Return of the Natural Philosopher

AI is not just automating method. It is exposing what method was always hiding. — from A Renaissance of Thought by Jake W. Casselman.

AI is not just automating method. It is exposing what method was always hiding.

Chapter 6 of A Renaissance of Thought This is the sixth chapter of a connected argument. It builds on everything before it — the conversational posture, the minimum viable scaffolding, the descent into understanding — and reads best after the earlier chapters. If you've arrived here first, you can still follow it, but the foundations are laid in the chapters that come before. A word of warning: this chapter dives considerably deeper into philosophy than the ones before it — into epistemology, ontology, and the history and sociology of science. It also runs longer and more demanding by design. In preparation for the forthcoming series First Paddle Out, I am using this chapter to test a few ideas. If parts of it ask more of you than usual, that is deliberate. Take it slowly.

How the Fracture Happened

There was a time when a single person could hold what we now call science and what we now call philosophy inside the same mind without any sense of contradiction. Newton's Principia is simultaneously a work of mathematics, physics, and natural philosophy — the categories had not yet been invented to separate them. Likewise, Darwin wrote with the precision of a naturalist and the breadth of a thinker who understood he was making a claim about the nature of life itself.

These people are called natural philosophers. The term sounds archaic. It is worth asking why we stopped using it.

The fracture was not a philosophical event. It was an institutional and economic one. Once method became reproducible — once you could train a person to run a regression, to conduct an assay, to execute a procedure — institutions optimized for that. Method was scalable. Epistemology was not. You could credential someone in technique. You could not credential someone in the ability to ask whether the technique was revealing what they thought it was revealing.

So the natural philosopher split. The epistemologist went one way — into philosophy departments, into corners of academia that science increasingly regarded with polite disinterest. The methodologist went the other — into labs, into grants, into the machinery that produces papers. The two stopped talking. Both stopped noticing they needed the other.

The logical positivists — Vienna Circle, 1920s, Carnap and Schlick — formalized the fracture philosophically. Their argument was clean and radical: only empirically verifiable statements are meaningful. Metaphysics is nonsense. Philosophy of science is housekeeping. What matters is the method, properly applied.

They did not win the philosophical debate. Popper dismantled them. Kuhn historicized them. Feyerabend burned the house down. But they won the institutional debate, and that is the one that determined how science actually got organized. The fracture they formalized became the architecture of the modern university. PhDs stopped knowing what the Ph in PhD stood for.

It stands for philosophiae doctor. Doctor of philosophy. That used to mean something.

The Justification That Is Now Dissolving

Here is the part that matters. The fracture was not simply a mistake. It had a justification, and the justification was real.

Experiment cost.

If running a physical experiment requires a lab, equipment, trained personnel, and years of setup, then you cannot afford to explore. You must proceed methodically, with clear hypotheses, disciplined procedure, rigorous documentation. The cost of a bad method is too high. You optimize for reliability and reproducibility rather than breadth — and the questions of whether you are even asking the right thing get deferred, because you cannot afford the alternative.

The methodological era produced everything it promised. The disciplined specialization of the last hundred and fifty years is why we can transplant organs and model the atmosphere and design materials at the atomic level. The fracture was load-bearing during the scaling phase. It would be dishonest to pretend otherwise.

But the justification was never epistemological. It was economic. And economics change.

I work with ERA5 — a global atmospheric reanalysis dataset covering over eighty years of the Earth's atmosphere at high resolution. To produce the equivalent understanding from physical weather stations and upper-atmosphere balloons would require decades of coordination across thousands of sites and billions of dollars. As a compute dataset, I can run it tonight. I can explore freely. I can follow a thread I did not plan to follow and see where it leads. The question what if I just try this costs nothing.

The natural philosopher did not return because someone decided to reverse the fracture. The fracture is being reversed by collapsing experiment cost. In fields where the world can be modeled computationally, the economic justification for methodological discipline over exploratory philosophy is weakening from underneath. And AI accelerates this collapse dramatically. The methodological work is not just cheaper. It is increasingly automated.

Which means the methodological layer — the layer that the fracture decided was the important one — is no longer the scarce resource. The question is what that leaves behind.

Simultaneous Discovery and the Shape of What Is True

There is a phenomenon in the history of science that has never received the explanation it deserves.

Newton and Leibniz discover calculus independently, within years of each other. Darwin and Wallace arrive at natural selection simultaneously, from entirely different bodies of evidence. Multiple inventors converge on the telephone within months. Oxygen is discovered within years by researchers who never met. The list goes on, longer than most people realize, and the standard explanation — that the idea was simply in the air — is no explanation at all. It restates the phenomenon in metaphor and moves on.

The actual mechanism is the minimum viable scaffolding.

When the scaffolding of a field reaches a particular configuration — when the accumulated prior knowledge points toward a specific unanswered question, when the conceptual vocabulary of the era has grown to the edge of a particular gap — multiple people standing in that position will look into the same region and find the same thing. Not because they copied each other. Not because creativity is an illusion. Because they were standing in approximately the same epistemic position (epistemic — having to do with knowledge: what is known and how one comes to know it), and from that position, only certain views were available.

This is Platonic in the most literal sense. The discovery was already there, waiting to be found, in the structure of the problem space. Philosophers of science have a name for the version of this view that survives modern scrutiny: structural realism — the position that what science actually discovers is real, mind-independent structure, even as the particular descriptions and frames we wrap around it keep changing. (Its older and broader ancestor is Platonic realism; if you want to follow the idea further, those are the two threads to pull.) The researchers did not invent it. They found it, each from their own vantage point, each rendering it in their own way. What they reached for was the same. What they produced was the articulation of it in the language and methods and preoccupations of the person doing the reaching.

The tragedy that usually follows — the priority disputes (the fights over who got there first), the accusations of theft, the bitterness of races that multiple people believed they were running alone — is the tragedy of people who do not understand what truth is. If you believe you created the idea, then someone else arriving at the same idea is a theft. If you understand that you found the idea — that it was already there, in the structure of things, and your contribution was the particular angle from which you first made it visible — then simultaneous discovery is not a threat. It is triangulation. It is confirmation that the thing is real.

The right response to finding out that someone else reached the same place you reached is to compare notes on exactly where you were standing when you saw it. What angle did you approach from? What did you see clearly from where you were that I could not see from where I was? What did I see that you missed?

That is the most productive Socratic dialogue possible. Two people pointing at the same feature of Platonic space from different positions, pooling the geometry of their respective views. We almost never do it. We fight instead.

I have learned to do a small version of this on purpose. When I have already worked out my own framework for something and I want to know whether it is real, I withhold it. I do not lead with it. I let the other person build their account from their own materials, uncontaminated by mine, and I listen for where they are standing. If they arrive somewhere different, I have learned something about the terrain I could not see. And if they arrive at the same place I did — by a route I never took, from evidence I never had — that convergence is worth more than any agreement I could have manufactured by showing my hand first. I suspect you have felt this too: the quiet, almost physical confirmation of hearing your own conclusion come back to you in someone else's words, reached entirely on their own. That feeling is triangulation. It is the structure of the thing announcing that it was there all along.

Confirmation Across Epistemic Vantage Points

Science already understands one kind of triangulation. Spatial replication — running the same study in different populations, different geographies, different datasets — is the methodological standard for robust findings. A result that holds up across independent reproductions is more credible than one that appeared once under controlled conditions. The logic is sound and it has served us well.

But we apply this logic only in physical space. We replicate across contexts. We do not replicate across ways of knowing.

This is not hypothetical. It happens whenever a researcher stands at the intersection of two fields that have been circling the same phenomenon from opposite sides — one with the vocabulary of one discipline, another with the vocabulary of a completely different one, neither aware the other was pointing at the same structure. The convergence is not planned. It is what emerges when someone occupies both vantage points and looks. That is not spatial replication. That is two fields confirming the same feature of conceptual space from opposite ends — which is a stronger claim about what is real than any single field running the same study twice.

Two independent spatial replications could share the same methodological assumptions, the same blind spots built into the paradigm of a single field. Two independent epistemic vantage points, arriving at the same structure from entirely different scaffolds, from different vocabularies, different measurement traditions, different ontological commitments (ontology — the study of what fundamentally exists and what kinds of things are real) — that is much harder to explain as coincidence or shared error.

Luckily, this is changing — a growing trend. Research communities increasingly cultivate interdisciplinary exposure for exactly this reason, and there are now cross-disciplinary journals, joint institutes, and grant programs that reward reaching across fields. The instinct that fields have something to learn from each other is alive and, in places, well funded.

But notice what those channels actually reward. They reward combination — bring the tools of one field to the problems of another, and the novelty of the pairing is the contribution. What they do not have a channel for is convergence as confirmation: the finding that two fields, working independently with different methods and different vocabularies, have been looking at the same underlying structure from opposite sides — and that this agreement is itself evidence the structure is real. You can publish the borrowing. You cannot easily publish the triangulation. You are a climate scientist or a thermodynamicist. You publish in climate journals or thermodynamics journals. The institutional architecture has a slot for "I used your method" and almost none for "we independently found the same thing, so it must be there."

Conceptual space is multidimensional. We have been treating it as flat.

Alongside that institutional shift, AI is beginning to dissolve the remaining walls. Not by resolving the disciplinary questions — by making the minimum viable scaffolding cheap enough that a climate scientist can actually understand thermodynamics, and a thermodynamicist can actually understand climate dynamics, and the conversation can begin across the gap that has kept them separate for decades.

What Feyerabend Saw

Paul Feyerabend spent most of his career as the science establishment's most inconvenient critic. Against Method, published in 1975, made a claim so uncomfortable that many scientists still refuse to engage with it directly.

The claim: the scientific method is a retrospective fiction. The actual process of discovery is anarchic — opportunistic, intuitive, driven by analogy and hunch and the particular personality of the researcher at least as much as by any formal procedure. Scientists write the methods section after they know the answer. They present a clean logical progression that bears little relationship to the actual path they walked. The method is how it gets reported, not how it gets done.

This was not well received. It still is not. But the response was always more sociological than logical — the discomfort of an institution having its self-presentation questioned, rather than a genuine refutation of the claim.

What Feyerabend saw is increasingly hard to deny in the age of computation. When you can query ERA5 at will, the pre-registration of hypotheses becomes an interesting constraint on what you publish, not a description of what you actually did. The exploration came first. The method is the account you give afterward.

There is nothing dishonest about this. It is how thinking works. Ideas arrive messily, with tangents and wrong turns and serendipitous collisions. The formal presentation cleans it up. The cleaning-up is not a lie — it is a translation. You are translating the anarchic process of discovery into the reproducible language of methodology so that other people can verify and build on what you found.

But if the method is translation rather than the thing itself, then automating the method does not automate the thing itself. It automates the translation layer. Which raises the question of what the thing itself actually is, if you strip the translation away.

The answer is: the vantage point.

The Vantage Point Is the Irreducibly Human Contribution

Paint a mountain. Walk around it until the view resolves — until the angle is right, until the light falls the way you need it to fall, until the composition does what you are looking for. Plant your easel (the stand that holds the canvas). Begin.

At what point in that process did you create the mountain? You did not. The mountain was there. Your contribution is entirely in where you chose to stand, which angle you found worth rendering, which features of the mountain your particular position and training and preoccupation allow you to see, and how the light falls differently from where you are than it would fall from anywhere else.

If another painter comes along after you and plants their easel in exactly the same spot — by coincidence, by following the same logic about where the view resolves — they did not steal your painting. They found the same vantage point. You would share a subject. The paintings would not be identical. Two people standing in the same position see the same mountain through the filter of everything they are. The rendering is personal even when the position is shared.

This is simultaneous discovery. Two scientists finding the same spot. Not copying. Finding.

Now: what happens if you name the mountain after yourself?

This is exactly what happened with the Hovmöller diagram — a way of flattening the weather into a single still picture, so that something traveling across the planet over days shows up as a diagonal streak you can see at a glance. Time runs along one axis, longitude (position east–west around the globe) along the other. It is a coordinate system for spotting atmospheric waves and anything that moves east-west across the tropics with a detectable rhythm. It is an extraordinarily useful visual frame for atmospheric scientists. It is also just a choice of axes. Hovmöller did not discover time. He did not discover longitude. He found a viewing angle that made certain features of the atmosphere visible in a way they had not been visible before. And yet his name is now stamped across it.

Science named it after him.

To be fair to Hovmöller, the eponym was not his doing — the field reached for his name, as fields habitually do. But the habit is worth examining, because naming an idea after a person quietly declares that they made the mountain. No one makes the mountain. You found the view. The idea existed before you stood where you stood. Your contribution was the standing, the looking, the craftsmanship of the rendering. That is a real contribution. It is also not ownership of the thing itself.

The frame you choose to render the phenomenon is a contribution of exactly the same kind as the view — real, skilled, creditable — and exactly as little a basis for ownership. You chose the axes, but you did not invent them; you found the ones that fit. Someone else could have found the same axes independently and arrived at the same diagram. If they were doing good work — if they were focused on the mountain and not on the real estate — they would not have been thinking about the coordinate system at all until they needed it. You only look down at where your feet are after you have found the angle you were looking for.

Science would be a different culture if it understood this. If priority disputes dissolved into geometry problems — you were here, I was here, together we have a better picture of the thing — and if the credit accrued to the quality of the vantage point — including the real credit owed to whoever opened it first, who showed the rest of us the angle was there and worth standing at — rather than to ownership of what was visible from it.

What Gets Left When the Method Is Automated

Here is where the threads close.

The methodological layer automates. Not completely, not immediately, not uniformly across all fields — the chemist with a wet lab is not in the same position as the climate scientist with a compute cluster. But the direction is clear and the acceleration is real. The things that made methodological discipline the primary credential of scientific work are losing their scarcity.

What remains is the vantage point.

Your biography. Your interdisciplinary range. The domains you traversed on the way to the question you are now asking, and what each of them left in you as a lens. Your phenomenological sensitivity — the trained capacity to notice what you are noticing, to hold an experience at arm's length long enough to see its structure. Your willingness to follow a thread into territory where you do not yet have a method, because you have learned to trust the thread before you trust the method.

Those of you who have been here since the beginning have already watched this happen in real time. The phrase Minimum Viable Scaffolding, from the second chapter, is a nod to the startup world — the world of the minimum viable product, which I am part of. I did not reach for that phrase to be clever. I reached for it because that is a domain I have traversed, and it left a lens in me: a particular instinct about how the smallest workable version of a thing is enough to start learning from. Carrying that instinct across the border into epistemology — letting a startup concept describe how understanding gets built — is exactly the move this whole section is about. The vantage point is not an abstraction I am asking you to imagine. It is the reason that chapter is called what it is called.

Biography, range, the lenses left by the domains you have crossed, the sensitivity to notice what you are noticing — these are not decorative qualities. In the methodological era, they were treated as decorative — the personality of the researcher, interesting but irrelevant to the science. What the fracture required was their suppression. The good scientist was supposed to be interchangeable, the method reproducible by anyone trained to execute it. The vantage point was variance to be controlled out, not signal to be cultivated.

That framing is inverting. The method can increasingly be executed by anyone — or anything. The vantage point cannot. It is constituted by everything the researcher has lived, read, built, crossed, failed at, and carried forward. It is the only part that remains irreducibly human.

You can already see this beginning, if you know where to look. Talk to professors assembling a lab and you will hear it: they recruit PhD students not only for technical competence but for the angle they bring — the physicist who wandered into biology, the engineer with a background in music, the candidate whose unusual path means they will see the problem from a side no one else in the group can. That instinct has always existed quietly. What is changing is that it is no longer a tiebreaker after the methodological credentials are checked. It is becoming the point.

What happens to the PhD defense in a world like this? It cannot rest entirely on the methods section — not because rigor stops mattering, but because rigor stops being scarce. If any sufficiently capable system can run the regression, then we ran this and found a correlation of 0.5 is no longer the interesting part of the room. The question worth defending becomes the one a system cannot answer for you: why were you the one to find this thing? What did your particular vantage — the fields you crossed, the assumptions you were positioned to doubt — let you see that the standard frame concealed? What did it necessarily blind you to? And how do you know the axes you chose revealed the structure rather than imposing it?

That is a question about vantage point. It is also, inescapably, a question about the person.

The Epistemological Monopoly and What Breaks It

The logical positivists did not defeat the other epistemologies. They starved them.

Phenomenology — the study of experience as experience, prior to the imposition of theoretical categories. Analogical reasoning — the ancient practice of understanding one domain by mapping its structure onto another. Narrative knowledge — the kind of understanding that accumulates through story and cannot be fully translated into propositional form. Wisdom traditions, accumulated pattern recognition, tacit knowledge that resists formalization. All of it got pushed to the margins because it could not be credentialed, could not be funded, could not be published in journals that required verifiable empirical claims.

The monopoly produced what monopolies produce. Efficiency in the production of a particular kind of output. Atrophy in everything the monopoly excluded.

Consider narrative knowledge — the kind of understanding that accumulates through story and cannot be fully translated into propositional form. A scientific paper on this blog, with the same set of ideas would have been perhaps three pages: an abstract, a few citations, a brief argument, a bullet-pointed conclusion. Efficient. Credentialed. Forgettable. This chapter is long. It has moved through history and physics and painting and the geometry of simultaneous discovery and the ego of a man who named a coordinate system after himself. If you have arrived here and feel like you understand something you did not understand at the beginning — not just know it, but understand it — that is narrative knowledge doing what it does. The structure of the argument did not arrive as a list. It arrived as an accumulation, one layer landing on the previous one, until the shape of the thing became visible. You could not have gotten there faster. The length was the mechanism.

And if you feel that understanding alongside a faint discomfort — a sense that this is not quite how serious intellectual work is supposed to look — that discomfort is also information. It is the positivist monopoly expressing itself from inside you. You absorbed it somewhere, in a classroom or a journal or a conference where slides had three bullet points and anything longer was self-indulgent. The discomfort is not evidence that the chapter is wrong. It is evidence of how deep the other lens goes.

AI breaks the monopoly not by defeating the positivists philosophically — that was done fifty years ago and had no institutional effect — but by automating the very output the monopoly was organized to produce. When the machine can run the regression, conduct the analysis, execute the method end to end, the credential for executing the method becomes hollow. The question of what else might count as knowing becomes urgent in a way it has not been since before the fracture.

This is the opening. Not just for science, but for the reunion of ways of knowing that the methodological era drove apart. The artist and the scientist were not always in separate buildings. The natural philosopher was not divided against herself. The reintegration is not nostalgia. It is the correction of an error that had a good justification until the justification dissolved.

The series this chapter belongs to is, among other things, an argument for that reintegration — made from the inside of a scientific career that took a particular route, through languages and cities and fields and failures, and arrived at questions that cannot be answered by method alone. The vantage point that asks them is not incidental to the questions. It is the reason the questions are visible at all.

The natural philosopher is not a thing of the past. It is what the present is slowly making room for again.

A Note on the Mountain

Two painters on a hillside, both looking at the same mountain from slightly different positions, will produce different paintings. Both true. Neither complete. The mountain is the same mountain. The paintings are different paintings. There is no contradiction.

Fighting over who painted the mountain first misses the point entirely.

The mountain was there before either of them arrived. It will be there after both of them are gone. What they leave behind is not the mountain. It is an account of how the mountain looked from where they stood, on that day, in that light, through the accumulated weight of everything they had learned to see.

The name you put on it is the smallest thing about it.

Written in Honolulu, June 2026. With thanks to M.A. for spurring the question that sent me looking deeper into this, in our discussions while touring Zermatt in 2025, the village beneath the Matterhorn — the mountain on the Toblerone.

If this resonated — and especially if you are a scientist who has felt the gap between the method you were trained in and the questions that actually matter — there is more coming. The next series, First Paddle Out, picks up where this one reaches. If you would like it to find you when it does, you can leave your email below.