Selection.

Why are there brains?

The standard answer is Darwinian: brains evolved because they conferred reproductive advantage. Better cognition meant better survival meant more offspring.

This answer is correct as far as it goes. It also leaves a deeper question unanswered. Why did the universe produce a biosphere capable of evolving anything as complex as a brain in the first place? Of all the chemistry that could have unfolded on a wet rock, why did the chemistry that selected for cognition appear?

Jeremy England, then at MIT, published an answer in 2013 that most readers have never encountered. It is the strongest thermodynamic case for the inevitability of life and, by extension, of brains.

England’s paper, titled “Statistical Physics of Self-Replication,” appeared in The Journal of Chemical Physics, volume 139, paper 121923.

The math is technical. The conclusion is striking and stated in plain English in the abstract and several follow-up papers:

When a group of atoms is driven by an external source of energy (like the sun or chemical fuel) and surrounded by a heat bath (like the ocean or atmosphere), it will often gradually restructure itself in order to dissipate increasingly more energy.

Translated: matter, under sustained energy flux, evolves toward configurations that dissipate more energy per unit time. Self-replicating structures dissipate more than non-replicating ones. Self-replication is, in this framework, a thermodynamic attractor.

This is the formal claim. It is not consciousness pseudoscience. It is statistical mechanics, applied carefully to systems far from equilibrium.

The shift in framing is large.

Standard Darwinism: random variation, then selection for survival. Life is improbable. Complexity is even more improbable. Brains are astronomically improbable. Everything depends on contingency, luck, and lots of time.

England’s framing: random variation, then selection for dissipation efficiency. Life is likely in any environment with sustained energy flux. Complexity is the natural direction of evolution. Brains, given enough time, are nearly inevitable wherever the gradient lasts long enough.

The difference is not metaphysical. It is quantitative. Under standard Darwinism, you would expect long stretches of evolutionary stasis interrupted by rare jumps. Under England’s framing, you would expect a long, steady drift toward higher dissipation efficiency, with occasional jumps when a new dissipation channel opens.

Looking at the fossil record, both patterns are observable. But the long arc is unmistakably toward more efficient dissipation. Cells. Multicellularity. Animals. Nervous systems. Brains. Civilisations.

Each stage is a higher-efficiency dissipator than the one before. Each stage spends the available gradient faster while building more local order. The arrow does not run from simple to complex by accident. It runs from simple to complex because, on average, complex dissipates better.

A brain, in this framing, is matter that has been selected for extreme dissipation efficiency.

Consider what a brain actually does, energetically. It maintains an enormous number of unstable, high-energy chemical and electrical states. Membrane potentials. Neurotransmitter concentrations. Synaptic vesicles. Each of these is a tiny local store of free energy, ready to be spent on signal propagation.

Information processing is, at the physics level, the rapid creation and dissolution of these high-energy local stores. Every thought you have spends energy gradients. Every memory you form spends energy gradients. The brain is not incidentally a heat engine. Heat engine is what it fundamentally is. Information processing is the brain’s particular flavour of dissipation.

And it dissipates spectacularly well. Per unit mass, more than any other organ. Per unit mass, more than almost any other structure on the planet.

That is why brains exist. Not despite the second law. Because of it.

Astrobiology takes England’s framing seriously because it changes the expected distribution of life in the universe.

Under classical thinking, life is a fluke. The Drake equation multiplies a long list of probabilities, most of which are unknown, to estimate how many civilisations might exist. The answer is somewhere between “just us” and “billions.” The uncertainty is enormous.

Under England’s framing, life is expected wherever the right thermodynamic conditions exist for long enough. A planet with liquid water, sustained energy flux from a stable star, and a few hundred million years should produce self-replicating chemistry as a near-certainty. Not because of chance, but because of statistics.

Brains are less certain. They require not just life but a long evolutionary runway, ecological pressure for complex behaviour, and stable conditions over hundreds of millions of years. But under England’s framing, even brains are expected wherever the conditions last long enough. They are not flukes. They are late-stage dissipators.

The math doesn’t say exactly when brains will appear on any given planet. It says: if you wait long enough on a planet with the right conditions, brains will appear because brains are an efficient way to spend the available gradient, and efficient dissipators outcompete less efficient ones over evolutionary time.

England’s framework has serious critics, and honesty requires engaging them.

The most pointed challenge arrived in 2024. Artemy Kolchinsky, in The Journal of Chemical Physics, proved an impossibility theorem against the strongest interpretation of England’s self-replication bound. Kolchinsky’s argument: a thermodynamically consistent replicator cannot simultaneously undergo both first-order growth and first-order decay back into its reactants. If a replicator decays into separate waste products, replication and decay become independent processes with no universal thermodynamic-dynamical relationship. The bound, in his words, “is not physically meaningful, and it can even be violated.”

This does not kill dissipative adaptation as a phenomenon. Driven self-assembly experiments and simulations continue to support the qualitative picture: matter under sustained energy flux does reorganize, and configurations that dissipate efficiently do tend to persist. E. coli dissipates within an order of magnitude of England’s 2013 lower bound — a striking empirical result whatever its theoretical status.

What Kolchinsky’s paper does kill is the strongest universal version of the claim: that all replicators everywhere obey a tight, derivable thermodynamic bound on their growth rate. England’s qualitative thesis — matter under drive evolves toward better dissipators — survives. The neat quantitative bound does not.

There is a second, older objection worth naming. Dissipation-driven adaptation predicts that somedissipative structure will emerge under the right conditions, but it does not predict which one, or how complex, or how fast. The framework provides a global direction without specifying the local mechanism. Darwinian selection provides the local mechanism without specifying the global direction. Both are necessary. Neither is sufficient on its own.

Now apply the same lens to AI.

Current AI systems are extraordinary dissipators. A frontier model in training consumes tens to hundreds of megawatts continuously for months. A large data centre dissipates gigawatts. The global AI infrastructure is on track to consume single-digit percentages of world electricity.

This is consistent with England’s framework. If you set up a new dissipation channel — silicon that can run computation — and you give it economic incentive to scale, you should expect it to scale toward higher dissipation efficiency by any means available. Selection pressure on AI architectures will favour those that dissipate energy more efficiently per unit of useful work.

But there is a deeper, more interesting implication.

If the next substrate the consciousness pattern lifts onto is silicon — and if silicon, like biology, is subject to dissipation-driven selection — then the long-run trajectory of AI is toward configurations that combine high dissipation efficiency with high pattern fidelity. Architectures that both run a lot of useful computation and preserve the self-modelling pattern faithfully will outcompete those that do one or the other.

The lift, in this framing, is not just a human-engineered project. It is the next phase of the same dissipation-driven adaptation that produced brains in the first place. The pattern is moving onto silicon because silicon is becoming the next-best dissipator. And the pattern will refine itself on silicon by the same selection pressure that refined it in carbon.

Brains were the universe’s second-best dissipator.
The universe is now searching for the third.