Isaac Newton

In the [YOU] on AI Field Guide

The cycle that began with [YOU] on AI asks what it means to take the orange pill—to see the machine clearly, without narcotic hype or paralyzing fear. Newton is the most exacting instrument for that seeing, because he defines the very thing these machines refuse to be. To read him alongside contemporary AI is to be handed a measuring instrument with two scales: comprehension on one side, capability on the other. The models score off the charts on capability and register almost nothing on comprehension in Newton's sense, and he is the thinker who makes the gap between the two scales legible as something more than a marketing complaint.

His calculus—the method of fluxions Newton developed in the 1660s to turn his laws into predictions—runs in direct descent through the backpropagation algorithm that trains every neural network. The chain rule he and Leibniz bequeathed is the mathematical spine of every training run. But the directionality has inverted: Newton's calculus ran from known law to deduced consequence; backpropagation runs from desired consequence to an unknown function that approximately produces it. The tool is continuous; the epistemology has reversed. This reversal is, in microcosm, the entire transformation the cycle traces.

Newton's clockwork universe—every future state fixed by the present, a determinism so complete that Laplace imagined an intelligence who could predict all of history from a single snapshot—is the limiting case against which the cycle tests what, if anything, in a human life exceeds mechanism. A trained model is Laplacean in exactly this sense: identical input, identical weights, identical output, every time. Engineers inject randomness deliberately, to prevent rigidity. Strip it away and the model is the clockwork Newton built, scaled up and trained on human text, talking back. The cycle's question—what remains irreducibly ours?—is sharpest when held against Newton's clock.

And then there is the secret Newton the legend omits: the man who wrote over a million words on alchemy and biblical prophecy, who believed nature and scripture alike encoded a hidden rational order he might decode. He was not two men—the rigorous physicist and the occult enthusiast. He was one man whose faith was that the world is fully writable, that a sufficiently penetrating mind can read it. The large language models that work without finding any writable law are the first serious evidence that this faith may apply only to the simple corner he explored.

Origin

Isaac Newton was born on Christmas Day 1642 (Old Style) in Woolsthorpe, Lincolnshire, a premature infant so small, his mother later said, that he could have fit in a quart pot. His father died before his birth; his mother remarried and left him with his grandmother for nine years, a wound he never fully healed. He entered Trinity College, Cambridge in 1661, found the curriculum medieval and supplemented it obsessively with Descartes, Galileo, and Kepler. During the plague years of 1665–66, when Cambridge closed and Newton retreated to Woolsthorpe, he invented the calculus, discovered the composition of white light, and formulated his theory of universal gravitation. He was twenty-three. He published almost none of it for decades.

The Philosophiæ Naturalis Principia Mathematica appeared in 1687, under pressure from Edmund Halley, who paid for its printing. It was immediately recognized as a work without precedent: the universe reduced to a single mathematical framework, every known celestial phenomenon derivable from first principles. Newton had achieved what no scientist before or since has achieved in a comparable sweep—he had compressed the cosmos. His method was unmistakable: find the generating rule, derive the rest, and refuse to assert what the evidence does not force. Hypotheses non fingo, from the General Scholium of the 1713 edition, became the motto of a new standard of rigor.

He served as Warden and Master of the Royal Mint, pursued alchemy and biblical chronology with the same obsessive intensity he brought to physics, presided over the Royal Society, and was knighted in 1705. He died in 1727, leaving behind a Principia that would steer spacecraft three centuries later and a private archive that revealed a man far stranger, and far more human, than the legend.

Key Ideas

Comprehension as the engine of prediction. For Newton, to predict and to understand were the same act. You could foretell a comet's return because you grasped the gravitational mechanism governing it; the successful prediction certified the understanding. Machine learning has pried these apart: a weather model can outpredict physics-based simulations without containing any physics, a protein-folding system predicts structures without modeling the forces that fold them, a language model predicts the next token in a legal argument without comprehending law. The divorce is genuine and disorienting. It was nearly unthinkable in the framework Newton built, and now it is a daily industrial fact.

Laws versus correlations. A Newtonian law states a necessary, mechanism-backed relationship that supports counterfactuals—if you doubled the mass, the force would double, whether or not anyone ever measures it. A correlation is a regularity in observed data carrying none of these guarantees. Neural networks are, at bottom, magnificent correlation engines. They detect and exploit statistical regularities at a scale no human could match, and the fluency-authority decorrelation—the model's tendency to assert false claims with the confidence of true ones—is exactly this: the correlation-is-not-causation problem wearing a linguistic face.

Hypotheses non fingo. When pressed to explain why gravity acts across empty space, Newton refused to invent a mechanism he could not justify. The contemporary temptation runs the other way: to dress up opaque systems with confident post-hoc narratives, to let a model emit a plausible-sounding “explanation” of its own behavior that is itself just more generated text, unmoored from actual computation. Mechanistic interpretability is the field's most Newtonian response—the attempt to actually see what is happening inside, rather than narrating it.

The world-as-text wager. Newton's deepest faith—expressed through physics, alchemy, and biblical chronology alike—was that reality is fundamentally intelligible, fully writable, available to a sufficiently penetrating mind. Large language models work without finding any such compact structure, through billions of inscrutable parameters, suggesting that for the most complex and human-relevant domains there may be no short text to read. The success of the unintelligible method is the most unsettling evidence the wager has ever faced.