What process philosophy actually claims
Most metaphysics starts from things. A chair, a coin, a person: each is treated as a substance, a bearer of properties that persists while its properties change. The coin is scratched, tarnished, spent; through all of it, something continues, and philosophers since Aristotle have argued about what that something is. Process philosophy rejects the starting point. It holds that the basic constituents of reality are events, not things — that what exists, fundamentally, are happenings, and that the things we call things are patterns stitched out of happenings, not the other way round.
Alfred North Whitehead's version, laid out in Process and Reality (1929), builds this from a unit he calls the "actual occasion": a momentary event that takes in the state of the world up to that instant, does something with it, and then perishes — but perishes into data, material that later occasions inherit and build on. Nothing simply vanishes. Nothing simply persists either. Each occasion is finished and gone almost as soon as it occurs; what looks like continuity is really an unbroken relay of inheritance, occasion handing to occasion, each new event conditioned by what came before it and available to what comes after.
On this view a rock is not a substance that persists through time while suffering minor changes. A rock is a very stable route through an enormous number of occasions — trillions of momentary events so alike in succession that nothing about the rock seems to move. Substance metaphysics treats change as something that happens to an enduring thing. Process metaphysics inverts this: endurance is what happens when change is arranged just so. Persistence is not the default state of the world with disruption layered on top. Persistence is an achievement, and achievements require maintenance. Stop the maintenance and the rock does not become a different rock. It stops existing as that route at all — slowly, in geological time, but the logic is the same as for a flame or a cell.
Where it came from
Whitehead did not arrive at this from armchair speculation about coins and rocks. He came to metaphysics from mathematics, having spent two decades co-authoring Principia Mathematica with Bertrand Russell, and he was watching physics dismantle the material particle as a stable basic category. Relativity made simultaneity relative to frame; quantum theory made the "particle" a probability distribution over events rather than a little billiard ball with a definite location. The substance ontology that had served Newtonian mechanics well was visibly failing to describe what the new physics was finding. Whitehead's response, worked out through the 1920s and published as Process and Reality in 1929, was to rebuild ontology from the event upward: take the actual occasion as basic, and treat enduring objects as high-level patterns — "societies" of occasions, in his term — sustained by continuous re-inheritance.
He was not without ancestry. Heraclitus had said as much, more tersely, twenty-four centuries earlier: you cannot step in the same river twice. Henri Bergson's notion of durée, real time as indivisible flow rather than a sequence of static instants, fed directly into Whitehead's picture. And the idea outlived him in a different discipline: Ilya Prigogine's work on non-equilibrium thermodynamics, decades later, gave process ontology an empirical home. Prigogine showed that certain structures — convection cells, chemical oscillators, living cells — exist only far from equilibrium, sustained by a continuous throughput of energy. Cut the throughput and the structure does not decay gently. It disappears. That is Whitehead's actual occasion, translated into watts.
The turn: intake as a metaphysical commitment before an engineering one
Here is the connection, and it is worth taking slowly, because it is easy to overstate and easy to dismiss.
Every model of the world has to decide how it takes the world in. A Large Language Model ingests a corpus: text fixed at a cutoff, a body of statements about how things were. It is substance metaphysics rendered as a data format — the world flattened into properties that were true, stated in the past tense, with all the transition that produced those properties already collapsed out. Ask it about a flame and it will tell you accurately that flames are hot, that they consume fuel, that they flicker in a draught. It has no access to burning, because burning is not a property; burning is a rate, and rates are exactly what a corpus discards on the way to becoming a corpus.
A Large World Model does better, for a while. It senses a scene as the scene unfolds — a room, a manipulation task, a driving segment — and for the duration of that episode it genuinely observes process: things changing because of other things, in real time, with real causal texture. This is not a small improvement. It is the difference between reading a transcript of a conversation and being in the room for it. But the episode ends. The camera stops, the scene closes, and nothing inherits what was observed into what comes next. Each episode is an actual occasion that perishes without contributing to a successor. The Large World Model has process without lineage — becoming, witnessed and then thrown away.
A Large Universe Model, on the argument this page is making, is the first intake regime shaped like the world Whitehead described. Streams that do not stop. Beliefs that carry forward as revisable inheritances from prior belief states, tagged with where they came from and how much confidence they still deserve. When two streams disagree, the belief updates rather than being silently overwritten. This is not a bigger corpus and not a longer episode. It is a structural claim: that observation should itself be a process, continuous with the process it is observing, rather than a series of freezes taken of something that does not hold still.
The instances that make this concrete are mundane, which is the point. A candle flame holds a stable shape for hours while no particular parcel of wax vapour persists inside it for more than milliseconds; photograph it and you have captured a shape whose entire causal content — the consumption rate, the draught, the wick's carbon accretion — is invisible in the image. A human body turns over roughly 330 billion cells a day; a cell atlas built as an inventory of types tells you what is present, not whether the turnover rate is heading somewhere bad, and it is the rate — a raised reticulocyte count, an accelerated crypt mitotic figure — that a clinician actually needs. The Amazon basin's status as forest or savanna is not a fixed classification but a balance between moisture recycling and dry-season length; a single-season satellite composite can show identical canopy cover for a stable forest and for one three dry seasons from collapse. In each case the frozen sample and the live rate can coincide in every visible respect while disagreeing completely about the future.
The objections, taken seriously
Whitehead's system is speculative philosophy with almost no working uptake in physics or chemistry. Invoking a 1929 metaphysical scheme to justify claims about data pipelines is decoration dressed as argument.
This lands, and it should narrow the claim. Process and Reality is not load-bearing in any laboratory; physics gets on fine with fields and state vectors, chemistry with molecules, and nobody needs actual occasions to balance an equation. But the argument does not require Whitehead's full metaphysical apparatus — only its negative half, that treating persistence as free is a modelling error. That half does not depend on Whitehead at all. Prigogine's dissipative structures establish it independently, inside physics: certain forms exist only under continuous energy flux, full stop. Whitehead supplies the vocabulary. The result belongs to thermodynamics. What follows is that cost and latency, not ontology, decide when continuous intake is worth paying for — but ontology decides whether it is describing something real when you do pay for it.
Any sensor samples at some frequency. A stream is a series of snapshots taken fast, not process. So a Large Universe Model differs from earlier generations by degree, not by category.
Also correct on its premise: all measurement is discrete, always was, always will be. But the distinction on offer is not sampling density. It is whether samples are stitched into a single inheriting lineage or left as an unconnected set. A corpus is a set. A stream whose states are each conditioned on the last, carrying provenance and revising under contradiction, is a sequence in Whitehead's specific sense — inheritance. A camera running at thirty hertz whose frames are classified independently of one another has no process representation whatever its frame rate. The same camera, feeding a belief that persists and updates across frames, does. The line is architectural. Frequency is beside it.
If nothing truly endures, then a retained belief is just as much a fiction of substance as a frozen fact — the Large Universe Model smuggles substance back in under the name of state.
This is the objection that should worry the thesis most, and it does not fully dissolve. Any persistent store reifies something. Whitehead's own answer is that endurance is legitimate precisely when it is maintained rather than assumed: a "society" of occasions held together by continuous re-inheritance, as opposed to a fact simply asserted to be still true. The practical test is whether the store degrades when intake stops. A belief that must be re-derived and re-timestamped against arriving evidence, and that decays in confidence when its supporting streams go quiet, is maintained persistence. A corpus fact, unmaintained by construction, is assumed persistence. They can be identical bit patterns on disk. Their ontological status is opposite, and the difference only shows up when the world moves and one of them notices.
The misreading to disown
The weak version of this argument says that everything is flux, therefore all fixed models are worthless, and only real-time systems deserve to be taken seriously. This is false and it defeats itself, because a claim that nothing is stable enough to model would also apply to the sentence making the claim. Whitehead's actual position is narrower and less dramatic: stability is produced, not given, and a snapshot cannot see the producing. That means frozen intake is entirely adequate wherever maintenance cost is low and slow relative to the question being asked — most of chemistry, most of engineering geometry, a great deal of settled history. It fails specifically where the object under study is being held together against a gradient: a flame, a fever, a fish stock, a bridge under load, a grid under frequency stress. The concept does not license contempt for corpora. It licenses suspicion of corpora used to answer questions about rate.
What this does and does not establish
Process philosophy does not prove that continuous ingestion is achievable, affordable, or wise in any given case. It does not show that a Large Universe Model, as an argued category, currently exists as working technology — it does not. It does not settle the many objections raised above so completely that nothing remains contestable; the second and third objections leave real seams. What it does establish is narrower and, I think, sufficient: that the intake axis running from frozen corpus to bounded scene to standing stream is not an arbitrary sequence of engineering upgrades stacked for convenience. It tracks a distinction that metaphysics worked out a century ago on entirely separate grounds — between things assumed to persist and things maintained in persisting — and it explains, without needing anything mystical, why the third position has no obvious fourth rung above it. You can always add sensors, span, fidelity, trust in the belief. You cannot add a fifth tense. That the axis has a top is not a claim about how clever machines can become. It is a claim about how many ways there are to stand in relation to a world that does not hold still.