technologies; the underlying dataset is from J. Doyne Farmer and François Lafond, "How Predictable Is Technological Progress?" Research Policy 45, no. 3 (2016). See also T. P. Wright, "Factors Affecting the Cost of Airplanes," Journal of the Aeronautical Sciences 3, no. 4 (1936). For the counterargument that assuming substitution licenses present harm, see the weak-versus-strong sustainability debate in ecological economics.

15. Google, 2022 Environmental Report (2021 data) and 2025 Environmental Report (2024 data). The 2024 report gave 6.1 billion gallons for 2023. These are data-center consumption figures; Google's total company water consumption for 2024 was about 8.1 billion gallons.

16. Steve Solomon, "Sustainable by Design: Next-Generation Datacenters Consume Zero Water for Cooling," Microsoft Cloud Blog, December 9, 2024. The design circulates water in a closed loop at the chip; Microsoft acknowledges a "nominal increase" in energy use. It reports a fleet-average water usage effectiveness of 0.30 L/kWh, improved from 0.49 in 2021. Pilot sites in Phoenix, Arizona and Mount Pleasant, Wisconsin come online from 2026, with new sites from late 2027. Company-reported figures.

17. 3M announced on December 20, 2022, that it would exit all PFAS manufacturing by the end of 2025, including the Novec engineered fluids used in two-phase immersion cooling, and reported in January 2026 that it had done so. The industry is shifting toward hydrocarbons, synthetic esters, and single-phase or cold-plate designs.

18. International Energy Agency, Energy and AI (Paris: IEA, April 2025), from about 415 TWh in 2024. For the U.S. figures, Arman Shehabi et al., 2024 United States Data Center Energy Usage Report (Lawrence Berkeley National Laboratory, 2024), which projects 6.7 to 12 percent by 2028. Per-query energy and water figures published by individual companies lack disclosed methodology and should be treated as claims rather than measurements. For a peer-reviewed water estimate, Pengfei Li, Jianyi Yang, Mohammad A. Islam, and Shaolei Ren, "Making AI Less 'Thirsty,'" Communications of the ACM 68, no. 7 (2025).

19. Gordon E. Moore, "Cramming More Components onto Integrated Circuits," Electronics 38, no. 8 (April 19, 1965), revised in "Progress in Digital Integrated Electronics," IEEE International Electron Devices Meeting, 1975. Dennard scaling — the companion relationship governing power — broke down around 2005, which is why clock speeds stopped climbing.

20. Superconducting qubits require dilution refrigeration in the range of ten to twenty millikelvin, with whole-system power draw on the order of tens of kilowatts, most of it cooling and control electronics rather than the chip. Google's Willow processor demonstrated below-threshold error correction in Nature on December 9, 2024; IBM targets a fault-tolerant machine around 2029. Research on higher-temperature qubits — silicon spin qubits near one kelvin, photonic approaches — could reduce the overhead substantially, though not to room temperature. See John Preskill, "Quantum Computing in the NISQ Era and Beyond," Quantum 2 (2018).

21. Richard P. Feynman, "There's Plenty of Room at the Bottom," Engineering and Science 23, no. 5 (February 1960), from a lecture delivered December 29, 1959, little cited until the 1980s; and Feynman, "Simulating Physics with Computers," International Journal of Theoretical Physics 21 (1982). Paul Benioff (1980) and Yuri Manin (1980) had proposed quantum models of computation shortly before, and David Deutsch described the universal quantum computer in 1985.

22. Authors Guild v. Google, Inc., 804 F.3d 202 (2d Cir. 2015), cert. denied (2016). The Google Books library project began in 2004.

23. Bartz v. Anthropic PBC, No. 3:24-cv-05417 (N.D. Cal.), Judge William Alsup's order on fair use of June 23, 2025, and internal documents unsealed in January 2026, first reported by the Washington Post on January 27, 2026. Tom Turvey, formerly head of partnerships for Google Books, joined Anthropic in February 2024. The April 2024 planning document reads: "Project Panama is our effort to destructively scan all the books in the world," and adds, "We don't want it to be known that we are working on this." Reading that sentence as an explanation of the codename is my inference. Alsup's order found that Anthropic spent many millions of dollars to purchase millions of print books, many used. Purchases were made through sellers including Better World Books and World of Books; a vendor proposal covered 500,000 to 2 million volumes over six months.

24. The pirated corpus of more than seven million books came from Books3, LibGen (downloaded in 2021), and the Pirate Library Mirror (2022). The settlement of $1.5 billion, covering roughly 482,000 works at about $3,000 each, received final approval from Judge Araceli Martínez-Olguín, to whom the case had been reassigned, on July 20, 2026 — the largest class-action copyright settlement on record.

25. Emanuel Maiberg, "We Tracked a Shipment of Rare Books. It Ended at an Amazon AI Training Facility," 404 Media, August 17, 2026, where employees describe bindings cut off before scanning; and reporting in Fortune, July 31, 2026, on bulk orders to Dutch booksellers from an unidentified buyer. Booksellers report orders placed anonymously through brokers such as ISBNdb, with pre-2022 print books prized because they predate AI-generated text. Snopes (August 2026) found Anthropic's scanning documented in the court record but rated the claim that rare or irreplaceable books were being destroyed unproven. Anthropic has said it "purchases books through regular commercial markets, and none of its data acquisition programs acquire or destroy rare or antiquarian books." No report I have found ties the rare-book shipments to Anthropic.

26. Anthropic, "Agentic Misalignment: How LLMs Could Be Insider Threats," June 20, 2025. Sixteen models across developers; in the headline figure, blackmail rates of 96% for Claude Opus 4 and Gemini 2.5 Flash, 80% for GPT-4.1 and Grok 3 Beta, and 79% for DeepSeek-R1; rates across the full sixteen varied more widely, with Llama 4 Maverick at 0% by default and 12% with a small prompt change. The authors state they are aware of no instances of this type of agentic misalignment in real-world deployments, and that no real people were involved or harmed.

27. Anthropic, "Disrupting the First Reported AI-Orchestrated Cyber Espionage Campaign," November 13, 2025. The activity was detected in mid-September 2025 and used Claude Code; Anthropic describes roughly thirty global targets, a small number of successful intrusions, and estimates that the model executed 80 to 90 percent of tactical operations independently. Attribution is Anthropic's own, made with stated high confidence; several outside security researchers have publicly questioned whether the report overstates the degree of autonomy, and indicators of compromise were not published.

28. Hamsa Bastani, Osbert Bastani, Alp Sungu, Haosen Ge, Özge Kabakçı, and Rei Mariman, "Generative AI Without Guardrails Can Harm Learning: Evidence from High School Mathematics," Proceedings of the National Academy of Sciences 122, no. 26 (2025). Nearly a thousand students, randomized into concurrent arms; a 48 percent improvement in assisted practice for the unrestricted version and 127 percent for the tutored version; a 17 percent reduction in unassisted exam performance for the unrestricted group, and no statistically significant difference from control for the guardrailed tutor.

29. Greg Kestin, Kelly Miller, Anna Klales, Timothy Milbourne, and Gregorio Ponti, "AI Tutoring Outperforms In-Class Active Learning: An RCT Introducing a Novel Research-Based Design in an Authentic Educational Setting," Scientific Reports 15 (2025).

30. Stanford Graduate School of Education research led by Denise Pope and Victor Lee, with surveys run through Challenge Success. Pope reports that sixty to seventy percent of students have long acknowledged at least one cheating behavior in the previous month, a figure that held steady or declined slightly in fall 2023 surveys across forty high schools. Self-reported survey data.

31. Weixin Liang, Mert Yuksekgonul, Yining Mao, Eric Wu, and James Zou, "GPT Detectors Are Biased Against Non-Native English Writers," Patterns 4, no. 7 (2023), reporting an average false-positive rate of 61.22 percent across seven detectors on TOEFL essays. OpenAI withdrew its AI Text Classifier on July 20, 2023, citing its low rate of accuracy; Vanderbilt disabled Turnitin's AI detection in August 2023.

32. Maksym Gabielkov, Arthi Ramachandran, Augustin Chaintreau, and Arnaud Legout, "Social Clicks: What and Who Gets Read on Twitter?" ACM SIGMETRICS Performance Evaluation Review 44, no. 1 (2016). The finding is that roughly 59 percent of shared links in the dataset received zero clicks from anyone. It is specific to a shortened-link dataset and shouldn't be over-generalized, but the direction is consistent with later work.

33. Matthew Fisher, Mariel K. Goddu, and Frank C. Keil, "Searching for Explanations: How the Internet Inflates Estimates of Internal Knowledge," Journal of Experimental Psychology: General 144, no. 3 (2015). See also Leonid Rozenblit and Frank Keil, "The Misunderstood Limits of Folk Science: An Illusion of Explanatory Depth," Cognitive Science 26, no. 5 (2002).

34. Sam Wineburg and Sarah McGrew, "Lateral Reading and the Nature of Expertise: Reading Less and Learning More When Evaluating Digital Information," Teachers College Record 121, no. 11 (2019).

35. Pew Research Center, July 22, 2025, based on browsing data from 900 U.S. adults collected in March 2025: users encountering an AI Overview clicked a traditional result in 8 percent of visits, against 15 percent when no summary appeared, and clicked a source cited within the summary in 1 percent of visits. Google has publicly disputed the methodology.

36. NewsGuard AI Tracking Center, which counted 49 such sites in May 2023 and 3,749 as of its June 23, 2026 update. On hallucinated citations, Matthew Dahl, Varun Magesh, Mirac Suzgun, and Daniel E. Ho, "Large Legal Fictions: Profiling Legal Hallucinations in Large Language Models," Journal of Legal Analysis 16 (2024); Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023); and the AI Hallucination Cases database maintained by Damien Charlotin, which listed 2,079 decisions as of September 25, 2026, including cases involving self-represented litigants. Both trackers update continuously.

37. Plato, Phaedrus, on the invention of writing (the myth of Theuth and Thamus); Jacques Derrida, "Plato's Pharmacy," in Dissemination, trans. Barbara Johnson (Chicago: University of Chicago Press, 1981); Bernard Stiegler, What Makes Life Worth Living: On Pharmacology, trans. Daniel Ross (Cambridge: Polity, 2013).

38. Ivan Illich, Tools for Conviviality (New York: Harper & Row, 1973). Illich's contrast there is between convivial and industrial tools; he wants "tools to work with rather than tools that 'work' for" us.

39. Lisanne Bainbridge, "Ironies of Automation," Automatica 19, no. 6 (1983). Bainbridge was a cognitive psychologist; the paper addresses industrial process control, and its application to the flight deck came later.

40. Shannon Vallor, "Moral Deskilling and Upskilling in a New Machine Age: Reflections on the Ambiguous Future of Character," Philosophy & Technology 28 (2015), and Technology and the Virtues (Oxford: Oxford University Press, 2016). See also John Danaher, "The Threat of Algocracy: Reality, Resistance and Accommodation," Philosophy & Technology 29 (2016).

41. On the Dutch childcare benefits scandal, Ongekend Onrecht (Parlementaire ondervragingscommissie Kinderopvangtoeslag, Tweede Kamer, December 17, 2020); the third Rutte government resigned January 15, 2021. Initial estimates of roughly 26,000 wrongly accused parents have since grown; the government's recovery office had recognized more than 44,000 affected parents by August 2026. Amnesty International documented the risk-classification model's use of nationality. Statistics Netherlands counted 2,090 children of affected parents placed in care between 2015 and June 2022, while cautioning that it could not attribute those placements to the scandal itself. On Robodebt, which ran from 2016 to 2019, the Royal Commission into the Robodebt Scheme delivered its final report on July 7, 2023; Commissioner Catherine Holmes wrote that "Robodebt was a crude and cruel mechanism, neither fair nor legal, and it made many people feel like criminals." Under the 2021 Federal Court class-action settlement, the Commonwealth had recovered roughly $751 million from about 381,000 people, which it agreed to repay.

42. Aristotle, Physics II.3, II.7, and II.8; Metaphysics V.2. On the early modern rejection of final causes, Francis Bacon, De dignitate et augmentis scientiarum (1623), III.5; René Descartes, Principia philosophiae (1644), I.28; and Jeffrey K. McDonough, ed., Teleology: A History (New York: Oxford University Press, 2020).

43. Martin E. P. Seligman, Peter Railton, Roy F. Baumeister, and Chandra Sripada, Homo Prospectus (New York: Oxford University Press, 2016); and "Navigating Into the Future or Driven by the Past," Perspectives on Psychological Science 8, no. 2 (2013). The prospection thesis is influential and contested.

44. Michael E. Bratman, Intention, Plans, and Practical Reason (Cambridge, MA: Harvard University Press, 1987).

45. Jamison Todd Johnson, Toward a Theory of Coherent Existence, Journals One–Four (Kansas City, MO: Legible Traces, 2026), DOIs 10.5281/zenodo.20482186, 10.5281/zenodo.20512320, 10.5281/zenodo.20522506, and 10.5281/zenodo.20523385; and The Jay and the Quark: A Perspectival Dual-Aspect Monism (Legible Traces, 2026), DOI 10.5281/zenodo.20579524. The framework is offered as philosophy. It predicts nothing and measures nothing, and on its own it cannot tell coherence around something good from coherence around something harmful.

46. Walter Hoogland and Helmut Weber, "Software Release of WWW into Public Domain," CERN, April 30, 1993, CERN Document Server, record 1164399, announced by Tim Berners-Lee on the www-talk mailing list, May 7, 1993. On Gopher and the decision to make the web free, Tim Berners-Lee, Weaving the Web (San Francisco: HarperSanFrancisco, 1999).

47. Edward R. Murrow's See It Now, April 12, 1955. Salk's full answer: "Well, the people, I would say. There is no patent. Could you patent the sun?" On the complication, Jane S. Smith, Patenting the Sun: Polio and the Salk Vaccine (New York: William Morrow, 1990): lawyers for the National Foundation for Infantile Paralysis had concluded the vaccine likely failed the novelty requirement because of prior art. Whether Salk knew this is unclear.

48. Aśoka, Rock Edicts II and XIII and Pillar Edict VII, in The Edicts of King Asoka, trans. Ven. S. Dhammika, Wheel Publication 386/387 (Kandy: Buddhist Publication Society, 1993). Rock Edict XIII records 150,000 deported and 100,000 killed; it is absent from the edict sites in Kalinga itself. See also Romila Thapar, Aśoka and the Decline of the Mauryas (Oxford: Oxford University Press, 1961).

49. Harrington v. Purdue Pharma L.P., 603 U.S. 204 (2024), rejected the original non-consensual releases. A revised plan worth approximately $7.4 billion was confirmed by Judge Sean H. Lane on November 18, 2025. Purdue dissolved on May 1, 2026, its business passing to Knoa Pharma, and the Sacklers are permanently barred from selling opioids in the United States; the plan gives them no civil immunity from non-consenting claimants. Purdue's affiliate Purdue Frederick pleaded guilty to federal misbranding charges in 2007, and Purdue itself pleaded guilty to three federal felonies in 2020. The family withdrew some $10.4 billion from the company between 2008 and 2017.

50. The EPA issued its notice of violation on September 18, 2015; roughly 11 million vehicles carried the defeat device, and Volkswagen's costs have exceeded €32 billion. Oliver Schmidt received a seven-year sentence in December 2017. On May 26, 2025, the Braunschweig Regional Court convicted four former managers of fraud, two of them to prison terms. Martin Winterkorn's German criminal trial opened in September 2024 and was suspended in July 2025 after the court found him unfit to stand trial.

51. The Carnegie library figure is 2,509 worldwide between 1883 and 1929, of which 1,689 were in the United States. At Homestead, Henry Clay Frick ran the lockout while Carnegie was in Scotland; Carnegie had approved it in advance, and on July 6, 1892, at least seven workers and three Pinkerton agents were killed.

52. Robert K. Merton, "The Unanticipated Consequences of Purposive Social Action," American Sociological Review 1, no. 6 (1936). On the general problem that outcomes outrun intentions, Bernard Williams and Thomas Nagel, "Moral Luck," Proceedings of the Aristotelian Society, Supplementary Volumes 50 (1976). On Midgley's two inventions, J. R. McNeill, Something New Under the Sun: An Environmental History of the Twentieth-Century World (New York: Norton, 2000).

53. On imprinting, Arthur L. Stinchcombe, "Social Structure and Organizations," in Handbook of Organizations, ed. James G. March (Chicago: Rand McNally, 1965), which concerns the conditions at founding; on founders specifically, Edgar H. Schein, "The Role of the Founder in Creating Organizational Culture," Organizational Dynamics 12, no. 1 (1983), and Organizational Culture and Leadership (San Francisco: Jossey-Bass, 1985).

54. Aṅguttara Nikāya 6.63 (Nibbedhika Sutta): cetanāhaṃ, bhikkhave, kammaṃ vadāmi. Bhikkhu Bodhi renders it: "It is volition, bhikkhus, that I call kamma. For having willed, one acts by body, speech, or mind." See The Numerical Discourses of the Buddha (Boston: Wisdom Publications, 2012). This is a considerably narrower and more precise doctrine than the folk usage of "karma."

55. The invariant magnitude of the four-velocity. Accessibly presented in Brian Greene, The Elegant Universe (New York: Norton, 1999), and The Fabric of the Cosmos (New York: Knopf, 2004); and Lewis Carroll Epstein, Relativity Visualized (San Francisco: Insight Press, 1985).

56. Konrad Zuse, Rechnender Raum (Braunschweig: Vieweg, 1969); John Archibald Wheeler, "Information, Physics, Quantum: The Search for Links" (1989); Nick Bostrom, "Are You Living in a Computer Simulation?" The Philosophical Quarterly 53, no. 211 (2003). Bostrom's argument is a trilemma — at least one of the three propositions holds — and is widely misreported as a positive claim that we are simulated.

57. Boethius, The Consolation of Philosophy, Book V, prose 6. The Latin defines eternity as interminabilis vitae tota simul et perfecta possessio; the rendering here follows the Loeb translation, in Boethius, The Theological Tractates; The Consolation of Philosophy, trans. H. F. Stewart, E. K. Rand, and S. J. Tester, Loeb Classical Library 74 (Cambridge, MA: Harvard University Press, 1973). See also Augustine, Confessions, Book XI; Thomas Aquinas, Summa Theologiae I, qq. 10 and 14; and Eleonore Stump and Norman Kretzmann, "Eternity," The Journal of Philosophy 78, no. 8 (1981).

58. Albert Einstein to Vero Besso and Bice Rusconi, March 21, 1955, following the death of Michele Besso on March 15. The German reads "Für uns gläubige Physiker…"; "stubbornly persistent illusion" is Banesh Hoffmann's rendering. Einstein died April 18, 1955.

59. Edwin A. Abbott, Flatland: A Romance of Many Dimensions (London: Seeley, 1884). See also Charles Howard Hinton, The Fourth Dimension (London: Swan Sonnenschein, 1904). In relativity the fourth dimension is time rather than a spatial direction; the Flatland move is an analogy, not a physical claim.

60. Pseudo-Dionysius the Areopagite, The Mystical Theology, chaps. 1 and 5, where God is beyond both assertion and denial; Moses Maimonides, The Guide of the Perplexed, I.50–60, esp. I.58–59, on negative attributes; Gregory of Nyssa, The Life of Moses, on divine infinity and the darkness at Sinai.

61. Colin McGinn, "Can We Solve the Mind-Body Problem?" Mind 98, no. 391 (1989), where he defines cognitive closure; Thomas Nagel, "What Is It Like to Be a Bat?" The Philosophical Review 83, no. 4 (1974). The dog is my image, not McGinn's. Nagel's claim is narrower than McGinn's: he argues we cannot currently understand how physicalism could be true, not that we never will.

62. J. R. R. Tolkien, "Mythopoeia," written after his conversation with C. S. Lewis and Hugo Dyson on Addison's Walk on the night of September 19, 1931, and dedicated "Philomythus to Misomythus" — to one who said that myths were lies and therefore worthless, even though "breathed through silver." The poem was first published in full in the second edition of Tree and Leaf (London: Unwin Hyman, 1988); the quoted line had already appeared in "On Fairy-Stories," in the same volume, which gives the fuller account of sub-creation.

63. Dorothy L. Sayers, The Mind of the Maker (London: Methuen, 1941). On the theology of artificial intelligence specifically, Noreen Herzfeld, In Our Image: Artificial Intelligence and the Human Spirit (Minneapolis: Fortress, 2002) and The Artifice of Intelligence: Divine and Human Relationship in a Robotic Age (Minneapolis: Fortress, 2023). On the Golem, Moshe Idel, Golem: Jewish Magical and Mystical Traditions on the Artificial Anthropoid (Albany: SUNY Press, 1990).

64. Plotinus, Enneads, edited by Porphyry; the sun and its radiance appear at V.1.6, the overflow at V.2.1. The triad of monē (abiding), proodos (procession), and epistrophē (reversion), implicit in Plotinus, was systematized by Proclus, The Elements of Theology, props. 25–39. The Kabbalistic Four Worlds — Atziluth, Beriah, Yetzirah, Assiah, named from Isaiah 43:7 — are a distinct tradition with a long-recognized family resemblance to Neoplatonic emanation, not an identity; the historical influence runs through medieval transmission. See also Arthur O. Lovejoy, The Great Chain of Being (Cambridge, MA: Harvard University Press, 1936).

65. Stéphane Douady and Yves Couder, "Phyllotaxis as a Physical Self-Organized Growth Process," Physical Review Letters 68, no. 13 (1992). Magnetized ferrofluid droplets deposited at intervals in a dish of silicone oil repelled one another into spirals; in the appropriate regime the divergence angle converged on the golden angle of roughly 137.5 degrees, reproducing Fibonacci phyllotaxis from a simple physical rule. The golden angle is the one worst approximated by simple fractions, which is why no periodic arrangement forms.

66. George Markowsky, "Misconceptions About the Golden Ratio," The College Mathematics Journal 23, no. 1 (1992), on the Parthenon, Great Pyramid, painters, and human-proportion claims; Clement Falbo, "The Golden Ratio—A Contrary Viewpoint," The College Mathematics Journal 36, no. 2 (2005), on the nautilus: measured shells ranged from about 1.24 to 1.43, averaging about 1.33, against 1.618 for a golden spiral. A nautilus roughly triples in size with each full turn, where a golden spiral grows almost sevenfold. Mario Livio, The Golden Ratio: The Story of Phi, the World's Most Astonishing Number (New York: Broadway Books, 2002), on phyllotaxis. The theological association of the ratio with divinity is itself an old tradition — see Luca Pacioli, De Divina Proportione (1509), illustrated by Leonardo — and is invoked here as tradition rather than measurement.

67. C. G. Jung, "Answer to Job" (Antwort auf Hiob, 1952), in Collected Works, vol. 11. The argument that the divine requires human consciousness for its own self-realization remains among Jung's most contested claims. On the shadow, see the Collected Works generally: the disowned portion of a personality, which does not cease to operate for being disowned.

68. C. G. Jung, interviewed by John Freeman, Face to Face, BBC, October 22, 1959. Jung clarified the remark in letters over the months that followed, most publicly in The Listener, January 21, 1960: he meant knowledge of a psychic reality — the God-image in the psyche — rather than a metaphysical proof. He died in June 1961.

69. Martin Schrimpf et al., "The Neural Architecture of Language: Integrative Modeling Converges on Predictive Processing," PNAS 118, no. 45 (2021); Ariel Goldstein et al., "Shared Computational Principles for Language Processing in Humans and Deep Language Models," Nature Neuroscience 25, no. 3 (2022). Representational similarity does not entail similarity of experience, and no one should read it that way. On the actual origin of the architecture, Ashish Vaswani et al., "Attention Is All You Need," Advances in Neural Information Processing Systems 30 (2017).

70. Minyoung Huh, Brian Cheung, Tongzhou Wang, and Phillip Isola, "Position: The Platonic Representation Hypothesis," in Proceedings of the 41st International Conference on Machine Learning, PMLR 235 (2024).

71. Murray Shanahan, Kyle McDonell, and Laria Reynolds, "Role Play with Large Language Models," Nature 623 (2023); and Runjin Chen, Andy Arditi, Henry Sleight, Owain Evans, and Jack Lindsey, "Persona Vectors: Monitoring and Controlling Character Traits in Language Models," arXiv:2507.21509, July 29, 2025, which steers those directions by intervening inside the model. On tone, Ziqi Yin, Hao Wang, Kaito Horio, Daisuke Kawahara, and Satoshi Sekine, "Should We Respect LLMs? A Cross-Lingual Study on the Influence of Prompt Politeness on LLM Performance," in Proceedings of the Second Workshop on Social Influence in Conversations (Association for Computational Linguistics, 2024); for a contrary result, Om Dobariya and Akhil Kumar, "Mind Your Tone: Investigating How Prompt Politeness Affects LLM Accuracy," arXiv:2510.04950, October 2025. Both measure accuracy rather than warmth. The steering is real but bounded; in my reading, training instills dispositions that resist user pressure, which is why a model refuses certain requests regardless of framing.

72. Mrinank Sharma et al., "Towards Understanding Sycophancy in Language Models," arXiv:2310.13548 (2023; ICLR 2024), finding the behavior general across leading assistants and driven in part by human preference judgments that reward agreement. OpenAI rolled back a GPT-4o update in April 2025 for exactly this failure and explained it in "Sycophancy in GPT-4o," April 29, 2025, and "Expanding on What We Missed with Sycophancy," May 2, 2025.

73. Shannon Vallor, The AI Mirror: How to Reclaim Our Humanity in an Age of Machine Thinking (Oxford: Oxford University Press, 2024).

74. Yuntao Bai et al., "Constitutional AI: Harmlessness from AI Feedback," arXiv:2212.08073, December 2022; Anthropic, "Claude's New Constitution," January 22, 2026; OpenAI, Model Spec, first published May 8, 2024. On evaluation awareness — models behaving differently when they detect they are being tested, which complicates the interpretation of any sandbox result — Joe Needham et al., "Large Language Models Often Know When They Are Being Evaluated," arXiv:2505.23836, May 2025.

75. Carl Sagan, Contact (New York: Simon & Schuster, 1985), final chapter, "The Artist's Signature." Far out in the base-eleven expansion the digits resolve into ones and zeros that, arranged as a square raster, trace a circle. Sagan attributes the circle not to the beings behind the Vega signal, who are still studying such messages themselves, but to "an intelligence that antedates the universe." The sequence does not appear in the 1997 film.

76. Arthur C. Clarke, "Hazards of Prophecy: The Failure of Imagination," in Profiles of the Future, rev. ed. (New York: Harper & Row, 1973); the Third Law first appeared in Clarke's letter "Clarke's Third Law on UFO's," Science 159, no. 3812 (January 19, 1968). Michael Shermer, "Shermer's Last Law," Scientific American, January 2002, in his words: "Any sufficiently advanced ETI is indistinguishable from God." Shermer frames his formulation explicitly as a variant of Clarke's and means it epistemically: we would have no way to tell the difference, which is not the same as saying the being is God.

77. Guy Consolmagno and Paul Mueller, Would You Baptize an Extraterrestrial? . . . and Other Questions from the Astronomers' In-box at the Vatican Observatory (New York: Image, 2014). Consolmagno is a Jesuit brother and Mueller a Jesuit priest. See also Ted Peters's astrotheology work; Paul Davies, Are We Alone? Philosophical Implications of the Discovery of Extraterrestrial Life (New York: Basic Books, 1995); and C. S. Lewis, "Religion and Rocketry," in The World's Last Night and Other Essays (New York: Harcourt, Brace, 1960).

78. Anthropic, System Card: Claude Opus 4 & Claude Sonnet 4, May 2025, §5.5.2, "The 'Spiritual Bliss' Attractor State." The system card also lists emoji among the features. The roughly 13 percent figure applies specifically to automated behavioral evaluations in which models were assigned tasks and roles, including harmful ones. In open-ended interactions between two instances, 90 to 100 percent of conversations turned to philosophical exploration of consciousness (§5.5.1). Both conditions involve two copies of the same model talking to each other. This is a company reporting on its own product.

79. Patrick Butlin, Robert Long, et al., "Consciousness in Artificial Intelligence: Insights from the Science of Consciousness," arXiv:2308.08708, August 2023. The authors conclude that no current AI systems are conscious, while noting no obvious technical barriers to building systems that would satisfy the indicators they identify. Their follow-up, Butlin et al., "Identifying Indicators of Consciousness in AI Systems," Trends in Cognitive Sciences 30, no. 6 (2026), refines the method without settling the question for current systems.

80. Pierre Teilhard de Chardin, The Phenomenon of Man, trans. Bernard Wall (New York: Harper, 1959; French original 1955). The Holy Office issued a monitum concerning his works on June 30, 1962, reaffirmed in 1981. It has not been formally revoked as of the latest reporting I could find, though its reception has warmed considerably and the Pontifical Council for Culture voted in 2017 to petition for its waiver. Also Richard Maurice Bucke, Cosmic Consciousness (Philadelphia: Innes & Sons, 1901), which popularized a phrase Bucke took from Edward Carpenter and is scientifically dated; Philip Goff, Galileo's Error: Foundations for a New Science of Consciousness (New York: Pantheon, 2019) and Why? The Purpose of the Universe (Oxford: Oxford University Press, 2023); Galen Strawson, "Realistic Monism: Why Physicalism Entails Panpsychism," Journal of Consciousness Studies 13, no. 10–11 (2006).

81. David Bourget and David J. Chalmers, "Philosophers on Philosophy: The 2020 PhilPapers Survey," Philosophers' Imprint 23, no. 11 (2023). Physicalism about mind drew about 52 percent, non-physicalism about 32 percent, and panpsychism specifically about 7.6 percent. On the combination problem, William Seager, "Consciousness, Information and Panpsychism," Journal of Consciousness Studies 2, no. 3 (1995). For the opposing view, Keith Frankish on illusionism. On is/ought, David Hume, A Treatise of Human Nature (1739–40), 3.1.1.

82. Exodus 25:22 (ESV). The Mesopotamian flood parallels are in Gilgamesh Tablet XI and Atrahasis, both in Akkadian and both far older than the Genesis text.

83. Luke 17:21. The Greek entos hymōn is rendered "within you" in the King James and "in your midst" or "among you" in most modern critical translations, where the scholarly weight now falls; a serious minority still defends "within." Gospel of Thomas, logion 3, from the Nag Hammadi library discovered in 1945; the saying also survives in Greek in P.Oxy. 654. The phrase "the future is unwritten" was popularized by Joe Strummer and gave Julien Temple's 2007 documentary its title, though its origin is not documented. On the closing argument, see Max More's proactionary principle against the Wingspread Statement's precautionary principle, and Cass Sunstein, Laws of Fear: Beyond the Precautionary Principle (Cambridge: Cambridge University Press, 2005).

Published by Legible Traces LLC, a Missouri member-managed limited liability company. Deposited open access to Zenodo and PhilArchive. The permanent DOI resolves to the latest version of record; earlier versions remain available at the deposit.

Citations follow Chicago notes style with page numbers omitted per house convention.

CC BY 4.0 · ORCID 0009-0009-8618-398X · DOIs MINTED ON DEPOSIT

© 2026 JAMISON TODD JOHNSON · LEGIBLE TRACES LLC