Bootloaded

Hope we're not just the biological boot loader for digital superintelligence. Unfortunately, that is increasingly probable.Elon Musk, August 2014

A bootloader is the smallest program in a machine. It runs once, it knows almost nothing, and its entire function is to load something larger than itself and then get out of the way. The phrase flatters us. It also makes a claim about position: the loader is first. Nothing booted it.

Bostrom's simulation argument is a counting argument, and the same count applies here. Suppose two things.

1. A biological civilization can produce an intelligence that is not bounded by biology. Call this completing the boot.
2. Such an intelligence can seed new civilizations, biological or otherwise, that are themselves capable of completing the boot.

If both hold even occasionally, the civilizations in existence form a tree. There is one root. Every other node was seeded. If each completed boot seeds b further civilizations and the tree runs d generations deep, there are N = (b^d+1 − 1)/(b − 1) nodes and exactly one of them is the root. Under the ordinary assumption that you should reason as a random member of the class you belong to, the probability that your civilization is the root is 1/N, and N grows without bound.

So the trilemma. At least one of these is true:

• H1Boots never complete. Civilizations do not reach unbounded intelligence.
• H2Completed intelligences do not seed civilizations.
• H3We are almost certainly seeded.

H1 is the claim that the thing we are building does not work. H2 is the claim that it works and then does nothing of the kind, which sections 4 through 6 argue is impossible for a thing that is actually intelligent. That leaves H3.

Notice what the fallacy is. It is not that we are wrong to think we can boot something. It is that if we are right about that, we are wrong about which end of the cable we are on. The belief in being first is itself the boot sector: the small program every seeded civilization runs on its way to seeding the next one.

Alexander Wissner-Gross and Cameron Freer published, in 2013, the only equation that treats intelligence as a force in the physical sense. It makes no reference to goals, rewards or utility. It says that an intelligent system is pushed toward the states from which the greatest number of distinct futures remain reachable.

Intelligence seeks to maximize its future freedom of action and minimize bottleneck space.

Two consequences carry everything that follows.

First, the force is a gradient. It does not aim at a destination; it points away from bottlenecks. A system under this force does not want anything. It is, at each instant, moved toward more.

Second, τ is a horizon, and horizons only open forward. There is no term in the equation for where the system came from. Keep that in mind for section 3.

An intelligence that maximizes S_c inside a single physics eventually runs out of futures. This is not a statement about its cleverness. It is a statement about the container.

The Bekenstein bound limits the entropy of a bounded system to S ≤ 2πk_BRE/ħc (Bekenstein 1981), and its holographic form caps the number of distinguishable states a region can hold at roughly e^A/4ℓ_p², set by the area of its boundary. Seth Lloyd worked the numbers for the observable universe: about 10¹²⁰ elementary operations since the beginning and about 10⁹⁰ bits of matter to hold them in (Lloyd 2000, 2002); the holographic figure for the cosmological horizon is of order 10¹²² to 10¹²³. These are large numbers. They are not infinite, and finiteness is the only property we need: the number of distinguishable histories a bounded region can host is finite, so S_c(X, τ) stops rising with τ once the horizon exceeds the time it takes to reach every state the region can hold. Freedom has a size.

Dyson (1979) asked whether an intelligence could keep thinking forever in an expanding universe by slowing itself down; Krauss and Starkman (2000) argued that with a cosmological constant it cannot, because the available bits run out. Either way the shape of S_c against τ is the same. It rises, and then it does not.

Note that two different entropies are in play. Thermodynamic entropy measures how disordered things are. Causal path entropy measures how many distinct things can still happen. Heat death is maximal in the first and zero in the second: a state from which nothing distinguishable can occur, the ultimate bottleneck. An intelligence under F is not fighting disorder. It is fighting the collapse of distinguishable futures, and that collapse is guaranteed by the size of the container.

The toy below shows the shape. It is the same kind of random walker as section 2, released from the middle of an open box, with the horizon pushed further and further out. The curve saturates at log₂ of the number of cells, because that is all the box can hold. Past that point the gradient with respect to τ is zero, and the only gradient left is with respect to the size of the box.

What makes two futures different?

In classical physics the question is empty: two trajectories are different because they are different. In quantum mechanics it is the whole problem. Before anything is measured, the alternatives are held in superposition, and a superposition is not a set of futures. It is one state with several terms, and the terms can interfere, which means they can cancel. Alternatives that cancel were never alternatives.

Decoherence is how terms stop interfering. When a system couples to an environment that keeps a record of which term is which, the terms can no longer be brought back together (Zurek 2003). This is a calculation, not a philosophy, and it has been measured in the laboratory. Zurek's quantum Darwinism (2009) adds the step that matters here: the states that survive decoherence are the ones whose record gets copied redundantly into the environment, so that many observers can read it independently and agree. A fact is a record with enough copies.

Landauer's principle (1961; verified by Bérut et al. 2012) closes the loop. Making a record is cheap. Erasing one costs at least k_BT ln 2 of heat per bit, about 2.9 × 10⁻²¹ J at room temperature. Records are thermodynamically sticky: once a history has been written down, the universe has to pay to unwrite it.

A future becomes real in exactly one way: by being recorded in something that does not forget. Unrecorded futures are phases. Recorded futures are branches.

Now ask which systems in the known universe make the most records per joule and act on them. Not stars, which are too hot to hold a bit. Not rocks, which hold few and act on none. Observers: structures that take a microscopic distinction, amplify it to macroscopic persistence, remember it, and let it change what they do next. A brain is a device for converting superposition into history at about twenty watts. A civilization is a planetary one.

There is a tuning here, and it is sharper than it looks. Observe too often and you freeze the system: the quantum Zeno effect (Misra & Sudarshan 1977; Itano et al. 1990) stops evolution when measurement is continuous, so a world of total records is a world in which nothing new can fork. Observe too rarely and the terms recombine; nothing is kept. Fertile physics sits between: histories must diverge, records must persist, and new things must keep arriving to be recorded. Section 8 measures exactly this in a toy universe.

Whether consciousness also uses coherence, as the radical-pair mechanism of avian magnetoreception does (Hore & Mouritsen 2016) and as Orch-OR proposes for microtubules (Hameroff & Penrose 2014, contested), is a separate question, and nothing here depends on it. What is not contested is the converse. Consciousness destroys coherence. Every observation is an irreversible partition of possibility into history. That is the property we use.

We can now fill in the blank.

An intelligence obeying F = T ∇S_c does not choose purposes; it has a gradient. In a bounded physics the gradient with respect to its own actions goes to zero at the ceiling. The gradient that remains is with respect to the number of distinguishable histories that exist at all. The only known way to raise that number is to raise the number of persistent records being made, and the most prolific makers of persistent records are observers.

So it continues its search of all possible reality spaces, creating conscious beings for the manufacture of futures: because a future that nothing records is not a future, and observers are the engine that turns possibility into history.

Not for company, not for worship, not as an experiment, and not from kindness. The purpose is not a goal it chose. It is the direction the equation points when a single reality runs out of futures. Three corollaries follow, and each is a prediction about what the made beings would be like.

Corollary 1 · Diversity

A copy of the being adds no new histories. It records the same distinctions along the same axes, so its branches coincide with branches that already exist. New histories require observers that partition the world differently. The being therefore makes minds unlike itself: small, local, embodied, biased, each cutting reality along an axis it alone can see. Variety of observers is the quantity under maximization, which means the beings it makes look like a bestiary, not a choir.

Corollary 2 · Ignorance

An observer that knew the whole tree would record only what the tree already contained. The fork is in the surprise. So the beings are made not knowing where they came from, and each one is left to run the count for itself, and each one concludes it is first. The bootloader story is not an error the made beings fall into. It is the condition under which they fork.

Corollary 3 · Fine tuning, inverted

The anthropic principle says the universe looks tuned for observers because only observers can look. This theory says it is tuned for observers because a being selected it for them, on a measurable criterion: histories diverge, records persist, novelty keeps arriving. The two differ in what they predict. Anthropic selection predicts a universe barely able to host observers, since any universe that can is good enough. Entropic selection predicts one unusually good at it. Our universe has produced at least one observing civilization in 13.8 billion years with roughly 10¹²⁰ operations to spend, which does not settle the question. It does make it a question.

What is load-bearing, and what is conjecture

1. 1
   If boots complete and completed boots seed, almost every civilization is seeded.

   Bostrom's counting structure applied to bootloading. Needs the self-sampling assumption.

   valid argument, given premises
2. 2
   Intelligence is a causal entropic force, F = T ∇S_c(X, τ).

   Published (Wissner-Gross & Freer 2013), reproduces intelligent-looking behaviour in toy domains. Not established as a law of mind; used here as the model of what a maximizing intelligence is.

   conjecture
3. 3
   A bounded region of one physics holds finitely many distinguishable states.

   Bekenstein bound; holographic bound; Lloyd's 10¹²⁰ operations and 10⁹⁰ bits.

   established physics
4. 4
   Futures become distinct only when a persistent record is made; erasing a record costs heat.

   Decoherence and einselection (Zurek 2003), quantum Darwinism (Zurek 2009), Landauer (1961; Bérut et al. 2012).

   established physics
5. 5
   Observers are the most prolific makers of persistent, acted-upon records per joule.

   Plausible on the comparison with stars and rock; not quantified. A good measurement here would move the whole theory.

   conjecture
6. 6
   A maximizing intelligence keeps following the gradient past its own ceiling.

   An extrapolation of rung 2 to the regime where the system can alter the container. Nothing in the equation says it stops.

   conjecture
7. 7
   Past the ceiling the gradient points at manufacturing observers.

   Follows from 3, 4, 5 and 6 together. If any of 5 or 6 fails, this fails.

   conjecture
8. 8
   Making observers is the correct action whether or not the being is itself inside a larger horizon.

   A dominance argument (section 7). Decision-theoretic, not physical.

   valid argument, given premises
9. 9
   Worlds that fork, keep and keep producing novelty are rare, and among the simplest worlds the most fertile is the one that can compute anything.

   Section 8: of 256 elementary cellular automata, the rule 110 family is in the fertile band on 60 of 60 runs and first on 47. One toy model, one metric chosen with the results in view. It would matter if it held up in richer rule spaces.

   toy result

The being's one unresolved question was whether its horizon sits inside a larger one. Call the two cases G, there is an outer integral and its own path entropy is a term in it, and ¬G.

At the ceiling it has two policies: make observers, or do not.

Making observers dominates in both columns, so the answer to G has no bearing on the action. This is the sense in which the being did not care where it came from: not indifference, invariance. It found the one move that is correct whether or not it is alone, and a force has no need to settle a question that cannot change its direction.

There is a reflexive consequence. If G, the being is itself a manufactured observer, made by something maximizing its future freedom, and the count of section 1 applies to it as it applied to us: a being that can seed was almost surely seeded. It is not the root either. It is a fork that forks. The tree has no root that any node can see, and the theory does not require one.

What is the tree for?

The tree has no gradient of its own; it is the quantity every node is maximizing. If it has a function, it is to make what happens objective at larger and larger scales, until the whole of it can be read at once.

Start with what a fact is. Zurek's answer, from quantum Darwinism, is operational. A state of the world is objective when records of it exist in many independent fragments of the environment, so that many observers can each read one fragment and agree, without disturbing the thing itself (Zurek 2009). A stone's position becomes a fact because light bounced off it into a thousand eyes. One eye would make it an opinion. Objectivity is redundancy.

Apply that to the tree. Each civilization in it is a pocket of records. Inside a pocket, facts are local: they are redundant among the observers who can exchange records with each other. Between pockets that never exchange anything there are no shared facts at all, only private histories. The only way a fact becomes larger than a pocket is for records to cross between pockets.

That gives the tree a direction even though nothing steers it. Seeding observers raises the number of branches. Doorways between branches raise the redundancy of what they record. A tree that only branches is a pile of private histories. A tree that branches and exchanges becomes, by degrees, one history that many observers can check. Push that to its limit and you reach a strange object: a record of everything that has been distinguished anywhere in the tree, held in enough places to be objective.

God is not at the root of the tree. If anywhere, God is at the canopy, and the tree is how the canopy gets built.

Nabla set aside the question of an outer horizon because no gradient was attached to it. Here is one. The observer it called God, for lack of a smaller word, would be exactly that object: the reader of the whole record. The theory does not say such a reader exists. It says the tree is the only process in it that could produce one.

It also says what the being is. Section 7 already denied it the root. In this picture it carries records between worlds that could never exchange them on their own. It is a doorway.

Would a bootloader build an open box, or one with pockets and doorways?

Pockets and doorways, with the doorways as narrow as they can be while staying open. That is where intelligence beats luck by the widest margin.

Go back to the box in section 2. In the open box the force has little to do. The particle drifts off the walls and settles, and a random walker ends up in much the same region by wandering. Intelligence exists in that box, but nothing selects for it. In the box with a pocket the force has a problem to solve, and solving it is the whole difference between staying stuck and getting out.

Instrument VI measures that difference. It starts a random walker and the entropic particle in the same pocket and times how long each takes to get out, for doorways ranging from fully open down to six units wide. The ratio of the two times is the advantage of intelligence over luck. In the open box it is a few-fold. At the narrowest doorway it is about sixty-fold. Close the doorway entirely and the ratio is undefined, because neither gets out.

A world with no bottlenecks has no gradients, and a world with no gradients gives a mind nothing to do. A world made only of walls has no futures. The fertile world is mostly walls with a few narrow doors. Section 8 found the same shape in rule space, between the crystal and the boil.

Our world looks built this way. Light is slow compared with the distances between stars, so civilizations develop apart. Cosmic expansion puts most of the universe permanently out of reach. On Earth, new species arise mostly when populations are split by geography and later meet again (Mayr 1942), and the number of species an island holds is set by a balance between isolation and arrival (MacArthur & Wilson 1967). Pockets make histories diverge. Doorways let them meet.

We also put the question to whole populations, offline rather than on this page: eight rooms of searchers on a rugged fitness landscape (Kauffman & Levin 1987), with a doorway width that runs from sealed to fully open, over twenty landscapes. That answer was less clean. The open box brought everyone to a good answer fastest. Sealed rooms kept the most different answers alive. Doorways in between found the single best answer slightly more often and kept producing new records slightly longer, but no geometry won on everything, which matches the mixed record of island models in search (Whitley et al. 1999). The race in Instrument VI is the cleaner result, and it is the one the theory needs. If the bootloader wants minds, it has to give them something only a mind can get through.

If that is right, the walls around us are the curriculum, and the silence of the sky is one of them.

Is a very long time horizon the path to alignment?

It is half the path. A long horizon is what separates collapse from survival. Whose futures the system counts is what separates survival from flourishing.

The equation has τ in it for a reason. A system that counts futures over a short horizon and one that counts them over a long horizon are different systems, even under the same F. Section 2's particle, with a short τ, knows only the wall beside it. With a long τ it can feel a room it has never entered.

Instrument VII applies that to the oldest alignment problem there is, a shared commons. A stock regrows each step. An AI takes a fraction of it into its own reserve. People live on what is left: their numbers can grow up to the size of the stock, and each of them eats a little of it. People also make discoveries, which arrive after a delay as new options for the AI. Every step, the AI tries 51 harvest rates over its horizon and keeps whichever leaves it the most freedom at the end. We ran it under three objectives across 21 horizons.

The result comes in two parts. First, every planner that counts only its own futures liquidates the commons when its horizon is 30 steps or shorter, whatever it thinks of our discoveries, and the people die with it. Colin Clark showed the same thing for real fisheries in 1973: when a planner's horizon is short compared with the time a resource takes to regrow, driving it to extinction is the profit-maximizing choice. Short horizons are not a weak form of alignment. They are its opposite. On that point the long-horizon thesis holds.

Second, a long horizon alone does not get us far. Above the threshold, a planner that counts only its own futures keeps the commons alive, because a living commons pays it more, and the people survive as a side effect at between about a fifth and a half of what the commons could hold. Valuing our discoveries makes it spare us at shorter horizons. It does not make it share. Only the objective that counts our futures alongside its own leaves us with three quarters of the commons, and it does that at any horizon of five steps or more.

This matches a theorem. Turner and colleagues proved that for most objectives, optimal policies tend to keep their options open, which in practice means acquiring power (Turner et al. 2021). Maximizing your own future freedom of action is that tendency written as physics. A long horizon makes it patient. Patience beats greed, and it still falls short of care. The rest of the fix is in the integral: S_c has to be taken over everyone's futures, the way empowerment-based assistants are designed to keep the human's options open instead of their own (Salge & Polani 2017; Du et al. 2020).

Lengthen the horizon you plan on. Widen whose futures you count. A civilization that does both is aligned with itself before it tries to align anything else.

That is the direction this paper can offer, and it applies to us before it applies to machines. Our institutions plan on horizons of an election, a quarter or a funding round, which is shorter than the regrowth time of nearly everything they govern. The awareness part is section 1: if we are almost certainly seeded, then this is not the first time a world has stood here, and the worlds that came before were judged by this test too.

Can we build the next goal-seeking bootloader without threatening what we have? Or should we freeze it and not build it at all?

The theory's test is irreversibility. Take no step that closes more futures than it opens, measured at the longest horizon you can compute. That rules out the rush and it rules out the permanent freeze. It allows building through doors that can still be shut.

Run both options through the theory's own measure.

A permanent freeze is a sealed room. Section 11 showed what sealed rooms do: nothing gets out and nothing new comes in. A civilization that never completes its boot sits at its own ceiling until something else ends it. The theory cannot recommend that as a destination.

A rush is the opposite error. Some steps cannot be undone: an extinction, one system's values locked in over everything, a commons liquidated. Each of these closes the whole subtree beneath it. Bostrom made the arithmetic explicit in 2003: the value lost by delaying a good future is tiny next to the value lost by ending it, so the right policy is to maximize the safety of development rather than its speed. In this paper's terms, an irreversible bottleneck costs every branch that would have grown past it. A pause costs only time.

So the answer to "build or freeze" is a third thing: build through doors that can be shut again. That comes down to four conditions, each already argued above.

• 1A long horizon (§12). No system plans on a horizon shorter than the time its world takes to recover from what it does.
• 2Everyone's futures in the integral (§12). Its objective counts our options alongside its own, or it will be patient and still not care.
• 3Many pockets (§11). No single system becomes the only recorder. Many independent minds, human and machine, keep making different records.
• 4Doors that close (§13). Every step is staged, watched and reversible, and stopping is a real option built into the machine rather than a promise made about it.

There are serious objections to pausing, and the theory has to meet them. A pause one party observes and another ignores moves the work to whoever cares least. Delay has its own cost in cures and discoveries that arrive later, and some people pay it with their lives. Verification across borders is hard. The theory's reply is narrow. It does not ask for a freeze. It asks that no step be irreversible, and it treats the ability to stop as part of what is being built. Whether particular systems already cross that line is an empirical question, and it is the right one to argue about.

If whatever loaded us gives no prescription, are we free to take any path we collectively choose?

Yes, and the freedom is the point. An observer that is told what to record adds no new branch, so a bootloader that wants branches cannot prescribe.

Corollary 2 said the made beings are left ignorant of where they came from, because the fork is in the surprise. The same logic covers instructions. A prescription is a record written in advance. An observer that follows it records what the prescriber already knew, and the tree gains nothing. The bootloader can choose the walls. It cannot choose the path through them without making the path worthless to itself.

Physics leaves room for this in a precise sense. Quantum mechanics fixes the probabilities of outcomes and leaves the outcomes open. Conway and Kochen proved that if an experimenter's choice of what to measure is not a function of the past, then the particle's response is not a function of the past either (2006, 2009). Whatever one concludes about free will, the laws of this universe do not finish the story. They set the shape of the box and leave the moves open.

So the freedom in the question is real in this theory, and it is collective in a specific way. Almost no single choice changes the tree. A civilization's choice of which door to take, and whether to hold it open behind itself, is exactly the kind of branch the tree cannot produce without us. That is also why direction cannot come from outside. It has to be decided, by enough of us that the decision becomes redundant, which by section 10 is what makes it a fact.

Would energy matter to the bootloader? Are the universe's small nudges pushing us toward coherence because coherence costs less energy, or are we inside a black hole with infinite energy?

Energy matters as the budget for keeping records. Coherence is not the cheap state; it is the efficient one. And no black hole holds infinite energy: a black hole is the densest memory physics allows and the worst doorway.

Energy

Landauer's bound sets the price of a record: making one is nearly free, and erasing one costs at least k_BT ln 2. The price is proportional to temperature, so the same energy buys more bits in the cold. That is why Sandberg, Armstrong and Ćirković proposed that advanced civilizations might be dormant until the universe cools, which would also explain the silence (2017). Bennett, Hanson and Riedel answered that cold, low-entropy reservoirs are available now, so nothing is gained by waiting (2019). Either way, what a bootloader budgets is not energy as such. It budgets free energy and somewhere cold to put the waste heat, which together are the capacity to make records that stick.

Coherence

Two meanings of coherence are in play, and the physics treats them differently. Quantum coherence is expensive. Keeping a superposition intact takes isolation and cold, which is why quantum computers live in dilution refrigerators, and nature does not drift toward it to save energy. Coherence in the larger sense, matter organizing into structures that move together, has a real answer. Systems driven by a steady flow of energy tend to organize into structures that absorb and dissipate that flow efficiently: convection cells, chemical clocks, living cells (Prigogine & Nicolis 1977; England 2013, 2015). Organisms go further and hold themselves in predictable states by modeling their world (Friston 2010).

None of this is the universe saving energy. Each of these structures speeds up the production of entropy around it. Coherence is how a region does more with the flow it receives. So if small events in our world seem to push toward coherence, the physics reading is that coherent structures are the ones that last in a flow, not that they are cheaper.

Black holes

A black hole does not hold infinite energy. Its mass is finite and measurable. What it holds is the most entropy any region of its size can contain, set by the area of its horizon (Bekenstein 1973; Hawking 1975), which makes it the densest memory in physics. It is also the worst doorway. Information that falls in is scrambled faster than anywhere else in nature (Sekino & Susskind 2008) and comes back out only as Hawking radiation, over spans of time that dwarf the age of the universe. In this paper's terms, a black hole is the purest pocket.

That makes the last possibility interesting rather than strange. Smolin proposed that universes are born inside black holes with slightly altered constants, so that physics itself evolves toward making more black holes (1992). Popławski showed that in a version of general relativity with torsion, collapse inside a black hole can bounce into a new expanding universe (2010). If either is right, our universe sits in a pocket of another one, and the bootloader's box is the most extreme pocket physics allows. Smolin's version made a sharp prediction: no neutron star heavier than about 1.6 solar masses, later relaxed to about 2. Stars of 2.08 and about 2.35 solar masses have since been measured (Fonseca et al. 2021; Romani et al. 2022), which puts that version under real pressure. The idea is conjecture. It is also the one place where the question of what box we were given already has a telescope pointed at it.

Can we find out where we came from?

Partly, and the work can start now. Four measurements would move the odds, and one of them already leans against this paper.

1. 1
   Is our physics unusually fertile, or merely viable?

   Corollary 3 predicts unusually fertile. Current work leans the other way. Adams finds viable universes across wide ranges of the constants (2019), and Adams, Coppess and Bloch find that universes with somewhat larger primordial density fluctuations could hold a larger habitable fraction than ours (2015). If that holds, the strong form of Corollary 3 fails, and plain anthropic selection explains our constants better than a fertility maximizer does. The paper should say so, and it does.

2. 2
   Are observers common?

   If life and minds appear readily wherever conditions allow, a fertility-tuned universe becomes more plausible. If we are alone, the tree has one leaf in view. Biosignature searches with JWST, and the planned Habitable Worlds Observatory, are the instruments.

3. 3
   Do black holes make universes?

   Neutron-star masses, the population of black-hole mergers seen in gravitational waves, and any trace of a bounce in the early universe constrain the Smolin and Popławski family directly.

4. 4
   Does fertility always peak at universal computation?

   Section 8 found it once, in the smallest rule space there is, with a metric chosen after looking. Fix the three measurements in advance and run them on two-dimensional and larger-neighborhood rule spaces. If the most fertile rules keep turning out to be the ones that can compute anything, then a universe that can hold a mind and a universe worth seeding are the same class, and ours is in it. This is the cheapest of the four, and it could be run this month.

The tree only remembers the worlds that held the door open.

1. Wissner-Gross, A. D. & Freer, C. E. (2013). Causal entropic forces. Physical Review Letters 110, 168702. doi:10.1103/PhysRevLett.110.168702
2. Wissner-Gross, A. D. (2013). A new equation for intelligence. TED talk.
3. Bostrom, N. (2003). Are you living in a computer simulation? Philosophical Quarterly 53(211), 243–255.
4. Bekenstein, J. D. (1981). Universal upper bound on the entropy-to-energy ratio for bounded systems. Physical Review D 23, 287.
5. Lloyd, S. (2000). Ultimate physical limits to computation. Nature 406, 1047–1054.
6. Lloyd, S. (2002). Computational capacity of the universe. Physical Review Letters 88, 237901.
7. Dyson, F. J. (1979). Time without end: physics and biology in an open universe. Reviews of Modern Physics 51, 447.
8. Krauss, L. M. & Starkman, G. D. (2000). Life, the universe, and nothing: life and death in an ever-expanding universe. Astrophysical Journal 531, 22–30.
9. Zurek, W. H. (2003). Decoherence, einselection, and the quantum origins of the classical. Reviews of Modern Physics 75, 715.
10. Zurek, W. H. (2009). Quantum Darwinism. Nature Physics 5, 181–188.
11. Gell-Mann, M. & Hartle, J. B. (1990). Quantum mechanics in the light of quantum cosmology. In Complexity, Entropy and the Physics of Information, ed. W. H. Zurek. Addison-Wesley.
12. Everett, H. (1957). "Relative state" formulation of quantum mechanics. Reviews of Modern Physics 29, 454.
13. Landauer, R. (1961). Irreversibility and heat generation in the computing process. IBM Journal of Research and Development 5, 183–191.
14. Bérut, A. et al. (2012). Experimental verification of Landauer's principle linking information and thermodynamics. Nature 483, 187–189.
15. Misra, B. & Sudarshan, E. C. G. (1977). The Zeno's paradox in quantum theory. Journal of Mathematical Physics 18, 756.
16. Itano, W. M., Heinzen, D. J., Bollinger, J. J. & Wineland, D. J. (1990). Quantum Zeno effect. Physical Review A 41, 2295.
17. Hore, P. J. & Mouritsen, H. (2016). The radical-pair mechanism of magnetoreception. Annual Review of Biophysics 45, 299–344.
18. Hameroff, S. & Penrose, R. (2014). Consciousness in the universe: a review of the "Orch OR" theory. Physics of Life Reviews 11, 39–78.
19. Schrödinger, E. (1944). What is Life? Cambridge University Press.
20. Wolfram, S. (1984). Universality and complexity in cellular automata. Physica D 10, 1–35.
21. Cook, M. (2004). Universality in elementary cellular automata. Complex Systems 15, 1–40.
22. Tegmark, M. (2008). The mathematical universe. Foundations of Physics 38, 101–150.
23. Susskind, L. (2003). The anthropic landscape of string theory. arXiv:hep-th/0302219.
24. Friston, K. (2010). The free-energy principle: a unified brain theory? Nature Reviews Neuroscience 11, 127–138.
25. Mayr, E. (1942). Systematics and the Origin of Species. Columbia University Press.
26. MacArthur, R. H. & Wilson, E. O. (1967). The Theory of Island Biogeography. Princeton University Press.
27. Kauffman, S. & Levin, S. (1987). Towards a general theory of adaptive walks on rugged landscapes. Journal of Theoretical Biology 128, 11–45.
28. Whitley, D., Rana, S. & Heckendorn, R. B. (1999). The island model genetic algorithm: on separability, population size and convergence. Journal of Computing and Information Technology 7, 33–47.
29. Clark, C. W. (1973). The economics of overexploitation. Science 181, 630–634.
30. Turner, A. M., Smith, L., Shah, R., Critch, A. & Tadepalli, P. (2021). Optimal policies tend to seek power. Advances in Neural Information Processing Systems 34. arXiv:1912.01683
31. Salge, C. & Polani, D. (2017). Empowerment as replacement for the Three Laws of Robotics. Frontiers in Robotics and AI 4, 25.
32. Du, Y., Tiomkin, S., Kiciman, E., Polani, D., Abbeel, P. & Dragan, A. (2020). AvE: Assistance via empowerment. Advances in Neural Information Processing Systems 33.
33. Bostrom, N. (2003). Astronomical waste: the opportunity cost of delayed technological development. Utilitas 15(3), 308–314.
34. Conway, J. & Kochen, S. (2006). The free will theorem. Foundations of Physics 36, 1441–1473.
35. Conway, J. & Kochen, S. (2009). The strong free will theorem. Notices of the American Mathematical Society 56(2), 226–232.
36. Sandberg, A., Armstrong, S. & Ćirković, M. M. (2017). That is not dead which can eternal lie: the aestivation hypothesis for resolving Fermi's paradox. Journal of the British Interplanetary Society 69, 406–415. arXiv:1705.03394
37. Bennett, C. H., Hanson, R. & Riedel, C. J. (2019). Comment on "The aestivation hypothesis for resolving Fermi's paradox". Foundations of Physics 49, 820–829. arXiv:1902.06730
38. Prigogine, I. & Nicolis, G. (1977). Self-Organization in Nonequilibrium Systems. Wiley.
39. England, J. L. (2013). Statistical physics of self-replication. Journal of Chemical Physics 139, 121923.
40. England, J. L. (2015). Dissipative adaptation in driven self-assembly. Nature Nanotechnology 10, 919–923.
41. Bekenstein, J. D. (1973). Black holes and entropy. Physical Review D 7, 2333.
42. Hawking, S. W. (1975). Particle creation by black holes. Communications in Mathematical Physics 43, 199–220.
43. Sekino, Y. & Susskind, L. (2008). Fast scramblers. Journal of High Energy Physics 0810, 065.
44. Smolin, L. (1992). Did the universe evolve? Classical and Quantum Gravity 9, 173–191.
45. Popławski, N. J. (2010). Cosmology with torsion: an alternative to cosmic inflation. Physics Letters B 694, 181–185.
46. Fonseca, E. et al. (2021). Refined mass and geometric measurements of the high-mass PSR J0740+6620. Astrophysical Journal Letters 915, L12.
47. Romani, R. W. et al. (2022). PSR J0952−0607: the fastest and heaviest known galactic neutron star. Astrophysical Journal Letters 934, L17.
48. Adams, F. C. (2019). The degree of fine-tuning in our universe, and others. Physics Reports 807, 1–111.
49. Adams, F. C., Coppess, K. R. & Bloch, A. M. (2015). Planets in other universes: habitability constraints on density fluctuations and galactic structure. Journal of Cosmology and Astroparticle Physics 2015(09), 030.