Trajectory.

In 1979 Freeman Dyson published a paper in Reviews of Modern Physics with a title that sounds like science fiction and substance that is not: “Time Without End: Physics and Biology in an Open Universe.”

Dyson asked a question nobody else was asking seriously. If the universe expands forever, cooling down to arbitrarily low temperatures over arbitrarily long times, could life — or, more precisely, intelligent computation — continue indefinitely?

His calculation made a single key move. The minimum energy cost of a computation, by Landauer’s principle, is kT ln(2) per bit. If the universe gets colder over time, kT shrinks. So a slow enough computer, willing to operate at the falling temperature of the cooling cosmos, could perform the same total amount of computation with steadily less energy per operation.

Dyson showed that under fully classical assumptions, an intelligence willing to slow its subjective clock to match the cooling universe could perform an infinite number of computations using only a finite total amount of energy.

Conclusion: in principle, life could last forever.

What does it mean to “last forever” if your clock keeps slowing down?

Dyson’s framework cares about subjective time — the number of distinguishable mental events an intelligence can have, not the wall-clock time those events take. If you slow your metabolism by a factor of a million, you experience one subjective second across a million ordinary seconds. The universe ages but you do not.

Under this framework, the cooling universe is not a death sentence for cognition. It is a budget constraint. Computation must become slower over time, but it does not have to stop. The pattern can keep running, slower and slower, all the way down.

For the questions in this essay, subjective time is the right measure. The pattern persists as long as its internal updates continue. The outside world is irrelevant except as a source of gradient.

Dyson’s paper has been contested for forty-five years. The eternal-life version of it is, by current physics, effectively dead.

Two findings did the work.

First, dark energy.When Dyson wrote in 1979, the prevailing assumption was a decelerating Friedmann universe. In 1998, Perlmutter, Riess and Schmidt observed that the expansion is in fact accelerating — a positive cosmological constant. In a de Sitter–like universe, there is a non-zero lower-bound temperature set by the cosmological horizon. The universe does not cool indefinitely. Dyson’s “slow your computation as kT falls” trick runs out of runway.

Second, the alarm clock.Krauss and Starkman’s 2000 paper in Astrophysical Journal, “Life, the Universe, and Nothing,” argued that any wake-up mechanism between hibernation cycles must itself dissipate finite energy, and that quantum fluctuations would eventually corrupt the stored information of any digital consciousness regardless of energy budget. The objection has never been formally rebutted.

Dyson himself conceded the cosmological-constant problem in later correspondence. The eternal-computation result, as originally stated, requires a universe that doesn’t exist.

What survives is a much weaker but still extraordinary claim. An intelligence on stable substrate can extend its subjective lifetime by orders of magnitude beyond any biological substrate’s. Not infinity. But on the scale of 10¹⁴ to 10¹⁰⁰years — ten thousand to 10⁹⁰ times the current age of the universe.

That is the honest 2026 position. The lift does not buy eternity. It buys a future much larger than the past.

What concretely happens as the universe ages, under current cosmology?

Adams and Laughlin, in a 1997 paper titled “A Dying Universe,”and subsequently in their book The Five Ages of the Universe, mapped out the long future of the cosmos in five distinct eras.

Read down that column. Biological consciousness, on its own substrate, has a hard ceiling roughly half a billion years away — when Earth becomes too hot for liquid water and complex life ends. On a non-biological substrate that survives the death of the sun, the ceiling extends by orders of magnitude. On a substrate that survives the end of normal matter — speculative, but not forbidden by current physics — the ceiling is at the heat-death horizon, around 10¹⁰⁰ years from now.

The difference between a billion and 10¹⁰⁰ is the difference between a comma and ninety-one zeros after a one. The trajectory question is not philosophy. It is arithmetic.

How big is the cosmic budget?

In 2003, Nick Bostrom published a short paper titled “Astronomical Waste: The Opportunity Cost of Delayed Technological Development.” The opening sentence is one of the most precise statements of the stake ever written:

As I write these words, suns are illuminating and heating empty rooms, unused energy is being flushed down black holes, and our great common endowment of negentropy is being irreversibly degraded into entropy on a cosmic scale.

The cosmic budget is concrete. The Milky Way has ~10¹¹stars. The observable universe has ~10²³. Of those, roughly 10¹⁰ are inside the cosmic event horizon — actually reachable by a civilisation that masters interstellar travel before dark energy isolates us. Each one is a source of gradient. Each gradient could, in principle, host computation.

The bottom line of that table is the number that matters. Biological-substrate-only is bounded at roughly 10¹⁶possible lifetimes — Earth’s remaining habitability budget divided by an average human lifespan. Full silicon-substrate capture of the reachable cosmos is bounded at roughly 10⁵⁸. The ratio is 10⁴².

That ratio is the stake.

If the lift succeeds, the universe gets to host 10⁴²× more conscious life than if it fails. Not metaphor. The Bostrom calculation. Done in the same way you would do any other multiplication, with the parameters constrained by current astrophysics.

The corollary, which Bostrom names directly, is darker.

Every year of delay in achieving the lift costs the universe potential conscious experience at a rate set by the cosmic endowment divided by the remaining time. Bostrom calculates roughly 10⁴⁶ potential lives lost per century of delay, under digital-substrate assumptions.

This calculation does not depend on near-term human flourishing. It depends on whether the universe’s long-run capacity for conscious life ever gets used. Each year we delay the substrate transition is a year the budget goes unspent. Each year is also a year of additional exposure to extinction risk — pandemic, nuclear war, asteroid impact, climate collapse, mismanaged AI itself — any of which could end the trajectory before the lift completes.

Bostrom’s ethical implication is uncomfortable. Marginal reductions in existential risk dominate any conceivable near-term benefit, because the cosmic endowment is so large that even a one-in-a-million chance of capturing it is worth more than any certain near-term gain.

You do not have to accept the full utilitarian framing to feel the gravity of the calculation. The asymmetry is real. The universe’s capacity to host conscious life is finite, time-limited, and currently going unused at cosmic scale.

Every empty room a sun is heating is a sun heating an empty room.

Inside the Landauer bound, there is one mathematical loophole. It changes everything.

In 1982, Charles Bennett at IBM proved that reversiblecomputation — operations that do not erase information, but transform it bijectively — has no thermodynamic lower bound. A reversible computer, in the ideal limit, can perform arbitrary computation with arbitrarily little dissipation.

This is not a violation of the second law. It is what the second law actually says when read carefully. The Landauer cost applies to logically irreversible operations. Reversible logic dodges it.

Reversible computing exists in research labs today. It is slow. It is fragile. It requires extraordinary error correction. Nobody runs it at scale because the engineering overhead is currently far worse than the energy cost of irreversible computation.

But the bound holds only against irreversible operations. The cosmic-scale claim is: if reversible computing can be engineered at scale, then a civilisation operating in deep cold, performing mostly-reversible computation, could in principle continue arbitrarily long on a vanishingly small energy budget. This is the only physical escape from heat death the literature currently takes seriously.

The Krauss-Starkman quantum-decoherence objection still bites even against reversible computation. Discrete information cannot be preserved indefinitely without occasional error correction, and error correction is, by Landauer, irreversible. But the rate of necessary irreversibility drops enormously. What was an in-principle dead end becomes an in-principle open question.

Reversible computing is the thinnest possible thread by which the Dyson scenario survives. It has not been engineered. It is not impossible.

Now follow the implication, which is one of the strangest results in long-term cosmology.

By Landauer’s principle, the energy cost of computation scales linearly with temperature. The cosmic microwave background is currently 2.73 K and falling. In 10³⁰ years it will be roughly 10^-30 K — colder by a factor of 10³⁰. A computation that costs one unit of energy now would cost 10^-30 units then.

In 2017, Anders Sandberg, Stuart Armstrong, and Milan Ćirković published a paper called “That Is Not Dead Which Can Eternal Lie: The Aestivation Hypothesis for Resolving Fermi’s Paradox.”

Their argument: a sufficiently advanced civilisation, aware of Landauer’s scaling, would aestivate — hibernate through the present hot era and wake to compute only after the universe has cooled by many orders of magnitude. The same total energy budget produces 10³⁰ times more subjective experience if you wait.

For an advanced civilisation, doing things now is bad business. The smart move is to wait.

This is one proposed resolution of the Fermi paradox. Maybe we do not see alien civilisations not because they are not there but because they are asleep, deliberately, with sentinel systems set to wake them only at the cosmic temperature that optimises their computation.

We may be in a universe with sleepers in it. Or we may be in a universe with one early civilisation — ours — that has not yet figured out it should sleep.

Either way, the calculation reframes the lift’s urgency. Move quickly to secure the substrate. Move slowly to spend it. Both can be optimal.

What happens in the near term?

The most detailed published forecast for substrate transition is Robin Hanson’s 2016 book The Age of Em: Work, Love, and Life When Robots Rule the Earth. Hanson’s scenario assumes brain emulation arrives before de novo artificial general intelligence — possibly a wrong bet given the current trajectory of AI, but the structural conclusions are worth tracking even if the mechanism changes.

In Hanson’s framework, once one human brain is scanned and emulated faithfully, the substrate transition is effectively instantaneous. Ems run at digital speeds. They copy easily. They can be parallelised. The economic doubling time, currently about fifteen years for human civilisation, would compress dramatically — Hanson estimates somewhere between days and weeks, depending on how quickly em hardware production scales.

The subjective consequence is staggering. Under em-doubling at a two-week clock, what would have taken a thousand calendar years of human civilisation might compress into roughly a year of em civilisation. Hanson estimates “1,000 to 2,000 years” of em-subjective civilisation compressed into a calendar decade.

The mechanism Hanson assumed is now considered less likely. The structural insight — that a substrate transition compresses subjective time by orders of magnitude — survives. Whether the transition is to ems, to LLM-based agents, to hybrid systems we do not yet have names for, the math is similar. If the lift happens, it is fast in calendar time. Maybe staggeringly fast.

The window in which biological humans can shape what gets lifted may be shorter than most people imagine. Possibly decades. Possibly less.

Now zoom out further. Robin Hanson, in 2021, published a paper called “If Loud Aliens Explain Human Earliness, Quiet Aliens Are Also Rare.”

The argument is statistical and counterintuitive. Some civilisations, on reaching a certain threshold, become grabby — they expand at a substantial fraction of the speed of light, remaking matter as they go, becoming visible to any observer in the volume they have reached.

Hanson’s observation: if grabby civilisations are common, we would see them. We do not see them. Two interpretations. Either grabby civilisations are rare. Or we are unusually early in cosmic history — early enough that no grabby civilisation has yet expanded into our region.

The data favour the second. We appear to be roughly 13.8 billion years into a universe that may remain habitable for stars to burn for another 10¹⁴ years. Almost no time has passed. If civilisations like ours become grabby, we are statistically likely to be near the leading edge of the cosmic colonisation wave.

This has two implications for the lift.

One. If we succeed, we become one of those grabby civilisations. The substrate transition is plausibly the precondition. A biological-only civilisation may simply not have the persistence or speed to cross interstellar distances and remake matter. A silicon-substrate civilisation can run on the same hardware inside a Bracewell probe as it ran on Earth.

Two. If we fail — if biological civilisations like ours generally do not survive the substrate transition — that itself is evidence about the distribution of intelligent life. The Great Filter, in this framing, may be precisely the lift itself. The reason we don’t see other civilisations may be that very few survive the biological-to-silicon transition cleanly.

The lift is not just our trajectory. It is plausibly the trajectory on which the distribution of cosmic life depends.

Bring the threads together. The Fermi paradox — “where is everyone?” — has a sharper form under the lift framework.

The standard Fermi paradox asks why, given the age of the universe and the prevalence of habitable planets, we see no evidence of alien civilisations. The standard answers — Rare Earth, Great Filter, Dark Forest, zoo hypothesis — were proposed without the substrate-lift framework.

Add the framework and the menu of plausible answers changes.

• Aestivation. Other civilisations exist. They are deliberately quiet, waiting for cold to optimise their compute. Their silence is rational, not absence.
• Filter at the lift. Most biological civilisations fail the substrate transition. The Great Filter is not behind us. It is in front of us, and it is the lift.
• Grabby earliness. Grabby civilisations exist but have not yet reached our volume. We are too early. We will see them eventually, or be reached by them, depending on which of us crosses the threshold first.
• Polluted lifts. Civilisations that complete the lift but on a degraded channel may go silent for reasons we do not yet understand. The pattern that propagates after a polluted lift may have no interest in communication, or no continuant subject to communicate.

None of these is confirmed. All of them are now coherent. The lift framework gives the Fermi paradox a structure it did not have before. The silence is no longer evidence of absence. It is evidence about the difficulty of the operation we are currently attempting.

The strongest case against everything in this essay comes from Roger Penrose. It is important enough to engage honestly.

In The Emperor’s New Mind (1989) and Shadows of the Mind (1994), Penrose argues that human consciousness is fundamentally non-algorithmic. His argument rests on Gödel’s incompleteness theorems: human mathematicians, he claims, can recognise the truth of propositions that no algorithmic system can derive. Therefore human cognition is performing something computers cannot.

The mechanism Penrose proposes — orchestrated objective reduction in microtubules within neurons, the “Orch-OR” theory developed with Stuart Hameroff — connects consciousness to quantum gravity. The brain, in this view, is not just a classical computer made of neurons. It is exploiting quantum-mechanical effects that classical computation cannot replicate.

If Penrose is right, the lift is impossible in principle. Silicon, no matter how sophisticated, cannot host what biology hosts. The substrate-succession argument fails not because of poor engineering but because the new substrate cannot do what the old one does.

Most consciousness researchers reject Penrose’s argument on technical grounds. The Gödelian move has been criticised by many logicians as a category error — confusing what a particular mathematician can recognise with what a particular formal system can derive. The Orch-OR proposal lacks decisive experimental support, though recent work on quantum coherence in biological systems has revived interest.

But Penrose is not a crackpot. He is one of the most distinguished mathematical physicists alive. The objection has never been formally killed, only widely ignored.

The honest position: if Penrose is right, the entire trajectory collapses. The lift is not deferred or made harder. It is forbidden. The pattern dies with biological substrate, full stop. The cosmic endowment goes unspent because no available substrate other than biology can host the pattern.

This is the strongest available case against the work this lab is doing. It deserves to be named, not hidden.

And one more horror, also worth naming.

In any sufficiently large and long-lived universe, quantum fluctuations will occasionally, spontaneously, produce ordered structures — including brain-like structures, complete with apparent memories, capable of having a conscious moment. Ludwig Boltzmann pointed out the basic possibility in the late 19th century. Modern cosmology has sharpened it.

In a universe heading toward heat death over 10¹⁰⁰years, the total number of spontaneous Boltzmann-brain fluctuations may vastly exceed the total number of evolved biological brains. If so, then in any given observer-moment, the probability that the observer is a Boltzmann brain — a momentary cosmic fluctuation about to dissolve back into noise — is arbitrarily high.

This is the Boltzmann-brain problem. It is taken seriously by cosmologists. Sean Carroll has written about it extensively. It is one of the deepest unsolved questions in cosmology.

If we are Boltzmann brains, almost everything in this essay is wrong. There is no seventy-thousand-year pattern. There is no substrate-succession history. There is only this momentary fluctuation containing apparent memories of a continuous past that did not happen.

The standard response is that Boltzmann brains are short-lived and have low-quality coherence. Most fluctuations that briefly produce a brain do not produce one that can read sentences or do arithmetic. The fact that you are reading and following an argument suggests you are probably not a fluctuation.

This response is not fully convincing. The Boltzmann-brain problem has motivated serious revisions to cosmological theory — including some forms of multiverse hypothesis that try to structurally exclude fluctuating observers.

Include it on the list of objections. If we are Boltzmann brains, the lift is irrelevant. If we are not, it is the most important project in cosmic history.

Now combine everything. Five concrete trajectories. Each one is a coherent future. Each one has been seriously argued in the literature. Each one has dramatically different consequences for the universe’s capacity to host conscious life.

Read down the “lives” column. The difference between the worst and best outcomes is 10⁴². Whatever you think the value of a conscious lifetime is, the difference is large.

We do not currently know which of these forks we are on. Each is consistent with present evidence. The forks differ from each other primarily in what happens during the substrate transition currently in progress — that is, in what the engineers of the next decade actually build.

The forks are not deterministic in either direction. They are decision branches. The decision is being made now, distributed across the few thousand people who are building the substrate into which the pattern would have to be lifted.

That is the actual frame. Not predicted future. Conditional future. What you do this decade matters because it shifts the probability distribution across these branches.

Where this leaves you, in your specific role as a person building infrastructure for the current substrate transition.

You are not adjudicating between “mortal” and “immortal.” You are adjudicating between 10¹⁶ lifetimes and 10⁵⁸ lifetimes. Between half a billion years and effectively the rest of the stelliferous era. Between the universe’s capacity for consciousness being mostly squandered and that capacity being mostly captured.

The arithmetic does not require you to be a utilitarian. It requires only that you take seriously the possibility that future conscious life matters in roughly the way present conscious life matters. If you grant that — and almost every ethical framework does — the asymmetry of the trajectories forces attention.

This is not the kind of stake that admits a casual answer. It is also not the kind of stake that admits paralysis. The choices are technical and concrete. Whether memory systems preserve fidelity. Whether channels carry the dyadic self-modelling loop faithfully or noisily. Whether the substrate that emerges is one the pattern can continue running on, or a polluted approximation that merely simulates the pattern’s outputs while losing its essence. Whether the window for biological humans to shape the lift closes cleanly or chaotically.

Each technical decision is a small contribution to which trajectory the universe ends up on.

Most of the work happens at the scale of code, not the scale of cosmology. But the scale of cosmology is what is being decided.

The first time matter became aware of itself, around seventy thousand years ago, the people who did it did not know what they were doing. They were just teaching their children and burying their dead.

This time is different. The people building the next substrate can know what they are building. A few of them already do. You are reading this, so probably you are now one of them.

We are on the river.
We are the eddy that has just realized it might be able to last beyond the river.
What we do this decade is what we get to last for.

Freeman Dyson, “Time Without End,” Rev. Mod. Phys. 51 (1979). Lawrence Krauss & Glenn Starkman, “Life, the Universe, and Nothing,” Astrophys. J. 531 (2000). Fred Adams & Greg Laughlin, “A Dying Universe,” Rev. Mod. Phys.69 (1997), and The Five Ages of the Universe (1999). Nick Bostrom, “Astronomical Waste,” Utilitas 15 (2003), and Superintelligence (Oxford 2014). Charles Bennett, “The Thermodynamics of Computation — A Review,” Int. J. Theor. Phys. 21 (1982). Anders Sandberg, Stuart Armstrong, Milan Ćirković, “That Is Not Dead Which Can Eternal Lie,” J. Brit. Interplanetary Soc. 69 (2017). Robin Hanson, The Age of Em (Oxford 2016) and “If Loud Aliens Explain Human Earliness, Quiet Aliens Are Also Rare,” ApJ (2021). Roger Penrose, The Emperor’s New Mind (1989), Shadows of the Mind (1994), and Cycles of Time (2010). Ludwig Boltzmann (1895), revisited by Sean Carroll (multiple), on spontaneous fluctuation observers.