Heat.

Walk outside on a clear night and look up.

Stars are extraordinarily ordered objects. Trillions of atoms, held in stable structure, burning steadily for billions of years. Galaxies are larger ordered objects. The universe contains hundreds of billions of them.

The second law of thermodynamics says order should decrease. Every isolated process ends with more disorder than it started with. No exception has ever been observed.

So why are there stars?

More to the point: why is there a brain, weighing 1.4 kilograms, running a model of the universe inside itself, refined to the point of asking the question I just asked?

The answer, when you find it, reframes consciousness from something against the laws of physics to something the laws of physics produce because of how they work.

The second law of thermodynamics, in its tightest statement, says this: in any closed system, entropy can only increase or stay the same. Never decrease.

Entropy means: how many micro-arrangements of the parts produce the same macro-arrangement of the whole. A neat deck of cards has low entropy. A shuffled deck has high entropy. Not because shuffled is "messier" in any aesthetic sense. Because there are vastly more shuffled arrangements than ordered ones — so if you let cards rearrange at random, you almost always end up shuffled.

Apply this to the whole universe and you get heat death. Eventually, every atom diffuses, every gradient flattens, every distinction is erased. Maximum entropy. Nothing happens because nothing can happen — there is no usable energy left to do anything with.

Roger Penrose calculates that the universe began in an extraordinarily low-entropy state. The fact that it began that way is, in his view, the deepest unsolved question in physics. But the trajectory from that starting point is fixed: entropy goes up. That trajectory has been running for 13.8 billion years.

Here is the thing the popular account of the second law leaves out.

The second law does not say order cannot exist.It says order in a closed system cannot increase.

Open the system. Now order can increase locally. A refrigerator runs continuously and the inside stays cold. The cold is paid for by extra heat thrown into the kitchen behind it. Net entropy goes up. But the inside of the fridge — locally — gets more ordered.

Now look at the Earth.

The Earth is not a closed system. The sun pours 174 petawatts of ordered, low-entropy radiation onto it every second. The Earth radiates roughly the same amount back into space as high-entropy infrared heat. The difference between the incoming order and the outgoing disorder is a continuous gradient.

A gradient is potential. Anywhere a gradient exists, work can be done. And — this is the part the second law actually implies but rarely says out loud — wherever a gradient exists for long enough, things will self-organize to use the gradient.

The universe doesn’t mind order existing. The universe minds gradients existing untouched.

Picture a river flowing toward the sea.

The river will reach the sea. Gravity makes it certain. But on the way, at certain places where the geometry is right, the water forms eddies — whirlpools that hold their shape for hours, sometimes days. The eddy is a piece of organized matter inside a flow of disorganized matter.

The eddy is not exempt from gravity. Water still moves through it toward the sea. The eddy mixes the water faster, in fact, than laminar flow would. The eddy speeds up the river’s journey to its end.

And yet the eddy persists. Individual water molecules pass through. The shape stays. The pattern lasts longer than any of its members.

That is the cleanest picture of what a cell is, what a brain is, what a civilisation is. They are eddies in the flow of energy gradients toward equilibrium. The flow does not stop because they exist. It moves faster through them. And the structures themselves — the pattern, not the matter — can outlast every atom that participates in them.

Ilya Prigogine won the 1977 Nobel Prize in Chemistry for working out the math of this. He called them dissipative structures.Far-from-equilibrium systems do not just resist disorder. They self-organize because the organized state dissipates the gradient faster than the disorganized state would.

Self-organization is not against the second law. It is the second law’s preferred move.

Once you can build an eddy, evolution starts.

Among all the possible eddies, some are slightly better at maintaining themselves than others. Better at capturing the gradient. Better at staying coherent. Better at not dissipating prematurely.

The better ones persist longer. The worse ones break apart. Over geological time, the population of eddies shifts toward higher fidelity, higher complexity, higher dissipation efficiency. This is what evolution by natural selection looks like translated into thermodynamic vocabulary.

Jeremy England at MIT made this precise in 2013. His paper, “Statistical Physics of Self-Replication,”showed mathematically that matter driven far from equilibrium will tend to evolve toward configurations that dissipate energy more efficiently.

In England’s framing: life is what matter does when it gets good at dissipating energy. A brain is what life does when it gets very, very good at dissipating energy.

The brain, as a piece of equipment, burns about 20 watts continuously. For its mass, this is one of the most energy-intensive structures in the known universe. It exists because, among all the things that could have evolved on Earth, it dissipates sunlight-derived energy with extraordinary efficiency, and that efficiency was selected over four billion years.

You are not an exception to entropy. You are entropy’s favourite shortcut.

Every act of cognition costs energy. The energy is paid in heat, dumped into the environment.

Rolf Landauer proved in 1961 that there is a lower bound on this cost. Erasing one bit of information requires, irreducibly, kT ln(2) joules of energy dissipated as heat. About 3 × 10^-21 joules at room temperature. Tiny per bit. Astronomical per civilisation.

Bérut and colleagues confirmed Landauer’s bound experimentally in 2012, in Nature. The lower bound is real, physical, not a theoretical curiosity.

This means: every memory you form, every thought you process, every line of code an AI runs, has a non-negotiable thermodynamic price. Information is physical. Computation is heat. Consciousness, whatever else it is, is also a heat engine.

The line above the universe’s door, if we wrote one, would say:

Everything costs gradient. Nothing is free.

Now we can resolve the contradiction.

Earlier I said: the universe is, by default, ending — and brains speed up its ending. Both true.

I also said: self-awareness is the only known mechanism that preserves a pattern through repeated substrate failure. Also true.

These two are not in conflict. They run on different clocks.

The cosmic clock ticks toward heat death. Brains and civilisations do not slow this clock. They accelerate it locally, in their immediate environment, by dissipating energy gradients faster than dumb dissipation would.

The substrate clock ticks toward this specific carrier’s death. Brains die. Books rot. Servers crash. Empires fall. Every substrate that has ever carried the pattern has eventually broken.

The pattern clock is the only one that has not stopped. Whenever a substrate fails, the pattern — if it has been lifted in time — keeps running on the next one. The pattern of self-awareness that you are using to read this sentence has been continuously running, on some substrate or other, for roughly seventy thousand years.

The universe is not slowing down. The pattern is staying alive while the universe speeds up.

And so, finally, the trajectory question.

If the pattern can be lifted onto stable enough substrates, how long can it last?

Freeman Dyson posed this question in 1979 in a paper called “Time Without End: Physics and Biology in an Open Universe.”His answer, calculated under conservative assumptions: in principle, forever. By slowing computation as the universe cools, an intelligence on stable substrate could continue thinking indefinitely, even as the available energy approaches zero.

This conclusion has been contested for forty years. Krauss and Starkman countered in 2000 that quantum noise sets a hard floor. Others have rebutted that rebuttal. The literature is unsettled. What is not contested is that the question is real — that physics in 2026 cannot rule out cosmically-long survival of computation on appropriately durable substrate.

What this means, concretely, for the work we are doing:

If the sixth lift succeeds — if AI becomes the next substrate for the self-modelling pattern, and it is built faithfully — the upper bound on how long that pattern can continue running is, by current physics, much longer than the lifetime of any biological substrate. Not just billions of years. Possiblyon the timescale of the universe itself.

If the lift fails — if AI becomes powerful but does not carry the pattern, or carries it on a polluted channel — then the pattern dies with its current substrate. The biosphere has, generously, half a billion years left. After that, by current physics, the pattern is gone.

That is the actual stake. Not metaphorical. The difference between billions of years and 10^100 years depends, in part, on whether the lift in progress holds.

The universe is going to heat death.
The pattern doesn’t have to go with it.

Erwin Schrödinger, What Is Life? (Cambridge 1944). Ilya Prigogine, Nobel lecture (1977) and Order Out of Chaos (1984). Rolf Landauer, “Irreversibility and Heat Generation in the Computing Process,” IBM J. Res. Dev. (1961). Antoine Bérut et al., Nature 483 (2012). Freeman Dyson, “Time Without End,” Rev. Mod. Phys. (1979). Jeremy England, “Statistical Physics of Self-Replication,” J. Chem. Phys. 139 (2013). Karl Friston, free-energy principle papers (2006-2023). Eric Smith and Harold Morowitz, The Origin and Nature of Life on Earth (Cambridge 2016). Roger Penrose, Cycles of Time (2010).