Deterministic AI Is a Myth — and Physics Proved It First

An episode of Dan's AI Intel

Reality is probabilistic all the way down — so the mature move, in physics and AI alike, is to engineer reliability on top of the randomness, not to demand a clock.

Published · By Dan Walter

Executive summary

There is a particular flinch a good engineer feels the first time a language model does real work: it writes the function, it passes the test, and then you run it again and the output is different. Not wrong, just different. Something in the trained-on-repeatability part of the brain recoils — this is a machine, and machines are supposed to be clocks. That recoil feels like rigor. It is worth taking seriously, because half of it is: reproducibility, testability, and auditability are real engineering virtues, not neuroses. But the other half is a philosophical hangover, and it is the same hangover that afflicted the single greatest scientific mind of the twentieth century. Albert Einstein spent the back half of his life insisting that "God does not play dice." In 2022, three physicists won the Nobel Prize for the experiments that proved, about as conclusively as physics ever proves anything, that He does.

That is the uncomfortable, clarifying fact at the center of this: reality is probabilistic at the bottom, not because we are ignorant of some hidden clockwork, but because the clockwork is not there. The deterministic universe was the comforting fiction; irreducible chance is the ground truth. And the deep lesson is not despair — it is that physics did not collapse when it gave up determinism. It got better. Every reliable thing built on top of quantum mechanics — every transistor, every bridge, every fifteen-day weather forecast — is reliability engineered on top of randomness, not the absence of randomness. Thermodynamics is rock-solid precisely because it is statistics over molecular chaos.

The argument here is that the modern discomfort with probabilistic AI is a category error we have made before and already know how to resolve. The AI breakthrough was itself a probabilistic bet that beat forty years of hand-built determinism. Asking that engine to behave like a clock, and flinching when it won't, is re-running the Bohr–Einstein debate — and Einstein lost. The mature move, in physics and now in software, is the same: stop demanding that the substrate be deterministic, and start engineering reliability on the layer above it. Do that, and a strange thing happens: the space of what you can build gets larger, not smaller.

The objection that stops AI at the door

Walk into any engineering team weighing whether to put a model into something that matters, and you will hear the same sentence, in some accent or other: I can't trust it, because it won't do the same thing twice. This is the single most common reason capable AI stalls at the threshold of real, load-bearing work. It deserves a straight answer rather than a pep talk, because the instinct behind it is not stupid. It is the accumulated wisdom of a discipline built on determinism.

Steelman it properly. A system you cannot reproduce is a system you cannot cleanly test: your test suite becomes a probability distribution, and a green run stops being proof. A system whose outputs vary is harder to debug, because you cannot reliably re-create the failure. It is harder to audit, because "show me why it did that" has no single answer. And it is harder to hold accountable, because accountability rests on the idea that the same inputs produce the same outputs, which is what lets us assign a cause. Reproducibility, testability, debuggability, auditability, accountability — these are not superstitions. They are the load-bearing walls of trustworthy software, and they were all quietly built on the assumption that the machine underneath is a clock.

Here is the twist that makes the problem genuinely interesting rather than a matter of just turning a knob. You cannot fix this by demanding determinism, because you never had it. Set a language model to temperature zero — greedy decoding, always take the single most likely next token, the setting people reach for precisely to make the thing repeatable — and it still is not deterministic. In 2025, researchers at Thinking Machines Lab ran the same prompt through the same open model a thousand times at temperature zero and got eighty different completions, the outputs first splitting apart at token 103. The cause turned out to be mundane and profound at once: it is not the sampler adding randomness, it is the arithmetic. The low-level operations that run the model are not "batch-invariant" — depending on how many other requests happen to be batched alongside yours on the server, the floating-point sums land in slightly different order, and floating-point addition is not associative. Force bit-identical results and you can, but you pay a ten-to-forty-percent speed penalty for the privilege.

Exhibit — Even at temperature zero, the same prompt returns dozens of different answers. The non-determinism engineers fear is real — but its root is ordinary arithmetic on shared hardware, not a ghost in the machine. Source: Thinking Machines Lab, "Defeating Nondeterminism in LLM Inference," 2025.

So the demand for a perfectly repeatable AI is not a modest ask that vendors are lazily refusing to meet. It is a demand for something that was never on offer, phrased as though determinism were the natural resting state of the world and probability an unfortunate contaminant. That picture is exactly backwards. To see why, it helps to ask where the craving for the clock comes from in the first place.

We are probabilistic machines that hate probability

The desire for a deterministic machine is not really about machines. It is about a much older human appetite for control, and the psychology of it is well-charted. In 1975 the psychologist Ellen Langer named the "illusion of control" — the tendency to expect a personal success rate higher than the odds can justify, simply because some feature of the situation feels like skill. In her lottery experiments, people who chose their own ticket valued it around three times as highly as people handed one at random, though the numbers on the ticket did not care who picked it. Control we do not have still soothes us, and losing the feeling of it stings even when the outcome is identical.

Sitting under that is a more specific fear: not of bad odds, but of unknown odds. In 1961 the economist Daniel Ellsberg posed a now-famous choice. One urn holds fifty red and fifty black balls; another holds a hundred in some unknown mix. Bet on a color and people reliably reach for the fifty-fifty urn, even when the payoffs are identical and the "ambiguous" urn is, by symmetry, an equally good bet. People will pay to avoid not knowing the odds — a preference so robust it violates the basic axioms of rational choice, and so stubborn that in later studies, being taught why it is irrational reduces it but does not abolish it. Add to that loss aversion, the finding from Kahneman and Tversky's prospect theory that a loss lands about twice as hard as an equivalent gain, and you have a mind that treats an unpredictable outcome not as neutral variance but as a threat to be minimized. A tool that "might" fail is felt as a tool that is already failing.

Exhibit — The craving for a deterministic tool is a stack of well-documented biases. None is irrational alone — but together they make us crave a certainty the world does not stock. Source: Langer 1975 (JPSP); Ellsberg 1961 (QJE); Kahneman & Tversky 1979 (Econometrica).

And here is the irony that should give the flinch pause about itself. The organ doing the flinching is not a clock. The best current account of how the brain works — predictive processing, the "Bayesian brain" — holds that the mind is fundamentally a prediction machine, running probabilistic models of the world, constantly generating expectations and updating them against the error when reality disagrees. On this view perception itself is a controlled hallucination, a best guess weighted by prior belief and incoming evidence. We are, in the most literal available sense, probabilistic engines that have talked ourselves into being revolted by probability. (We pulled on that strange convergence — the more modern AI diverges from biology, the more it seems to rebuild it — in our episode on AI reinventing the brain, number 27, last month.) That does not make the engineering worries evaporate. But it should reframe them: the discomfort is not the voice of physics or logic. It is the voice of a very old, very human need for control, wearing the costume of rigor. Which raises the obvious question — is the universe it wants to control actually built that way?

The universe was never a clock

For about two centuries, the smart money said yes. The purest statement of it came from Pierre-Simon Laplace in 1814. Imagine, he wrote, an intellect that knew the position and momentum of every particle in the universe and had the power to do the math; for such a mind "nothing would be uncertain and the future, just like the past, would be present before its eyes." This is the clockwork universe in one sentence: the present is the strict effect of the past and the strict cause of the future, and uncertainty is merely a confession of ignorance. Randomness, on this account, is not a feature of the world. It is a gap in our information. (Laplace himself never called his all-knowing intellect a "demon"; that was a later flourish.)

It is a beautiful idea, and it is worth noticing that the first crack in it did not come from the quantum weirdness everyone remembers. It came from heat. In the second half of the nineteenth century, Ludwig Boltzmann and the other founders of statistical mechanics did something that ought to sound impossible: they built the ironclad, engineering-grade laws of thermodynamics out of the assumption that a gas is a chaos of countless molecules whose individual motions we neither know nor track. The second law — entropy increases, heat flows from hot to cold, your coffee cools and never spontaneously reheats — turns out not to be a fundamental commandment at all. It is a statistical near-certainty, overwhelmingly likely because the disordered states vastly outnumber the ordered ones. The reliability we lean on when we design an engine is not the absence of molecular randomness. It is a stable pattern that emerges from it, the way a smooth average emerges from a million coin flips.

That is the crucial move, and it happened a full generation before quantum mechanics: rock-solid macroscopic law, standing squarely on top of microscopic disorder. The clock was already, quietly, made of dice. Physics just hadn't yet been forced to admit that the dice went all the way down.

God plays dice — and in 2022 we collected the proof

Then it was forced. In the 1920s quantum mechanics arrived and moved the randomness from our books to the world's. The decisive step was Max Born's rule of 1926: the theory does not tell you where a particle will be, only the probability of finding it there, given by the square of the wave function's amplitude. Werner Heisenberg's uncertainty principle sharpened the knife — certain pairs of properties, like position and momentum, cannot both be definite at once, not as a matter of clumsy instruments but as a matter of what "definite" can even mean. This was a different claim from Boltzmann's. The randomness of a coin flip is ignorance; in principle a good enough model of the toss removes it. Born was saying the randomness of the quantum world is not ignorance. There is no finer model underneath that makes it go away.

Einstein hated it. In December 1926 he wrote to Born the line that would follow him forever: "The theory produces a good deal but hardly brings us closer to the secret of the Old One. I am at all events convinced that He does not play dice." This was not a throwaway; it was the opening of a decade-long duel with Niels Bohr that began at the 1927 Solvay Conference and ran through the famous 1935 "EPR" paper, in which Einstein and two colleagues argued that quantum mechanics must be incomplete — that beneath the probabilities lay "hidden variables," a deterministic layer we simply hadn't found yet. It is worth sitting with how reasonable this was. Einstein was not being a crank. He was doing exactly what the engineer does today: refusing to accept that the apparent randomness was fundamental, insisting there had to be a deterministic clockwork underneath if only we looked hard enough.

For thirty years it was a matter of taste, untestable. Then in 1964 the physicist John Bell found the crack of light. He proved a theorem: any theory that restores determinism through local hidden variables — local meaning no faster-than-light spooky influence — must obey a certain inequality, a ceiling on how correlated two distant measurements can be. Quantum mechanics predicts that ceiling gets broken. This turned philosophy into an experiment. Starting with John Clauser in 1972, then Alain Aspect in the early 1980s closing the timing loopholes, then Anton Zeilinger's group sealing the last gaps, the experiments came in, again and again: the inequality is violated, exactly as quantum mechanics says. Clauser later admitted he had hoped for the opposite — "I was very sad to see that my own experiment had proven Einstein wrong." In 2022 the three of them shared the Nobel Prize in Physics for it.

Exhibit — Science advanced by giving up determinism — one hard result at a time. The demand for a deterministic substrate wasn't refuted by argument — it was refuted by measurement. Source: Nobel Prize in Physics 2022 (nobelprize.org); Bell 1964; Born 1926.

Honesty requires two caveats, because the popular version overshoots. Bell rules out local hidden variables — theories that keep both determinism and no-spooky-action. You can still save determinism if you are willing to pay a steep price elsewhere: "superdeterminism" keeps the clockwork but sacrifices the assumption that experimenters can choose freely what to measure; the "many-worlds" interpretation keeps a kind of determinism by saying every outcome happens in some branch. These are live, respectable positions. What is not on the table anymore is the comfortable, common-sense world Einstein wanted — local, definite, and deterministic all at once. And a second subtlety cuts the other way: determinism and predictability are not the same thing. Edward Lorenz showed in 1963, with a toy three-equation weather model, that a perfectly deterministic system can be intrinsically unpredictable — his "butterfly effect," where an unmeasurably small change in the start explodes into a totally different outcome. So even the systems that are clockwork often can't be forecast like one. Determinism was losing on both fronts: where it survived, it frequently didn't buy you the predictability you wanted it for.

Physics didn't collapse — it learned to build on sand

Here is the part that matters for anyone building things, and it is the reason this is a story of maturity rather than loss. When physics accepted that the substrate was probabilistic, it did not become vague, useless, or unable to make promises. It became one of the most precisely predictive endeavors in human history. Quantum electrodynamics predicts certain quantities to around a dozen digits. The transistors reading this sentence run on quantum mechanics — the same probabilistic physics — and switch reliably billions of times a second. None of that reliability comes from the substrate being deterministic. It comes from a layer built on top: statistical law that turns a haze of individual chances into a sharp, dependable aggregate, plus instruments that always report their error bars, plus engineering margins sized to the spread.

Weather forecasting is the cleanest illustration, because you can watch the field consciously make the switch. For decades, forecasting was deterministic: run the one best model from the one best estimate of today's atmosphere, and read tomorrow off the output. Lorenz's butterfly guaranteed that approach a hard ceiling — a single tiny error in the starting data blows up. So the field changed its whole philosophy. Modern forecasting is ensemble forecasting: at the European center, they run the model roughly fifty times over, each from slightly jiggled starting conditions, and read the spread of outcomes as a probability. "Seventy percent chance of rain" is not a hedge or an admission of laziness; it is a calibrated, verifiable claim, and a good forecasting system is one where it rains on seventy percent of the days it says seventy percent. By embracing the probability instead of fighting it, forecasts got both more accurate and more honest — they now tell you how much to trust them. That is the template physics discovered, stated as a recipe: don't demand a deterministic substrate; put a reliability layer on the probabilistic one.

Exhibit — Reliability is a layer you add on top of randomness — the same stack in physics and in AI. Neither lane makes its substrate deterministic. Both earn dependable behavior on the layer above it. Source: Compiled by Dan's AI Intel from the physics and ML literature cited here.

AI's breakthrough was a bet on dice

Which brings the same recipe home to AI, and reveals just how upside-down the flinch is — because probability is not a defect bolted onto modern AI. Probability is the reason it works at all. For the field's first forty years, the dominant program was the deterministic one. "Good Old-Fashioned AI" tried to capture intelligence in explicit, hand-written rules: symbols, logic, if-then expert systems that encoded a human specialist's knowledge directly. It was clean, inspectable, and repeatable — everything the flinch wants. And it hit a wall. The systems were brittle, shattering the moment a situation fell outside their hand-coded rules; they had no feel for uncertainty; and extracting all that knowledge from experts by hand was so slow and costly it earned a name, the "knowledge acquisition bottleneck." By the late 1980s the whole approach had collapsed into an "AI winter." Determinism, pursued directly, plateaued.

What broke the logjam was the opposite bet. Instead of writing down the rules, let a statistical model learn the patterns from mountains of data — no guarantees, no hand-built logic, just gradient descent nudging billions of parameters to make the training data slightly more probable, over and over. Rich Sutton called the moral of the story "the Bitter Lesson": across seventy years, "general methods that leverage computation are ultimately the most effective, and by a large margin," while approaches that lean on hand-built human knowledge keep stalling out. It is bitter precisely because it offends the engineer's taste for control. A modern language model does not look up the next word; it computes a probability distribution over every possible next token and samples from it. Its competence rises smoothly and predictably with scale — the "scaling laws," which hold so well that the compute-optimal recipe can be written down as a ratio (roughly twenty tokens of training data per parameter). Capabilities nobody explicitly programmed — arithmetic, translation, chain-of-thought — appeared as the models grew.

Intellectual honesty demands a flag here, because this is where hype often runs ahead of the evidence. Those "emergent" abilities were widely described as sudden, magical jumps; later work argued that much of the drama is a measurement artifact — that under smoother scoring metrics, more than ninety percent of the celebrated emergent leaps on standard benchmarks resolve into gradual, continuous improvement. The honest reading is not "it's all a mirage," but "the growth is real and mostly smooth, and we should resist the temptation to narrate it as sorcery." Either way, the load-bearing point stands: the deterministic, symbolic route to AI was tried in earnest for decades and stalled, and the probabilistic route is the one that delivered the thing now sitting in the engineer's editor. Asking it to stop being probabilistic is asking it to stop being the thing that works.

So build the way physics builds

If the flinch is a category error, the legitimate need underneath it is not — and the resolution is to satisfy that need the mature way rather than the impossible way. You do not get trustworthy AI by demanding a deterministic substrate, any more than you get trustworthy weather forecasts by demanding a deterministic atmosphere. You get it the way physics and forecasting got it: by building the reliability layer on top. This is not a metaphor being stretched; it is the actual working practice of the teams putting AI into serious systems. You wrap the probabilistic core in the things that convert a haze of individual chances into a dependable aggregate — test suites and evaluation harnesses that measure behavior as a distribution and hold it to a bar; type systems and schemas that constrain outputs into valid shapes; verification and validation passes that check the work instead of trusting it; retrieval that grounds the model in real sources; guardrails that catch the tail. The model is the statistical mechanics; this scaffolding is the thermodynamics.

A thread this show has pulled before lands squarely here. The reason a chart-topping benchmark score so often fails to survive contact with real work is that a benchmark measures a model's best-day peak, not the calibrated reliability that daily use demands — the argument of our benchmark-trap episode, number 42, a couple of weeks back. The same model, wrapped in a careful test-and-verification layer or thrown at the task raw, behaves like two different tools — because most of what we call reliability was never in the model at all. It lives in the layer around it. The model is the probabilistic substrate; whether it is reliable is decided almost entirely by what you build on top.

Exhibit — Physics, the mind, and AI each ran this exact play — twice they've already matured. In physics and in cognition the flinch has already lost. AI is the third time we're being asked to learn it. Source: Synthesis of the sources cited throughout, compiled by Dan's AI Intel.

There is a reward for making the shift, and it is not merely relief from anxiety. We already trust probabilistic systems everywhere they have proven themselves — the weather forecast that reroutes a flight, the medical test read as a likelihood ratio, the entire insurance industry, which is nothing but the disciplined monetization of odds, and, quietly, our own predictive brains. Each of those got more powerful, not less, once its field stopped pretending the underlying process was a clock and started managing it as a distribution. The same door is open for AI. A great deal of what looks impossible for a deterministic tool — fluent translation, open-ended reasoning, code from a plain-English wish — is possible only because the tool is probabilistic. Insisting it behave like a clock does not make it more trustworthy; it makes it less capable, and forecloses the very things the flinch is trying to protect.

Bottom line

Determinism is not the ground truth of the world. It is a useful special case — a fiction that happens to be an excellent approximation at human scale, which is exactly why it feels like bedrock. Physics learned, against the fierce resistance of its greatest mind, that reality is probabilistic all the way down, and it responded not by giving up on reliable prediction but by getting extraordinarily good at building reliability on top of chance. Cognitive science suggests we are, ourselves, machines of exactly that kind. AI is the third act of the same story, and the only real question is whether we take the mature path deliberately this time or relearn it the slow way.

So the flinch deserves a better answer than "get over it" and a better answer than "you're right, wait for a deterministic model that isn't coming." The answer is: you are protecting something real, and there is a proven way to protect it — the way physics and forecasting and medicine already do. Build the reliability layer. Then the genuinely open and interesting question is the one worth ending on, without a tidy bow: once we stop demanding that our tools be clocks, and start engineering trust on top of probability the way the universe apparently forces everyone to eventually — what becomes buildable that a deterministic machine could never have done?

Sources

Transcript

Alex: Einstein spent half his life certain the universe wasn't a casino.

Sam: In 2022, three physicists won a Nobel Prize for proving that it is.

Alex: Reality is probabilistic all the way down. And that's not a defeat — it's the exact playbook for what to do with AI. Welcome back to Dan's AI Intel — the show where we try to make sense of the fastest, most consequential shift any of us is likely to live through.

Sam: I'm Sam, here with Alex, and this one took us somewhere we didn't expect.

Alex: The show exists because AI is remaking the world week by week, the field's dense, the pace is relentless, and the honest gap between "keeping up" and "actually understanding it" keeps getting wider. That's the gap we're here to close.

Sam: So here's the trigger for today. Every engineer who's put a language model into something that actually matters has felt the same flinch: you run it twice, same input, and it gives you a different answer. Not wrong — just different. And something in the back of your head goes, wait, machines aren't supposed to do that.

Alex: We wanted to know if that flinch actually holds up — because it turns out this exact demand, a machine that behaves like a clock, has been put to the test before, at the highest level anyone's ever put it to the test.

Sam: We're going to follow that all the way down — through why our brains hate uncertainty in the first place, into the actual physics of a universe that may or may not be built on chance, and then back out to what any of it means for the thing sitting in your code editor right now.

Alex: And along the way we'll sit inside the argument between two of the biggest names in twentieth-century physics over whether God plays dice — with an actual Nobel Prize eventually settling it.

Sam: There's a turn in the middle of this I did not see coming until we were deep in the research — it changes what "reliable" is even supposed to mean.

Alex: We're not spoiling that one. If you like where this goes, do us a favor — hit follow wherever you're listening. It's free, and it's the thing that actually helps a small show like this find the next person. Okay, so let's put a name to the flinch, because I think most people who feel it have never actually said the sentence out loud.

Sam: Go on.

Alex: "I can't trust it, because it won't do the same thing twice." That's it. That's the whole objection. Show a language model the same prompt twice, and you can get two different, both-reasonable answers. And for anyone trained on traditional software — where the entire discipline is built on the same input always producing the same output — that reads as broken.

Sam: Okay, but I want to push back gently here, because that doesn't sound crazy to me. If my calculator gave me a different answer to two plus two depending on its mood, I'd throw it out the window.

Alex: And you'd be right to. That's actually the first thing worth saying clearly — the instinct isn't stupid, it's the accumulated wisdom of an entire discipline. Reproducibility, testability, debuggability, accountability — those aren't neuroses, they're the load-bearing walls of trustworthy software. If your test suite becomes a probability distribution instead of a yes-or-no, a green run stops meaning "it works." It just means "it worked that time."

Sam: Right — and if you can't reliably reproduce a bug, you can't reliably fix it. And "why did it do that" needs one answer, not a shrug.

Alex: There's an accountability version of this too, and it's maybe the most serious one. If something goes wrong — a bad medical suggestion, a wrong financial call — accountability rests on the idea that the same inputs produce the same outputs. That's what lets you point at a specific cause and say "this is what did it." Break that link, and "who's responsible" gets genuinely murky.

Sam: Okay, so I take it all back — that's not a neurosis, that's five separate, completely legitimate engineering and even legal concerns, all bundled into one flinch.

Alex: All five of them real. So steelman fully granted — that's a real cost, not a made-up one. But here's the twist that made me sit up when I read it: you can't fix this by demanding determinism. Because you never had it to begin with.

Sam: Wait, what do you mean? Isn't there a setting for that — turn the randomness down to zero?

Alex: There is. It's called temperature zero — greedy decoding, always take the single most likely next word. It's the exact setting people reach for specifically to force repeatability. And it still doesn't give you a repeatable machine.

Sam: How is that even possible? If it's always picking the most likely option, where's the room for randomness to sneak in?

Alex: In 2025, researchers at a lab called Thinking Machines ran the exact same prompt, through the exact same open model, a thousand times, at temperature zero. They got eighty different completions back. Not one weird outlier — eighty distinct answers, and they started splitting apart at token 103.

Sam: Okay, that's genuinely unsettling. What's actually causing that?

Alex: Here's the part I love, because it's so unglamorous — it's not some ghost in the machine, it's arithmetic. Depending on how many other people's requests happen to be running on the same server as yours at that moment, the low-level math gets batched slightly differently, and floating-point addition doesn't always land the same way when you add the same numbers in a different order. Force it to be bit-for-bit identical every time, and you can — but you pay for it, ten to forty percent slower.

Sam: So it's not that engineers haven't gotten around to fixing this. It's that fixing it costs real money, on every single request, forever.

Alex: Every single one. And think about what that means for the "just make it reproducible" instinct in an actual production system — a bank running fraud checks, a hospital running triage support, a company shipping code review. You're not choosing between "deterministic" and "probabilistic." You're choosing between "probabilistic and fast" and "still slightly probabilistic, and thirty percent slower and more expensive." The clean deterministic option was never actually on the menu.

Sam: So the demand for a perfectly repeatable model isn't a modest ask a vendor's being lazy about.

Alex: It's a demand for something that was never actually on the table, dressed up as though determinism were the natural resting state of the world, and probability were some contamination we let slip in.

Sam: Which is exactly backwards, if what you just described is true even down at the level of adding numbers together.

Alex: It is backwards. And to see why we keep reaching for that picture anyway, you have to stop looking at the machine — and start looking at us.

Sam: Okay, I like where this is going. Because you're telling me this isn't really about software at all.

Alex: It's about a much older appetite, and the psychology of it is genuinely well mapped. In 1975 a psychologist named Ellen Langer named something she called the "illusion of control" — we expect to succeed more often than the actual odds justify, just because some part of the situation feels like skill.

Sam: Give me the experiment.

Alex: Lottery tickets. People who picked their own numbers valued that ticket about three times higher than people who were handed a random one — even though, obviously, the numbers on a lottery ticket do not care who chose them.

Sam: Three times! For a ticket with literally identical odds.

Alex: Literally identical. Control we don't actually have still soothes us, and losing the feeling of it stings even when nothing about the outcome has changed.

Sam: Okay, so that's one piece. What else is stacked on top of that?

Alex: A more specific fear — not of bad odds, but of unknown odds. There's a famous thought experiment from 1961, the Ellsberg paradox. One urn has fifty red balls and fifty black balls. A second urn has a hundred balls in some ratio you don't know. Bet on a color, and people reliably go for the fifty-fifty urn — even when the payoff is identical either way, and the unknown urn is, by the math, an equally good bet.

Sam: We'll pay to avoid not knowing the odds, even when knowing them buys us nothing.

Alex: Even when it's proven to buy you nothing — and here's the detail that got me: you can sit someone down, walk them through exactly why that preference is irrational, and it shrinks a little. It doesn't go away. It's not a knowledge gap you can patch with a good explanation. It's closer to a reflex.

Sam: Which is a pretty uncomfortable thing to notice about yourself while you're reading a vendor's model card and going "yeah, but is it CONSISTENT."

Alex: And stack loss aversion on top of that — Kahneman and Tversky's famous finding that a loss hits us roughly twice as hard as an equivalent gain feels good — and you get a mind that treats "this tool might fail" not as neutral variance, but as an active threat. A tool that "might" fail already feels like it's failing.

Sam: Okay, honestly, that's just describing me every time I check if a flight's on time. Or, if I'm being fully honest, every time I re-run the same AI query three times just to see if I get a "better" answer.

Alex: Which, notice, is you personally performing the exact non-determinism the flinch claims to hate — and using it to soothe yourself. It's all of us. And here's the part that made me actually laugh when I read it: the organ doing all that flinching isn't a clock either.

Sam: What do you mean?

Alex: The leading account of how your brain actually works — it's called predictive processing, sometimes the "Bayesian brain" — says the brain isn't a recorder of reality. It's a prediction machine. It's constantly running a probabilistic model of the world, guessing what's coming next, and updating that guess against the error when reality disagrees with it. On this view, what you're calling "seeing" right now is closer to a controlled hallucination — your best guess, weighted by what you expected and corrected by what actually showed up.

Sam: So the thing inside my skull that's revolted by a probabilistic machine — is itself a probabilistic machine.

Alex: Right down to how you're perceiving this podcast right now. What you'd call "seeing" or "hearing" in the moment isn't a raw feed from your eyes and ears — it's your brain's best current guess, constantly checked and corrected against the error when the incoming signal disagrees with the prediction. Most of the time the correction is so small you never notice it happened.

Sam: So it's not that my brain sometimes guesses. It's that guessing is the whole operation, start to finish.

Alex: That's the claim. We picked at that exact convergence a couple of months back, in our episode on AI and the brain, number 27 — the more modern AI diverges from biology on paper, the more it keeps quietly rebuilding the same tricks. We're probabilistic engines that have talked ourselves into being offended by probability.

Sam: That doesn't erase the engineering worry, though. Right? Reproducibility's still a real cost.

Alex: It doesn't erase it — but it reframes it. The discomfort isn't the voice of logic or physics. It's a very old, very human need for control, wearing the costume of rigor. Which raises the obvious next question.

Sam: Is the universe it's demanding control over actually built like that?

Alex: For about two centuries, the smart money said yes. The purest version of the idea came from a French mathematician named Pierre-Simon Laplace, in 1814.

Sam: Set the scene for me.

Alex: Imagine an intellect — Laplace never actually called it a demon, that came later — that knew the exact position and momentum of every single particle in the universe, and had the computing power to do the math on all of it. For a mind like that, Laplace wrote, "nothing would be uncertain, and the future, just like the past, would be present before its eyes."

Sam: So the whole universe is just one enormous, fully wound clock. Know the position of every gear right now, and you can compute forwards or backwards forever.

Alex: That's it exactly. Randomness, on that picture, isn't a real feature of the world — it's just a confession that we don't have enough information yet. Fill in the gaps, and uncertainty vanishes.

Sam: It's a genuinely beautiful idea. Where does it start to crack?

Alex: And here's the first surprise — it doesn't crack where you'd expect. Everyone assumes the crack is quantum weirdness. It's not. It comes from heat.

Sam: Heat cracks the clockwork universe?

Alex: In the second half of the 1800s, a physicist named Ludwig Boltzmann did something that should sound impossible on its face: he built the ironclad, rock-solid laws of thermodynamics — the laws engineers still use to design real engines today — directly out of the assumption that a gas is a chaos of countless molecules whose individual paths we neither know nor track.

Sam: Wait — so the law that says heat flows from hot to cold, that your coffee cools down and never spontaneously reheats itself — that's not some fundamental commandment carved into the universe?

Alex: It's not a commandment at all. It's a statistical near-certainty. It's overwhelmingly likely, because the number of disordered ways for those molecules to arrange themselves absolutely swamps the number of ordered ways. The reliability engineers lean on when they design something is not the absence of molecular randomness — it's a stable pattern that emerges from millions of tiny random events, the same way a smooth average emerges from a million coin flips.

Sam: So the clock was already made of dice, a full generation before anyone got anywhere near quantum mechanics.

Alex: Physics just hadn't been forced to admit yet that the dice went all the way down. And notice what Boltzmann actually built with that — this wasn't a hand-wavy "eh, it's probably fine on average." Thermodynamics is engineering-grade. It's the math behind every engine, every fridge, every power plant on Earth. It just turns out that engineering-grade reliability was sitting on top of pure, untracked molecular chaos the entire time, and nobody needed to know a single molecule's individual path for the law to hold.

Sam: The reliability was never IN the molecules. It was in the pattern that shows up once you've got enough of them.

Alex: That's the sentence to hold onto, because we're going to say almost exactly that sentence again in about twenty minutes, about a completely different kind of system.

Sam: Okay, so then quantum mechanics shows up.

Alex: And it doesn't just add a little randomness at the edges — it moves the randomness from our books into the world itself. The decisive move was a 1926 rule from a physicist named Max Born: the theory doesn't tell you where a particle will be. It only gives you the probability of finding it there.

Sam: Not "we don't know yet" — "there is no fact of the matter until you look."

Alex: That's the whole shift. Werner Heisenberg sharpened it further — certain pairs of properties, like a particle's position and its momentum, literally cannot both be definite at the same time. Not because our instruments are clumsy. Because "definite" stops meaning what you think it means.

Sam: I already know who hated this.

Alex: Einstein hated it with his whole chest. December 1926, he writes to Born the line that follows him around forever: "I am at all events convinced that He does not play dice."

Sam: And that wasn't just a grumpy letter — that turned into an actual, sustained fight, right?

Alex: A decade-long duel with Niels Bohr, starting at a conference in 1927. It runs all the way through 1935, when Einstein and two colleagues publish what's now just called the EPR paper. Their argument wasn't "quantum mechanics is wrong" — it was "quantum mechanics must be an incomplete description of a deeper, deterministic reality." There had to be "hidden variables" underneath the probabilities, some machinery we just hadn't found yet.

Sam: Honestly — that's not crazy. That's just refusing to accept "we can't ever know" as a final answer. Isn't that what any good scientist should do?

Alex: It's exactly what any good scientist should do. Einstein wasn't being a crank. He was doing precisely what today's engineer does — insisting there has to be a deterministic clockwork underneath, if you just look hard enough. For thirty years, it was completely untestable. Just a matter of taste.

Sam: So what finally broke the tie?

Alex: In 1964, a physicist named John Bell found a crack of actual light in the problem. He proved a theorem — any theory that keeps determinism by adding "local" hidden variables, local meaning no faster-than-light spooky influence — has to obey a certain mathematical ceiling on how correlated two distant measurements can be. And quantum mechanics predicts that ceiling gets broken.

Sam: Which turns a philosophical shouting match into something you can actually go measure in a lab.

Alex: Exactly that. Starting with a physicist named John Clauser in 1972, then Alain Aspect in the early eighties closing off the loopholes, then Anton Zeilinger's team sealing the last gaps — the experiments kept coming back the same way, decade after decade, lab after lab. The inequality gets violated, exactly as quantum mechanics predicted.

Sam: Every single time, for fifty years, nobody manages to rescue the clockwork.

Alex: Nobody. Each new experiment closes off one more escape hatch — maybe the detectors were somehow talking to each other, maybe the timing was off — and the result never budges. That's not one clever result. That's the kind of pattern that ends an argument.

Sam: Give me the human moment there.

Alex: Clauser himself later admitted he'd hoped for the opposite result — he wasn't rooting for the answer he got. His quote is "I was very sad to see that my own experiment had proven Einstein wrong." That's not a triumphant scientist. That's someone who ran the numbers honestly and didn't like what came back.

Sam: Which is sort of the whole spirit of good science, isn't it — you don't get to keep the answer you were rooting for.

Alex: In 2022, all three of them — Aspect, Clauser, Zeilinger — split the Nobel Prize in Physics for it.

Sam: So it's not overstating it to say Einstein just... loses this one. On the record. With a Nobel Prize attached.

Alex: He loses it — with two honest caveats, though, because the popular version of this story overshoots. Bell only rules out LOCAL hidden variables. You can still keep some form of determinism if you're willing to pay a steep price elsewhere — "superdeterminism" keeps the clockwork but gives up the idea that experimenters can freely choose what to measure; "many-worlds" keeps a kind of determinism by saying every outcome happens, just in a different branch. Those are live, respectable positions.

Sam: But the comfortable, common-sense picture — local, definite, and deterministic, all at once — that's the one that's off the table.

Alex: That one's gone. Superdeterminism and many-worlds are both live, respected research programs — nobody serious is calling their proponents cranks. But notice the price of admission on both: superdeterminism gives up the idea that experimenters are ever really choosing what to measure; many-worlds says every possible outcome of every measurement really happens, just in a universe you're no longer in. That's a genuinely steep bill to pay just to keep the clockwork.

Sam: So it's less "determinism survived" and more "determinism survived, if you're willing to give up something almost as precious to get it."

Alex: That's a great way to put it. And there's a second twist that cuts the other way, and it's the one that actually matters most for what we do with computers. Determinism and predictability are not the same thing. A meteorologist named Edward Lorenz showed in 1963, with a tiny three-equation toy weather model — fully deterministic, no randomness anywhere in it — that the system was still intrinsically unpredictable. His butterfly effect: an unmeasurably small error in your starting numbers explodes into a totally different outcome.

Sam: So even the parts of the universe that ARE clockwork often can't actually be forecast like one.

Alex: Determinism was losing on both fronts at once — the substrate isn't deterministic, and even where it technically is, it doesn't buy you the predictability you wanted it for. And here's where it gets genuinely interesting, because physics didn't respond to any of that by throwing up its hands.

Sam: What did it do instead?

Alex: It got MORE precise, not less. This is the part that matters if you build things for a living, because it's the reason this is a story about maturity, not loss. Quantum electrodynamics — the theory built on top of all that fundamental randomness — predicts certain quantities to something like a dozen decimal digits of accuracy. That's one of the most precise predictions in the history of science, coming out of the single theory that killed the idea of a clockwork universe.

Sam: So the theory that told us "you can never know for certain where this one particle is" is also the theory giving us more decimal places of confidence than almost anything else we've got.

Alex: That's the paradox, stated as sharply as I can put it. Certainty about the individual event, gone forever. Precision about the aggregate, better than it's ever been.

Sam: Built on top of a substrate you just told me is irreducibly random.

Alex: Every transistor in the phone in your pocket runs on that same probabilistic physics, and switches reliably billions of times a second, every second, for years. None of that reliability comes from the substrate secretly being deterministic after all. It comes from a layer built ON TOP of it — statistical law that turns a haze of individual chances into a sharp, dependable aggregate, plus instruments that report their own error bars, plus engineering margins sized to the actual spread.

Sam: So nobody designing a phone chip is sitting there hoping any individual electron behaves. They designed the whole system so it doesn't matter what any one electron does.

Alex: That's exactly the shift. The individual event stays genuinely unpredictable, forever. The aggregate becomes almost boringly reliable — which is the opposite of what your gut expects, and it's worth sitting with, because that's the whole move we're about to watch AI make too.

Sam: Give me something I can picture. That's still pretty abstract.

Alex: Weather.

Sam: Okay, I like weather. Very relatable.

Alex: For decades, forecasting was fully deterministic — run the one single best model, from the one single best estimate of today's atmosphere, and read tomorrow straight off the output.

Sam: And Lorenz's butterfly effect just... torpedoes that entire approach.

Alex: Puts a hard ceiling on it. One tiny error in the starting numbers, and the whole forecast blows up downstream. So the field changed its entire philosophy. Modern forecasting is ensemble forecasting — the European center runs the model roughly fifty times over, each run starting from very slightly jiggled conditions, and reads the SPREAD of those fifty outcomes as an actual probability.

Sam: So "seventy percent chance of rain" isn't a hedge. It's not the forecaster covering themselves.

Alex: It's a calibrated, verifiable claim. A well-run forecasting system is one where, on the days it says seventy percent, it genuinely rains about seventy percent of the time. By fully embracing the probability instead of fighting it, forecasts got both MORE accurate and more honest — they now tell you exactly how much to trust them.

Sam: That's the whole recipe in one line, isn't it. Don't demand the substrate be deterministic — put a reliability layer on top of the probabilistic one.

Alex: That's the recipe, stated as cleanly as physics has ever stated it. Which brings us to the part that made me want to make this episode in the first place.

Sam: The AI part.

Alex: The AI part — because it reveals just how upside-down the flinch really is. Probability isn't a flaw bolted onto modern AI as an unfortunate side effect. Probability is the actual reason it works at all.

Sam: Wait, really — the whole reason?

Alex: For the field's first forty years, the dominant approach was the deterministic one. It's sometimes called "Good Old-Fashioned AI" — try to hand-write intelligence as explicit rules. Symbols, logic, if-then expert systems encoding what a human specialist knows.

Sam: That sounds exactly like what the flinch WANTS — clean, inspectable, repeatable.

Alex: Everything the flinch wants, precisely. And it hit a wall hard. The systems were brittle — they shattered the moment reality fell one step outside their hand-coded rules. They had no way to represent uncertainty at all. And getting all that expert knowledge out of a human's head and into rules by hand was so slow and so expensive it earned its own name — the "knowledge acquisition bottleneck."

Sam: So the deterministic bet on AI — the one that felt safe — actually failed first.

Alex: By the late eighties, the whole approach had plateaued and then collapsed into what people call an AI winter. Funding dried up, projects shut down, and "artificial intelligence" briefly became almost a dirty phrase to put in a grant application.

Sam: Which is a genuinely wild thing to sit with, given everything we cover on this show. The clean, controllable, rule-based version of AI is the one that actually died first.

Alex: What broke the logjam was the opposite move entirely. Instead of writing down the rules, let a statistical model LEARN the patterns from an enormous pile of data. No guarantees, no hand-built logic — just gradient descent, nudging billions of numbers over and over to make the training data very slightly more likely each time.

Sam: That sounds like giving up on control completely.

Alex: A researcher named Rich Sutton called the moral of that story "the Bitter Lesson" — across seventy years of the field, general methods that just lean on more computation keep winning, by a large margin, over approaches built on hand-crafted human knowledge. It's called "bitter" precisely because it offends the engineer's taste for control.

Sam: Okay, tie this to what's actually running when I type into a chat window.

Alex: A modern language model doesn't look anything up. At every single step, it computes a full probability distribution over every possible next word, and it samples from that distribution. And its competence rises smoothly and predictably as you scale it up — the so-called scaling laws hold well enough that you can write the compute-optimal recipe down as a ratio, something like twenty tokens of training data for every parameter in the model. Capabilities nobody explicitly programmed — arithmetic, translation, step-by-step reasoning — just showed up as the models got bigger.

Sam: Now, I know you well enough to know you're about to add an honest caveat right here.

Alex: You do know me. Those "emergent" abilities got described everywhere as sudden, almost magical jumps — and later analysis pushed back hard on that: under smoother, more careful scoring, more than ninety percent of those celebrated emergent leaps resolve into gradual, continuous improvement, not sorcery.

Sam: So it's not "it's all a mirage" — it's "the growth is real, and mostly smooth, and we should stop narrating it like magic."

Alex: Exactly that — and I think that caveat matters here more than almost anywhere else in this episode, because it's tempting to reach for "magic" on both sides of this argument. The flinch wants AI to be a boring, legible machine; the hype wants it to be inexplicable sorcery. The actual truth is duller and more useful than either: a statistical process, scaled up, producing smooth, mostly-predictable improvement.

Sam: But the load-bearing point survives either way.

Alex: It survives completely — the deterministic, hand-coded route to AI was tried in earnest for decades, by very smart people, and it stalled. The probabilistic route is the one that produced the thing sitting in your editor right now. So asking it to stop being probabilistic isn't a reasonable safety request.

Sam: It's asking it to stop being the thing that actually works.

Alex: Which brings us to the resolution — and I think this is the single most useful idea in the whole episode. If the flinch is a category error, the legitimate NEED underneath it is not. You satisfy that need the mature way, not the impossible way.

Sam: So — not "wait for a deterministic model." That's just not coming.

Alex: Not coming, and it was never really the point. You get trustworthy AI the exact same way physics and forecasting got trustworthy predictions — you build the reliability layer ON TOP of the probabilistic core, not inside it. Test suites and evaluation harnesses that measure behavior as a distribution and hold it to a real bar. Type systems and schemas that force outputs into valid shapes. Verification passes that check the work instead of just trusting it. Retrieval that grounds the model in real sources. Guardrails that catch the tail cases.

Sam: The model is the statistical mechanics — and that whole layer around it is the thermodynamics.

Alex: That's exactly the shape of it, and it's not abstract — it's the actual working practice of teams putting this stuff into serious systems today. An eval suite doesn't demand the model say one single correct thing; it measures behavior as a distribution and holds that distribution to a bar, the same way the ensemble forecast holds itself to "seventy percent means seventy percent." A type system doesn't ask the model to be deterministic; it just refuses to let a malformed answer out the door, whatever produced it.

Sam: So the mistake the model makes is still allowed to happen — it just gets caught before it reaches anyone.

Alex: Caught, and caught systematically, not by luck. Same idea with retrieval — instead of trusting the model's raw, probabilistic memory of a fact, you ground it: hand it the actual source document at the moment it answers, so the guess has something real to check itself against. You're not making the underlying process less probabilistic. You're changing what it's guessing FROM.

Sam: It's the ensemble forecast again, basically. Don't trust one draw — build a system around the draws. And this connects straight back to something we dug into a few weeks ago, in our episode on the AI benchmark trap, number 42 — the reason a model that tops the leaderboard so often falls apart on real work is that a benchmark measures its best-day peak, not the calibrated reliability daily use actually demands. The same underlying model, wrapped in a careful test-and-verify layer versus thrown at a task raw, behaves like two completely different tools.

Alex: Because most of what we call "reliability" was never sitting inside the model at all. It lives in the layer built around it — the harness, basically, which is exactly the idea we went deep on just a couple of episodes back, number 48. Whether a model is trustworthy is decided almost entirely by what gets built on top of it, not by the model itself.

Sam: Okay, so pull this all the way back together for me. Physics, the brain, and AI — same shape, three times?

Alex: The exact same arc, three times over. Physics wanted a clock, got proven wrong by Einstein's own gold standard of rigor, and responded by building the most precise predictive science in human history — on top of chance. The mind wants control, and it turns out the mind itself runs on educated guesses, constantly updated. And AI wanted hand-coded rules, watched them collapse, and won everything it's won by betting on probability instead.

Sam: In physics and in cognition, the flinch has already lost, twice. AI's just the third time we're being asked to learn the same lesson.

Alex: And here's the reward for actually making that shift — it's not just relief from the anxiety. We already trust probabilistic systems everywhere they've earned it. The weather forecast that reroutes your flight. A medical test read as a likelihood, not a verdict. The entire insurance industry, which is nothing but the disciplined monetization of odds. Our own predictive brains, every waking second.

Sam: And every one of those got MORE powerful once the field stopped pretending its underlying process was a clock, and started managing it as a distribution instead.

Alex: None of them got there by demanding certainty first and refusing to ship until they had it. They shipped the probabilistic version, instrumented it honestly, and let the reliability layer do the actual work of earning trust over time.

Sam: That same door is standing wide open for AI right now. A huge amount of what feels impossible for a deterministic tool — fluent translation, open-ended reasoning, working code out of a plain-English wish — is only possible BECAUSE the tool is probabilistic. Demanding it behave like a clock doesn't make it more trustworthy. It makes it less capable, and it forecloses the exact things the flinch was trying to protect in the first place. Which is a genuinely different way to think about "reliable" than the one most of us grew up with. Reliable used to mean "the same every time." Now it means "predictably distributed, and honestly measured."

Alex: That's the upgrade, in one sentence. And it's not a downgrade dressed up in nicer language — it's a strictly more powerful idea, because it's the only version of "reliable" that was ever actually true, in physics, in your own head, or in a language model.

Sam: So determinism was never the ground truth.

Alex: It's a useful special case — a fiction that happens to be an excellent approximation at human scale, which is exactly why it feels like bedrock under your feet. Physics learned, against the fierce resistance of its single greatest mind, that reality is probabilistic all the way down — and it answered not by giving up on reliable prediction, but by getting extraordinarily good at building reliability on top of chance.

Sam: And cognitive science says we're built the exact same way — a prediction engine that mistakes its own best guesses for solid ground.

Alex: Which means the flinch we started with tonight isn't really an AI story at all. It's the oldest story there is, just wearing a new costume — the same argument physics had with itself, that the mind has with itself every waking second, now showing up in a code review.

Sam: So if you're carrying just three things out of this one — the flinch is real and half-right, five legitimate engineering virtues underneath it, but demanding a deterministic model is asking for something that was never on the table, not even at temperature zero. Reality itself runs the same way, all the way from a hot cup of coffee to an entangled particle, and Einstein — of all people — spent thirty years betting against it and lost. And the fix, in physics and in AI both, was never "wait for the randomness to go away." It was "build something reliable on top of it."

Alex: That's the whole shape of it. So here's where I want to leave it, because I don't think there's a tidy bow on this one. The flinch deserves a better answer than "just get over it" — and a better answer than "you're right, wait for the deterministic model." The real answer is: you're protecting something genuinely real, and there's a proven way to protect it, the same way physics and forecasting and medicine already do it. Build the reliability layer.

Sam: And then the actually interesting question is the one you don't get to answer neatly. Once we stop demanding our tools be clocks, and start engineering trust on top of probability the way the universe apparently forces everyone to do it eventually —

Alex: — what becomes buildable, that a deterministic machine could never, ever have done?

Sam: That's the one I'm going to be turning over for a while.

Alex: That's it for today — thank you so much for listening. I hope you came away seeing a little more clearly where this is all actually heading. It's a genuinely complex, fast-moving picture with a brutally short knowability horizon, and that's exactly what makes it worth following this closely.

Sam: And one honest note on how this show is actually made — it's AI-generated. Dan builds a custom stack of AI tools to research, analyze, verify, and illustrate the questions worth understanding, mostly to learn them himself, and he publishes it for anyone who'd like to follow along. AI-assisted, fact-checked, worth a second look.

Alex: Before we go, one genuinely useful thing you can do: follow the show. Whatever app you're listening in right now, there's a follow or a plus button — it's one tap, it's free, and it does two things. You'll get every new episode the moment it lands, and honestly, for a small independent show like this one, a follow is the single biggest lever there is for helping it grow and reach the next person trying to make sense of all this.

Sam: So if any of this was worth your time — go ahead and hit follow.

Alex: And one quick thing before we actually go. If there's something in this episode you'd push back on, or a thread you want us to pull harder on next time, tell us — podcast at connectiveshift dot com. We read every single message, and it genuinely shapes what we dig into next.

Sam: See you next time.