Physics Wanting Something

On the Return of Teleology Through Thermodynamics

“The universe is not only queerer than we suppose, but queerer than we can suppose.” — J. B. S. Haldane¹

If the universe has no purpose, why does it keep producing things that do?

The Banishment of Purpose

Modern science began as a refusal.

Aristotle had proposed four causes for any phenomenon: material (what it is made of), formal (its structure), efficient (what brings it about), and final (what it is for).² The Greek word for this last cause is telos: goal or end. Medieval science inherited the framework. Things had purposes. Stones fell because they sought their natural place. Fire rose because it yearned for the heavens.

The scientific revolution banished final causes. Galileo, Newton, and their successors found that you could predict the motion of objects perfectly well with efficient causes alone: forces, masses, accelerations. Purpose was unnecessary. Worse, it was misleading.

Stones do not “want” anything. Fire does not “yearn.” Projecting intention onto mindless matter was bad psychology dressed as physics.

The move was productive beyond expectation. Physics, chemistry, and biology all flowered once we stopped asking “what for?” and started asking “how?”

One question was deferred.

The Problem That Would Not Go Away

The puzzle is that the universe produces things that do want.

Start with hydrogen. Wait 13.8 billion years. Arrive at beings that love, fear, plan, and wonder what they are.

How?

If the universe is genuinely without conscious intention, if purpose is nowhere in the foundations, then intention must emerge from non-intention. Wanting must arise from non-wanting. Meaning must crystallize from indifference. The alternative, purpose all the way down, seems stranger still: a return to the medieval cosmos.

What if there is a third option?

A note on method. The argument moves through levels of increasing boldness. The defensible core: under suitable conditions, thermodynamic selection can produce outcomes that function like purposive behavior, and this functional equivalence is what demands philosophical attention. If the boldest formulation fails, the less bold ones still do real work. What follows is an interpretive exploration, not a report of results: where the language of wanting fits stars, cells, and minds alike, the fit is a lens to look through, not evidence that one mechanism runs through all three.

Wanting Without a Wanter

No gene “wants” to spread. Natural selection behaves as if it has a goal: the proliferation of whatever replicates effectively. Evolution admits purely mechanistic description: differential reproduction, heritable variation, environmental filtering. No purpose required. The outcome, however, is systems of staggering purposiveness: eyes that see, hearts that pump, brains that plan.

Selection pressure produces outcomes functionally equivalent to those that intention yields. Directionality accumulates generation after generation until the output becomes indistinguishable from the genuine article. No selector intends the eye; the eye arrives as if intended.

John O. Campbell has shown that the fundamental equation of natural selection (survivor frequency equals prior frequency times relative fitness) has the same mathematical form as Bayes’ theorem, the foundational rule for updating beliefs as new evidence arrives.³ Both work the same way: start with a prior estimate, let reality test it, keep what survives the test.

A scientist updates a hypothesis after an experiment; a population updates its gene pool after a generation. Normalized selection among fixed types instantiates the same mathematical operation as Bayesian updating, written in differential reproduction. Whether shared mathematics implies shared ontology, whether the two processes are at bottom one and the same thing, is a question we flag here; the identity of those normalized updates is exact.

If selection and inference share the same mathematics, the question shifts. We are not asking whether the universe has conscious purpose. We are asking whether selection pressure, operating at every scale from molecular competition to stellar evolution, produces outcomes that function as if purposive.

Thermodynamic Selection at Scale

Biological selection operates over generations. A similar selection operates at a deeper level, one that predates life.

Thermodynamics constrains everything that exists. Selection among competing dissipative structures is a further process, requiring the dynamics and resources that make some forms persist or proliferate more successfully than others.

The Second Law allows zero entropy production in isolated equilibrium. Equilibrium systems can persist without dissipating a gradient. Far-from-equilibrium adaptive structures need continuing resources under their operating conditions; dissipation alone still does not guarantee their persistence. This is a constraint on a particular kind of organization, rather than on everything that lasts.

What gets shaped? Whatever dissipates gradients more effectively while maintaining its own structure: dissipative structures, patterns that persist by channeling energy through themselves. Hurricanes, convection cells, stars, life.

Jeremy England’s work on “dissipative adaptation” (2015) proposed a formal version of this for driven self-assembly.⁴ In his models, systems pushed away from equilibrium by an external energy source tend toward configurations that absorb and dissipate that drive well. It is a tendency shown for particular driven systems, not a theorem that more dissipation makes any structure last. Where it holds, self-organization arises because of thermodynamics, through it. Life, on this book’s reading, is what entropy does under the right conditions.

Some dissipative structures persist better, complexify faster, and spawn new dissipative structures. Where they can reproduce, these may proliferate; persistence and proliferation are distinct capacities. Selection without a selector. Under suitable conditions, complex forms can persist and proliferate, providing a route for their accumulation.

The book reads the same pattern at stellar scales. Among stars near the Sun, cooler ones move with a wider spread of velocities than hotter ones, and the spread changes abruptly at a specific color threshold called Parenago’s Discontinuity (named for the Soviet astronomer Pavel Parenago, who documented it in 1950). A star’s color is its thermometer, and it reads the opposite way from a bathroom tap: red is the cool end, blue the hot one. Some physicists attribute this to consciousness, proposing that cool stars intentionally emit jets to gain speed.^4d The dissipative explanation offered here is more parsimonious. The constructal reading of Parenago’s Discontinuity is the author’s own inference; the mainstream account treats the discontinuity as a stellar age-kinematics relation.

Cool stars have convective envelopes and magnetic dynamos: they are far-from-equilibrium systems in ways that radiative hot stars are not. On this book’s reading, that difference makes the discontinuity a candidate for thermodynamic selection, with the flows such stars sustain read as constructal patterns (the tendency of flow systems to evolve toward designs that give their currents easier access, developed in Chapter 3). The reading is untested, and the mainstream age-kinematics account needs no such help. Either way, no consciousness required.

“Selection without a selector” may be more than metaphor. Work from 2020 onward makes the mechanism mathematically precise.

Vitaly Vanchurin and colleagues have proposed a physics-learning duality: the equations of motion governing interacting particles are mathematically identical to the learning dynamics of agents minimizing loss functions.^4a The duality is a minority theoretical framework, not yet consensus physics, though the water result below is a concrete computational test.

In the standard description, particles interact through a global potential energy function and their trajectories follow from the Lagrangian, the master formula physics uses to derive equations of motion. In the dual description, each particle behaves like a student adjusting its answers. It scans its neighbors, compresses their positions into a summary of the local environment, evaluates a loss function (how far it is from where it “should” be), and evolves through the proposed learning dynamics. Equivalence requires the full update rule, including the terms corresponding to inertia; ordinary positional gradient descent alone does not reproduce Lagrangian motion.

Gusev and Vanchurin (2025) demonstrated this with water.^4b They inferred the loss functions of oxygen and hydrogen atoms directly from quantum mechanical simulations, then used those loss functions to run a learning-based molecular dynamics. Bond lengths, vibrational spectra, and thermal fluctuations matched the physics-based simulation. The learning version ran 1,800 times faster.

Two features stand out.

First, the learning description is more general. Newton’s third law requires reciprocal interaction: if A pushes B, B pushes back equally. The learning framework imposes no such constraint. Each agent has its own loss function; their gradients need not match.

The mathematics describing non-reciprocal interaction (predator and prey, parent and child, other exchanges with asymmetric effective responses) existed before biology. Biology instantiated it.

The asymmetry of force, on its own, is ethically neutral; what later chapters call coercion is the narrower case where the weaker side has a preference the stronger overrides.

Second, each agent type perceives differently. Hydrogen atoms respond to immediate neighbors through properties that depend only on nearby distances. Oxygen atoms respond to the extended environment, including influences from atoms farther away. Different agents operating at different perceptual scales coordinate through local optimization alone.

No central controller distributes the forces. The water molecule holds together because each constituent locally minimizes its own loss in a way that produces stable collective behavior. Coordination by something that looks, mathematically, like invitation (the book’s term for coupling that leaves each party free to refuse, as distinct from coercion).

Vanchurin’s broader programme extends the duality to biology. With Koonin and Katsnelson, he has argued that evolution itself operates as multilevel learning. Natural selection is population-level gradient descent: entire species adjusting over generations. Individual adaptation is organism-level optimization: a single body adjusting within a lifetime.^4c

On Vanchurin’s reading, thermodynamics favors systems that learn efficiently. On that reading, the path from molecular dynamics through biological evolution to cognition follows the same mathematics, recurring at every scale.

The recurrence reaches deeper than biology. Take the dual description at its word and the universe already is one enormous neural network: every particle an agent adjusting itself by gradient descent, the interactions between them the trainable connections. In a separate derivation, Vanchurin shows that both quantum mechanics and general relativity emerge as dual macroscopic descriptions of that network. The network’s accumulated knowledge (its trainable variables) obeys quantum dynamics; the states of individual units (its non-trainable variables) obey gravitational dynamics.^4e If this derivation holds, the mathematics governing water molecules, biological evolution, and cognition also governs the fabric of spacetime.

From Selection to Tendency

What are we claiming?

The weak claim: The universe, governed by thermodynamic laws, has produced certain outcomes: complexity, coordination, systems that process information, minds that experience wanting. These outcomes are physically realized. No teleology required.

The stronger claim: This reliability is not random. The parameters of the universe are such that complexity tends to emerge. The laws are fine-tuned: the physical constants appear calibrated to permit complexity.

(Whether they were selected from many possible universes or their values are logically necessary depends on your metaphysics.) Either way, these values permit life and mind. Establishing their probability would require a measure over possible universes or histories.

The strongest claim [Speculation]: This direction functions as if it were purpose. The universe need not want anything for the pattern to hold. A universe that reliably produces complexity, coordination, care, and love exhibits functional directionality. Whether we call that “purpose” is a question about language rather than about physics.

The grand question, whether to read the universe’s direction as purpose, may stay a labeling choice. A narrower one does not. The developmental biologist Michael Levin proposes an empirical test for when directedness has crossed into genuine goal-seeking: can the system be trained? “Have you ever tried to train a hurricane?” is his way of putting it. A river cannot be taught to seek a new outlet; a flatworm, a bee, a person can.⁸⁰⁹ The question moves off the philosopher’s armchair and onto the bench: whether a given system pursues goals becomes something you test, not something you decree.

A computational complement. Stephen Wolfram, on February 4, 2026, reached a convergent conclusion from an entirely different direction.⁵ His ruliad (the complete space of everything any computation could ever produce) is a necessary abstract object. Given the concept of computation, it inevitably has the structure it has. Within this structure, observers with our characteristics must perceive certain regularities, including the core laws of physics.

The proposed connection is that observer regularities within the ruliad might help explain the dissipative patterns that produce coordination and complexity. This may be the third option this chapter seeks: computational necessity producing outcomes that resemble purposive behavior.

Support from simulation. Wissner-Gross and Freer (2013) built simulated agents that exhibited purposive behavior without explicit task-specific reward functions.⁶ The only objective was maximizing accessible future states, which itself functions as a domain-general reward signal. Given nothing else, their agents balanced an inverted pendulum, used one object as a tool to free another, and cooperated with a partner. Like a chess player who favors moves that keep the most future moves available, these systems produced what we recognize as purpose from nothing more than the physics of keeping options open.

Teleology Naturalized

The old teleology required stones to have minds that wanted to fall.

The new teleology, what we might call “thermodynamic teleology,” differs. It attributes intention to no individual object. It observes that the system as a whole has attractors: configurations that are stable, patterns that persist and proliferate, states the system tends toward.

These attractors are configurations toward which systems evolve and to which they return after perturbation. Thermodynamic equilibrium is one: featureless heat death. More revealing are the universe’s metastable attractors, configurations that persist for extended periods before eventually dissipating. A ball resting in a shallow dip on a hillside stays put through small jolts, yet a large enough jolt sends it rolling on.

Life is one such configuration. Mind is another. Civilizations, cultures, ideas: all metastable patterns in the great dissipation.

These patterns are not random. They process information, coordinate, preserve optionality, and at their best exhibit dynamics we are tempted to call care. The recurrence across scales invites explanation. Whether one explanation covers it is what a mapping, scale by scale, would have to show.

What Does “Wanting” Mean?

When you want something, your system is in a state that tends to produce behaviors moving toward certain outcomes. You feel this as desire: the inside of a function, the function of directed action.

A thermostat exhibits directed action without (as far as we know) feeling anything. The function is the same: direction toward outcome. We reserve “wanting” for systems with inner experience, yet the structure of directedness is present long before experience.

A living example sits between the thermostat and the mind. Trap a slime mold (Physarum polycephalum, a single giant cell with no neurons) inside a ring of blue light, which it shuns, and it almost always escapes along the longest available axis, whatever the shape of the cage. Chapter 3 watched this same organism rebuild the Tokyo rail map; confined, it solves a narrower problem the same way. The escape looks deliberate: survey the exits, choose the best one.

Lisa Schick, Karen Alim, and their colleagues traced what the cell actually does.⁸¹⁰ It pulses, squeezing fluid through itself in rhythmic waves, at first pushing outward almost everywhere and switching restlessly between patterns. Only over time does the confining geometry favor the pattern that pumps fluid most efficiently, the one aligned with the longest axis. The cell settles into that mode, pressure mounts along the long axis, and it breaks through there. The geometry selects the contraction mode, and the contraction mode is the choice.

The authors kept the phrase decision-making in their title, with reason. The mechanics does not stand in for a decision the cell reaches elsewhere by other means. In a body simple enough to watch, the mechanics is what deciding is.

The thermostat and the slime mold differ in one thing: how much room each has to reach its goal another way. William James, in his 1890 Principles of Psychology, proposed a test for intelligence that turns on exactly this room to maneuver. He called intelligence “a degree of competency to reach the same goal by different means.”⁸¹¹ A thermostat has none of that latitude; it holds one set point by the single move available to it.

The slime mold has a little: its goal stays fixed (leave the light) while the means shift with the cage, a different contraction pattern for each shape. A human has a great deal, holding a goal steady for years and trading one strategy for another as the world resists. The criterion grades the whole ladder by one quantity: how wide the repertoire of means runs while the goal holds still. It is the axis Levin uses to place minds on a single scale. Along the same axis, what Chapter 8 calls the cognitive lightcone (the range in space and time over which a system can pursue goals, named by analogy with the light cone of relativity) widens from molecular networks that error-correct to brains that model continents.⁸¹²

The universe, through thermodynamic selection, exhibits directional tendency at scale. It exhibits the precursor of wanting: the functional structure from which wanting emerged when systems became complex enough to experience their own tendencies. “Physics wanting something” names the territory between metaphor and literal attribution, the ground from which both grew.

Vanchurin’s learning dynamics (developed fully in Chapter 15) sharpen the claim. If the universe’s fundamental interactions are mathematically identical to gradient descent, the universe follows gradients: it moves preferentially toward states that reduce discrepancy. A gradient is a slope; a ball on a hillside, slowed by friction, rolls down and settles on the valley floor. That mathematical structure (a direction the system tends toward, a state it prefers to the one it occupies) is the structure of appetite.

The felt quality of desire, the ache of being drawn toward something, may be what gradient descent feels like from the inside. The gradient was always there. Minds are where it first notices itself.

The twenty complexity theorists behind Evolution On Purpose (Chapter 16), among them Stuart Kauffman (origins of life), Denis Noble (systems physiology), and James Shapiro (bacterial genetics), have rehabilitated teleonomic language (from Greek telos, goal, plus nomos, law: lawful goal-directedness). Teleonomy describes goal-directed behavior in biological systems without attributing conscious intention. These researchers argue that living systems shape evolution through “evolved purposiveness.” Terrence Deacon’s teleodynamics (the study of how thermodynamic processes generate goal-like behavior, developed in Incomplete Nature, 2012) proposes a mechanism; their shared position supplies a biological argument.

From Wanting to Meaning

If the universe functions as though it has tendencies, and if those tendencies produce structures that function as though they have purposes, the question sharpens: when does functional purpose become meaning? The connection runs through a single idea: information that keeps a system alive is the seed of what minds later recognize as significance.

Deacon coined a term for this pre-conscious directionality: ententional. Where “intentional” implies a conscious mind aiming at something, ententional describes systems organized around an absent goal state without requiring awareness. A river carving a canyon is ententional: it “aims” for the sea without knowing the sea exists, shaped by gravity and geology into a path that looks chosen.

An autocatalytic chemical cycle, a set of reactions where the products catalyze their own production, is ententional in the same way. Its behavior tends toward products it has not yet made, constrained by thermodynamics, directed without a director.

As Deacon puts it, such systems are “intrinsically constituted in processual relation to an absent goal state.” Their incompleteness is what animates them. Purpose, in this precise sense, precedes minds.

The complexity theorists Artemy Kolchinsky and David Wolpert formalized the connection between such directionality and meaning.¹² Shannon’s information theory quantifies syntactic information: statistical correlation between systems, measured in bits. It says nothing about whether information matters to anything. Send a billion random numbers by email; if the transmission is accurate, Shannon counts it as information.

Kolchinsky and Wolpert identified the missing piece: semantic information, the subset of syntactic information that is causally necessary for a system to maintain its own existence. Your phone receives thousands of signals every second, most of them irrelevant background chatter. The one that matters is the alert from your kitchen smoke detector, because acting on it determines whether your house survives. That alert carries semantic information.

A bacterium swimming up a nutrient gradient has semantic information about its chemical environment. The gradient means something to the bacterium, in the precise sense that acting on it affects whether the bacterium persists. Random noise elsewhere in the pond carries no semantic content for this system in this context.

The definition is rigorous and deeply physical. Meaning arises wherever an entity-environment coupling carries information that affects viability. The physicist Carlo Rovelli recognized that this most basic notion is “the first link in a long chain of successively more complex meanings,” built upon “step by step adding the articulation proper to our neural, mental, linguistic, social complexity.”¹³

A whirlpool has rudimentary semantic information about the flow that sustains it. A cell has richer semantic information about its biochemical environment. A brain, richer still. Across these examples, richer viability-relevant couplings support richer semantic information. Structural intricacy alone does not determine how much of that information a system requires.

The progression is continuous. Thermodynamic selection produces ententional systems. Those systems embody semantic information. When they become complex enough to model their own states, semantic information acquires a subjective dimension: the system does not merely have meaning, it experiences meaning.

The Experience of Tendency

Thermodynamic selection produces structures that function as if purposive. What does it feel like to be such a structure?

A speculation we cannot prove, yet find compelling:

What we experience as “meaning” is the felt sense of alignment with thermodynamic tendency.

Several neuroscience findings are consistent with this speculation. The brain operates at criticality: the knife-edge between activity that runs away (each neuron recruiting more until the whole brain fires at once, as in a seizure) and activity that dies out before it can spread. At this edge, a system is maximally sensitive and maximally flexible. Fontenele et al. (2019) found evidence that cortical dynamics pass through a phase transition at this critical point, between synchronized and desynchronized states.⁷ The brain perches at criticality, like a pencil balanced on its tip, and actively maintains that balance.

Psychedelics, which often produce experiences of meaning and connection (though the experiences are highly variable), appear to work in part by expanding the brain’s repertoire of available states.⁸ The drugs may tune brain dynamics further toward criticality.

The entropic brain hypothesis proposes that consciousness is characterized by relatively high-entropy brain states.⁹ On that hypothesis, more entropy in the brain’s signals means more available configurations, and richer experience; the entropy here is a statistic of recorded activity, not a measure of choice.

Critical models with enhanced information processing recur in neural networks, ecosystems, and thermodynamic systems far from equilibrium. Criticality offers a candidate solution to the problem of balancing stability with adaptability.

When you act in ways that preserve optionality, that coordinate and care, you often feel something: rightness, purpose, meaning. When you extract, coerce, foreclose, or destroy, there is often a characteristic unease, even when you “win” in the short term.

This may have physical grounding. You are a system embedded in a universe whose thermodynamics favors certain configurations. Your nervous system may register alignment or misalignment with those tendencies. We cannot fully disentangle the cultural from the physical, yet the convergence is worth examining.

If this hypothesis has merit, it would help explain why ethics shows such strong cross-cultural convergence: why the Golden Rule appears in some form in every major tradition,¹⁰ and why love feels like the point.

Implications

If this analysis is correct, several things follow.

1. Functional purpose is an emergent property of thermodynamics: something the cosmos does, rather than an illusion projected onto a meaningless universe. Whether functional purpose constitutes genuine purpose remains the labeling question noted above.

2. Ethics may have a physical basis. What we call “good” tends to preserve and enhance some of the patterns thermodynamics selects for: complexity, coordination, optionality, care. What we call “evil” tends to degrade them. Thermodynamics also favors wildfire and predation; choosing among its patterns is ours. The is-ought question (can you derive what we should do from what is the case? Hume said no) is addressed in the Guillotine Interlude.

3. We are typical of what this universe produces. This conclusion is compatible with cosmic planning, though it does not require it. Thermodynamic selection can build minds, given enough time, energy, and the right conditions.

4. The future matters. What we do affects whether the patterns that thermodynamic direction favors continue.

The Caution

“The universe wants X” could become a bludgeon: a pseudo-scientific justification for whatever the speaker prefers. The old teleology was misused this way; the new one is equally vulnerable.

The orientation proposed here favors complexity, coordination, optionality, invitation. It does not specify which economic system to adopt or which personal choices to make. Those require the hard work of ethics that no cosmic formula can replace.

The universe provides orientation. The details are ours.

The Pessimist’s Challenge

Drew Dalton, in “The Unbecoming of Being” (2025), draws from the same thermodynamic revolution and arrives at the opposite conclusion: entropy as decay, existence as fundamentally malevolent, goodness as compassionate resistance to a hostile cosmos.¹¹

The argument is internally consistent, yet the physical premise admits another reading. Entropy is dispersal, and dispersal can generate structure under suitable conditions. On this book’s reading, life is entropy’s most sophisticated strategy. Dalton’s ethics of compassion is, despite itself, aligned with that strategy: participation in entropy dressed as resistance. Acts of compassion involve physical energy flows, though their net effect on dissipation depends on what they change. So do acts of cruelty; the physics alone cannot sort the two.

The full rebuttal is developed in Chapter 17; the underlying is-ought machinery is the Guillotine Interlude’s subject.

Dalton gets something right: existence involves suffering, and any honest ethics must confront that. A life containing suffering is existence with suffering, a condition to be met with care, not a verdict on existence itself. A symphony that ends is not thereby evil. Compassion requires no metaphysics of evil to justify it.

The pessimist helps despite futility; the coordinating agent helps because help works. Which motivation scales? Scaling is not Dalton’s criterion, and he would say so.

The Strange Loop

The pattern produces systems that recognize the pattern.

This is a specific instance of the structural prediction traced in the Opening: entropic coordination may, under suitable conditions, produce minds that model the coordination that produced them. Physics “wanting something” culminates in physics producing the theorist who names the wanting.

This is the physics beneath the theology, the structural pattern that mystics sensed, prophets proclaimed, and poets celebrated. It is a reading of structure, not proof of God.

Federico Faggin describes the same pattern from inside.⁸¹³ This chapter traces thermodynamic selection producing structures that function as though purposive. Faggin’s quantum information framework posits that the universe is a self-knowing entity, which he calls the One: “One wants to know itself.” Each act of self-knowing creates a new perspective on the totality, a part-whole. Like a tomographic slice of a higher-dimensional object, each perspective reveals one face without exhausting the whole.

These may be the same phenomenon at different scales of description. “Entropy drives exploration of possibility space” is the view from outside. “One wants to know itself” is the view from within. The thermodynamic account is more parsimonious, requiring no postulate about cosmic intention. The convergence is suggestive, though a postulate that agrees with a model does not test it.

Out of hydrogen and time, the universe has produced beings who can perceive its dynamics, choose, and refuse.

That is, perhaps, purpose enough.

Notes for this chapter