Anthropics Explained (by Rejecting the Copernican Principle)

Many paradoxes in philosophy, like the Simulation Argument, the Boltzmann brain argument, and the Doomsday argument, rest on what’s called the Copernican Principle. This is the idea that among all of the observers that exist in the universe you are equally likely to be any one of them. There is little reason to think this is true, and these paradoxes give us a strong reason to reject the Copernican principle. Instead, the question of what observer you are should be approached like any other scientific question: be open to many theories and judge them by their parsimony and their fit with your observational data.

This is easiest to understand in the “anthropic” puzzles in eternal inflation theories. In these theories the process of inflation creates an infinite number of “bubble universes” which may have different effective laws of physics (although the rules for how these different effective laws of physics come about, as well as what happens in high-energy experiments with these bubble universes, should be determined by shared universal laws of physics). Already the mere possibility of there being an infinite number of observers kills the Copernican principle as a necessary truth, as when this occurs it is impossible weigh each observer with an equal probability. This is a practical as well as a theoretical problem. There are many proposals for how to sample bubble universes within the full universe/multiverse, and they give different predictions for what a “typical” bubble universe is like. I think such sampling procedures for where our observable universe fits in the full universe should be considered to be a component of any specific eternal inflation theory. This means that if a single inflation mechanism has multiple substantially different ways a universe like ours might be picked out these different ways should be considered different theories. It also means that a simple inflation mechanism which requires a complex sampling procedure to describe how our currently observed universe arises should be penalized as more complex.

Your knowledge of where you are in the universe is something philosophers of science call an auxillary hypothesis. No scientific theory makes predictions entirely on its own, but it must be augmented with auxillary hypotheses that relate the theory to things we can directly observe. For example, if you test Maxwell’s equation by measuring the force between a current through a wire and magnet, the test depends on the auxillary hypothesis that your battery generates a voltage, your wire conducts electricity, and your magnet generates a permanent magnetic field. An unexpected result can be explained by changing either the theory or the auxillary hypotheses. The famous example from astronomy: When Uranus’s orbit did not match predictions this lead to the discovery of Neptune whose gravity was affecting Uranus, so Newtonian gravity still worked but required a new auxillary hypothesis. When Mercury’s orbit diverged from predictions a planet Vulcan was theorized to affect it, but Mercury’s orbit was eventually explained by general relativity, a new theory. People usually don’t explictly include auxillary hypotheses when assessing the elegance or complexity of a theory, but it is necessary to do so when comparing explanations that involve changing the auxillary hypotheses. Perhaps some fine-tuned invisible Vulcan is what’s really affecting Mercury, but general relativity is the simpler theory.

Every theory about cosmology or the nature of reality implicitly includes our place in the universe as an auxillary assumption. After all, it is tested through observations we make. You expect to see a galaxy only when you’re actually looking through a telescope. In particular, if you are actually in a VR or simulated environment then what see through a virtual telescope won’t be determined by cosmology but by the will of the creator of your environment. That means that when we consider the hypothesis that everything we’ve ever experienced is simulated, then we have absolutely no direct evidence on the cosmology and laws of physics of the underlying “real” universe. These theories must be re-examined from scratch, and for the underlying universe to have the same laws and cosmology is possible but highly speculative. That is not how the simulation argument is usually presented. Instead, it’s thought that simply by taking our current cosmological model seriously a false egalitarianism forces us to take far more seriously our cosmological model existing in a simulation which in turn exists in the exact same cosmological model, in spite of the latter model being obviously more contrived. That’s because the simulation argument treats the auxillary hypothesis of self-location through a dogma rather than as an aspect of a scientific model.

I only covered two examples in depth, but I hope you get the idea. Again, every paradox that works using the Copernican principle can inverted to produce an argument against the Copernican principle:

• It looks like the future population will be a lot larger than the current population, so there will be far more observers in the distant future than all observers that ever lived up to the present day. Therefore living in the present, we are not typical observers.
• It looks like our descendants will make many simulations of us that are indistinguishable to an observer from the real thing. These simulated observers will vastly outnumber us, but unlike us, they won’t be able to make simulations in turn. Therefore we are not typical observers.
• If the universe ends in heat death, then afterwards thermal or quantum fluctuations can create conscious observers. This is incredibly rare, but if the heat death is eternal such observers will eventually outnumber those that live in a non-equlibrium universe. Most such observers only experience the bare minimum of conscious experience that counts as such, rather than the rich lives we live. Therefore we are not typical observers.
• In an infinite inflationary universe the very notion of a “typical observer” doesn’t make sense. Therefore we aren’t typical observers.

Strictly speaking, nothing here produces an actual contradiction, but it’s all evidence. The Copernican principle forces us to take many contrived positions about the past and future of our universe, when we could just not do that. Reject the Copernican principle!

Relativistic Colonization

The main inspiration for my ideas here has been Stuart Armstrong and Anders Sandberg’s paper “Eternity in six hours: Intergalactic spreading of intelligent life and sharpening the Fermi paradox”. It’s main point is to argue that intergalactic colonization of an intelligent civilization is highly feasible in a cosmic timescale, and to discuss the implications of this on the Fermi paradox. In doing so, it also discusses a particular method intergalactic colonization can occur: it argues a single starting solar system has enough materials and energy to directly send a probe to every reachable galaxy, and in turn each probe can self-replicates and sends probes to every star in that galaxy. While thinking through this scenario, I decided that there’s a more efficient and more plausible method that intergalactic colonization can occur. This does not substantially affect Armstrong and Sandbergs’ main points about the Fermi paradox. While re-examining this paper I found it responded to Robin Hanson’s paper “Burning the Cosmic Commons: Evolutionary Strategies for Interstellar Colonization”, which in many respects is closer to my picture of intergalactic colonization, and by all rights should have been an inspiration for me.

Armstrong and Sandberg were careful justifying their assumptions on the technological capability of an intelligent species, and trying to make their results robust to conservative technological assumptions. My approach is more optimistic — roughly speaking, anything that appears to be physically possible I assume to be technologically possible with enough research time. A good follow-up question to my proposal to figure out the exact technological requirements and their feasibility.

In Armstrong and Sandberg’s strategy, a single probe is created at the starting solar system and sent directly to a target galaxy. It spends most of its journey — a vast amount of time — completely inert. This is wasteful. Instead, the probe should spend that time gathering resources from the surrounding space while remaining in motion. It should use those resources to self-replicate while in transit rather than when it reaches a target star and galaxy. That way, it will be able to settle an entire galaxy at one rather than make a time-consuming second hop to colonize the galaxy. Though now there is no reason for a probe’s target to be exactly one galaxy. Instead, single probe now targets a cone-shaped region with the starting solar system as the apex.

Even if this method is more efficient, does that matter? If both strategies are more than capable of colonizing the future lightcone, isn’t it just a matter of which method the civilization chooses, rather than which one is “better”? No it is not, because the second stage for the inert probe really adds a serious delay. Imagine people implemented Armstrong and Sandberg’s proposal first, and launched a probe at 99% lightspeed to every reachable galaxy. Then, it takes ten thousand years until someone successfully launches a self-replicating-in-transit probe at the same speed to a particular galaxy. For comparison, Armstrong and Sandberg’s most pessimistic estimate is that will take eleven thousand years to launch every probe, and a more representative estimate is in the paper’s title, six hours[1]. Then the inert probe arrives to the galaxy twenty thousand years earlier, and has time to create and send secondary probes to a twenty thousand light year radius at best. Meanwhile, the self-replicating-in-transit arrives at the entire galaxy at once. If the galaxy is as large as the Milky Way, one hundred thousand light years across, then the active probe gets to most of the galaxy first. The fifty thousand years it takes the inert probe to colonize the rest of the galaxy is tiny from a cosmological perspective, which is why it was ignored in Armstrong and Sandberg, but huge from a human/historical perspective. Since we are comparing two approaches that can both be initiated by the same technological civilization as part of its historical development, it is the historical timescales that are relevant for comparing which approach wins over.

An active probe in transit won’t just gather resources and self-replicate, it can host an entire society. It can host intelligent entities, whether AIs or real or simulated humans, that are thinking and planning for their future. In particular, they can make scientific and technological advantages that will make the probe work better, either decrease its mass or energy requirements or increase its speed. This gives another reason to expect active-in-transit probes to be more successful than inert probes: If in situ research can lead to an early-launched probe accelerating from 99% lightspeed to 99.9% lightspeed then that speed difference will really add up over millions or billions of light years, beating inert probes launched earlier. It will also beat probes launched later at 99.9% lightspeed from the starting solar system due to its head start.

To make reasonable guesses on the behavior of these intelligent entities and their society, we should think about their incentives. Of all these entities, all in motion, whichever one moves fast would have first access to galaxies, stars, interstellar gas, and other natural resources. These resources have energy in a low entropy form, and both the energy and the low entropy are useful for performing computations and self-replicating. However, these resources are (relatively) stationary, and usable energy from the perspective of our fast-moving society must also have a lot of momentum. So the matter in this space must be accelerated to the speed of our society to be usable, with the remaining matter accelerated to an equal and opposite momentum. This opposing momentum will make it more difficult for any later entity from making use of these resources, especially if it’s also trying to move very fast. Moreover, due to transformation law for energy-momentum, the faster you go the more energy in stationary form is necessary to obtain the same amount of usable energy from your mobile perspective. So the faster the first movers are going, the more they’ll need to propel naturally-occurring matter in the opposite direction and make it difficult to use. So there’s a huge first-mover advantage.

This is absolutely terrible.

It’s bad because it’s saying a large proportion of the resources in the future lightcone, perhaps most of them, will be burnt for nothing. Specifically, burnt as rocket fuel to make a huge number of rockets which, due to time dilation, only exist for long enough for their inhabitants to figure out how to make the rocket go so fast. I’m not sure what will be left behind after such a process, whether there will be an absolute vacuum or a maximum-entropy heat bath or whether there will still be some sort of usable energy after this entire process. Either way, it will be a terrible loss. This is what I believe will happen if intergalactic colonization is guided by the incentives of individual colonizers.

[1] Both estimates from Table 5 of the paper.

Is There Slack in Evolution?

In a recent blog post, Studies on Slack, Scott Alexander discusses how slack (particularly the rationalist conception of it) interacts with competition in various situations, and models it in terms of two-layered competitive systems. Many of his example use evolution as the competitive system, and his very first example claims that in certain circumstances a useful adaption is more likely to occur if there is less evolutionary pressure:

Imagine a distant planet full of eyeless animals. Evolving eyes is hard: they need to evolve Eye Part 1, then Eye Part 2, then Eye Part 3, in that order. Each of these requires a separate series of rare mutations. Here on Earth, scientists believe each of these mutations must have had its own benefits – in the land of the blind, the man with only Eye Part 1 is king. But on this hypothetical alien planet, there is no such luck. You need all three Eye Parts or they’re useless. Worse, each Eye Part is metabolically costly; the animal needs to eat 1% more food per Eye Part it has. An animal with a full eye would be much more fit than anything else around, but an animal with only one or two Eye Parts will be at a small disadvantage. So these animals will only evolve eyes in conditions of relatively weak evolutionary pressure. In a world of intense and perfect competition, where the fittest animal always survives to reproduce and the least fit always dies, the animal with Eye Part 1 will always die – it’s less fit than its fully-eyeless peers. The weaker the competition, and the more randomness dominates over survival-of-the-fittest, the more likely an animal with Eye Part 1 can survive and reproduce long enough to eventually produce a descendant with Eye Part 2, and so on. There are lots of ways to decrease evolutionary pressure. Maybe natural disasters often decimate the population, dozens of generations are spend recolonizing empty land, and during this period there’s more than enough for everyone and nobody has to compete. Maybe there are frequent whalefalls, and any animal nearby has hit the evolutionary jackpot and will have thousands of descendants. Maybe the population is isolated in little islands and mountain valleys, and one gene or another can reach fixation in a population totally by chance. It doesn’t matter exactly how it happens, it matters that evolutionary pressure is low.

Is this situation realistic?

The first part that struck me is that Scott presupposed that each of these mutations occurs separately. This would make sense in ordinary evolution, where each mutation provides a benefit and reaches fixation. If each mutation is harmful, why bother going through normal competition that the organism is going to lose, and why not just have all three Eye Parts arising by chance at once? It may seem absurd that an organism would spontaneously form a fully-fledged eye, but actually all that indicates is that the entire pathway envisioned here is absurd: As creationists are so eager to point out, something as complex as a compound eye cannot just arise by chance. It requires evolution, which requires intermediate stages that are beneficial to the organism, exactly what has been positted not to occur. You can imagine we’re not actually talking about evolving eyes, but rather a much simpler adaption still goes against the flow of the fitness landscape. If you do it becomes easier to imagine that the entire adaption occurs as a single mutation at once.

But actually, Scott is right: An adaption like this would develop in stages rather than all at once. Let’s calculate this with some made-up numbers. I’ll suppose that each one of these Eye Parts has a 10-5 probability of occurring by chance. Multiply that for all three parts and you get a probably of 10-15 of all three mutations occurring at once. A mutation like that is only likely to occur in a species once 1015 organisms of that species have been born.

Let me emphasize the last sentence: The chance a mutation will occur in a population depends on the number of individuals born in that population, or in other words the total number of individual that have ever existed in the population. That is the product of the population size and the number of generations the population existed. In particular, larger populations acquire beneficial mutations faster. This will be important later.

Now, let’s compare this to developing eyes in stages. Suppose Eye Part 1 only reduces fitness by 1%. Then on average, an individual with Eye Part 1 has 1% fewer descendants, so on average 0.99 of its descendant will also have Eye Part 1. Adding up the geometric series, if an individual acquired Eye Part 1 through a spontaneous mutation it is likely to have around 100 descendants with Eye Part 1. The chance that one of them will acquire Eye Part 2 is $10^2 \cdot 10^{-5} = 10^{-3}$. If an individual with both Eye Parts 1 and 2 loses another 1% in fitness than it will have 50 descendants with both Eye Parts 1 and 2, assuming perfect genetic linkage. Then there’s a $50 \cdot 10^{-5} = 5 \times 10^{-4}$ chance that one of these descendants will acquire Eye Part 3. Multiplying these together, including the probability for the initial Eye Part 1 mutation, there is a $10^{-5} \cdot 10^{-3} \cdot (5 \times 10^{-4}) = 5 \times 10^{-12}$ chance of eyes evolving per individual of eyes evolving. This is still very small, but significantly larger than the 10-15 odds of all mutations happening at once.

In contrast, imagine if each mutation is beneficial and immediately reaches fixation. It takes 105 individuals until one acquires Eye Part 1, and immediately everyone has it. Afterwards it takes 105 individuals each to evolve Eye Parts 2 and 3, a total of 3×105. Since it takes time for traits to reach fixation and even for beneficial traits this doesn’t always occur, it is expected to take a larger number of individuals until eyes evolve, but this is still much less than the numbers required in the above two scenarios where eye parts are maladaptive.

Okay, this is the model for evolving eyes in ordinary circumstances. But what if you add slack? Let’s look at Scott’s first scenario: A disaster wipes out most of the population and leaves behind a resource-rich environment for the survivors, so even individuals with below-average fitness can reproduce above replacement and not have their lineage die out. First of all, I want to question the notion that there is less “evolutionary pressure” here, whatever that means. Less fit individuals can propagate their genes when they wouldn’t have otherwise, but fitter individuals still propagate their genes even more. If by chance a high proportion of the population right after the disaster had some fitness-lowering gene, then by the time the population rebounded the gene would be much less frequent, because the fitter individuals without the gene will repopulate faster than the less fit individuals with the gene. So it’s a matter of perspective: Are we looking at the absolute number of offspring and descendants an individual will have, or are we interested in the proportion of a trait relative to the whole population? If the former then the population boon meaningfully reduces the evolutionary pressure, if the latter than it doesn’t affect it at all.

But we’re not asking a broad qualitative question that could depend on the perspective we take, we’re asking a concrete question: Will this disaster-and-repopulation make evolving eyes more likely? Let’s think. Normally an individual with Eye Part 1 has around 100 descendants like it each which can develop Eye Part 2. During the time of plenty it can have many more than that. Sounds good?

Not so much, once you compare this with what every other individual is doing. If this individual has 100 relatedness-weighted offspring, so does every other individual during time of plenty. If there is sexual reproduction this can be tricky to think about: The first individual with Eye Part 1 has more than 100 descendant, but only some retain Eye Part 1. Other individuals also typically have more than 100 descendants. However, each descendant has many ancestors, so it’s hard to count how many descendants there should be. It’s easier to count genes: Each descendant gene comes from exactly one ancestral gene. If the spontaneously formed Eye Part 1 gene spreads to 100 descendants, then in that time every other gene should also spread to 100 descendants, or a bit more since Eye Part 1 gene reduces fitness. Since we need individuals to store all of these genes — what else are individuals good for? — the population size increases by roughly one hundred.

Now remember what I said earlier about the effect of population size on evolution? If the population increased by a factor of one hundred, that means that at the start — right after the disaster occurred — the population was one hundred times less than the ordinary stable population. That means that the mutation for Eye Part 1 was one hundred times less likely to occur in the first place! This nullifies any advantage that seems to have been gained by the plentiful conditions increasing the chance Eye Part 2 develops. Overall, a disaster that decreases population does not increase the speed the eye evolves.

Lightspeed delays lead to multiple technological singularities

By Yudkowsky’s classification, I’m assuming the Accelerating Change Singularity: As technology gets better, the characteristic timescale at which technological progress is made becomes shorter, so that the time until this reaches physical limits is short from the perspective of our timescale. At a short enough timescale the lightspeed limit becomes important: When information cannot traverse the diameter of civilization in the time until singularity further progress must be made independently in different regions. The subjective time from then may still be large, and without communication the different regions can develop different interest and, after their singularities, compete. As the characteristic timescale becomes shorter the independent regions split further.

Ribosome

My favorite example of a physical manifestation of an abstract concept is ribosome. The concept is that of making physical manifestations of abstract concepts.

You Must be a Coward to be Afraid of Losing Your Mind

I recently read two authors both make a similar kind of argument, which I disagree with.

Jacob Falkovich responds to a comment on his blog, about the insights gained from taking drugs:

Do you believe your own epistemology to be so fragile, your disbelief in woo so contingent, that they will crumble in the face of a few hours of altered consciousness?

A recent essay of Paul Graham, about having children:

On the other hand, what kind of wimpy ambition do you have if it won’t survive having kids? Do you have so little to spare?

Both excerpts are accusing the reader of a lack of confidence in her principles if she is afraid some transformative experience will change them: If are afraid drugs will make you believe in woo, you must not have much confidence in your current disbelief in woo. If you are afraid parenting will make you lose your ambition, you must not be very ambitious.

I don’t like the way these two excerpts shame being afraid of an external factor changing the way we think. My minds are fragile and I shouldn’t be afraid to admit it. A bullet can shut it down completely, and so many things can affect it in subtler ways I don’t understand.

For comparison, I don’t think either author would argue this way about more frequent and short-term events that change our motivation and reasoning. If someone worries that he’s unmotivated when he works at home, you don’t tell him that it’s not a problem if he cares enough about working, you tell him he shows he cares about working by setting up the environment where he can be motivated.

Linear Logic, Topos Theory, and Algebraic Geometry

Note: The stuff below is some idle thought without much backing to it and extremely speculative. Don’t take it seriously.

Something seems right about linear logic: Both classical logic and intuitionistic logic embed in it, and although sequent calculus loses duality symmetry going from classical logic to intuitionistic logic this symmetry is regained in linear logic. It seems to me like intuitionistic logic is more fundamental than classical logic, and although there’s no good theory for how to fully do mathematics in linear logic (as Bishop did in intuitionistic logic), I suspect that linear logic is more fundamental still.

The category of sheaves over a topological space or a site is a model of intuitionistic set theory. Conversely, given a model of intuitionistic set theory we can try to interpret it as being a category of sheaves over some kind of space. We can construct a dictionary of logical equivalents of geometric concepts and vice versa. This is the core idea of topos theory. If linear logic is more fundamental than intuitionistic logic then a natural generalization of this idea is to geometrically interpret models of linear dependent type theory to get a more general notion of space. As far as I’m aware a good theory of linear dependent type doesn’t exist yet.

A speculative idea I have for how to apply this is that it can reconcile to conflicting intuitions in algebraic geometry: On one hand, algebra laws over a field say that if $a \neq 0$ and $b \neq 0$ then $a b \neq 0$. For an arbitrary commutative ring $R$, even though this statement is no longer true it is still sometimes useful to think of it as holding in some “local” sense. For example, suppose $R$ is a ring of polynomials of some variety $X$. For each element $x \in X$ and $f, g \in R$ the evaluations $f (x), g (x)$ are elements of a field, and so if $f (x) \neq 0$ and $g (x) \neq 0$ then $f (x) g (x) \neq 0$. It follows that for any ideal $I$ in $R$ if $V (I) \subseteq D (f)$ and $V (I) \subseteq D (g)$ then $V (I) \subseteq D (f g)$. Finally, by Nullstellensatz this is the same as saying that if $I + (f) = I + (g) = R$ then $I + (fg) = R$, an purely algebraic fact that can be proven for an arbitrary commutative ring $R$.

On the other hand, scheme theory makes use of nilpotent elements of a ring and these elements contain significant geometric information. That is, we cannot say that $x^2=0$ is equivalent to $x=0$ since the schemes corresponding to these equations are different.

Perhaps this can be reconciled if we work in a logic where $P \land P$ is not equivalent to $P$. Then we can say that $x \neq 0 \land x \neq 0$ implies $x^2 \neq 0$, but the former is not implied by $x \neq 0$. Indeed we do not have such an implication for the multiplicative conjunction $\land$ of linear logic. So maybe the right way to think of a scheme with nilpotent functions over it is as a space where the logic over the space is intrinsically linear.

A somewhat natural average-case NP-complete problem

A sampling NP problem is determined by an NP language $L$ and distribution $D$ that can be sampled in probabilistic polynomial time. The problem is to determine for an $x$ sampled frome $D$ whether $x \in L$. This is a model for the average-case complexity of an NP problem. It is a little known fact that there are “sampling-NP-complete” problems in a certain sense (definition here). Such problems are rarer and less likely to arise naturally than NP-complete, and so are less well known. An example of such a problem is: Given a Turing machine $M$, sampled by an exponential distribution on codes for Turing machines, and a uniformly random bitstring $x$, solve the 3SAT problem which is produced by running $M$ on $x$ for $|x|^2$ steps.

It strikes me that idealized forms of roguelike games such as Nethack could provide a somewhat natural example of a sampling NP complete problem, and this could give an intuitive picture of what such problems could look like. In such a game, the player explores a procedurally generated map trying to reach a particular target. The player must overcome a diverse variety of obstacles and can acquire a diverse variety of special items to help them. To make this fit the sampling NP framework assume this game is deterministic and perfect information and the player has a limited number of turns to reach the target (none of these are the case in most real roguelike games). Then in an arbitrary-sized world the player has a fixed probability of getting into this situation: The player is trapped in some region of the map with only a fixed particular set of items. The effects of these items interact in a peculiar way that imitate running a nondeterministic Turing machine with features of the map as input. If the playey can make the Turing machine accept then they can get to the next stage of the map, after which winning is routine.

Here’s a roughly outline of a more formal model that fits this schema:

Consider a finite 2D lattice of locked boxes (this generalizes easily to the infinite lattice). You have keys to boxes (0,0) and (1,0) and you want to open box (2,0). Within each box is either a white or black token, sampled uniformly. Each box also has two compartments containing a blue token and a red token, built such that opening one compartment destroys the contents of other. Finally, boxes $(x,y)$ for $x \geq 0$ each have special keys that can open any box under suitable conditions: specifically, for each such key there’s a sequence $(x_0, y_0, c_0), \dots, (x_{n-1}, y_{n-1}, c_{n-1})$ where $x_i, y_i \in \mathbb {Z}$ and $c_i \in \{\mathrm {black}, \mathrm {white}, \mathrm {blue}, \mathrm {red}\}$, such that the key opens box $(x,y)$ when given the $c_i$-colored token from box $(x+x_i, y+y_i)$ for each $i$. Tokens are reusable across multiple keys. At each box the key is sampled from some probability distribution supported on all the (countably-many) possible keys; for example, $n$ could be sampled from an exponential distribution, each $c_i$ sampled uniformly and $x_i, y_i$ sampled from a discretized Gaussian distribution. Then the following happens with finite probability: The first two boxes have white tokens and the keys $[(0, -1, \mathrm {white}), (1, -1, \mathrm {white})]$ and $[(-1, -1, \mathrm {white}), (0, -1, \mathrm {white})]$. This allows opening boxes $(0,k)$ and $(1,k)$ as long as both have white tokens. This occurs for $n$ layers granting certain predetermined keys. Conditional on these keys being just right, the black token nucleates a pattern of open boxes with $x<0$ encoding the history of a 1D cellular automaton with both random and controllable inputs. This can emulate any NP problem with any sampling procedure. An ‘accept’ state can create a pattern of open boxes that spreads to the start site and to (2,0).

Prefix-free codes and ordinals

Consider the problem of representing a number in computer memory, which is idealized as a sequence of zeros and ones. The binary number system is a well-known solution to this problem — for example, the sequence “01101” represents $0 \cdot 2^4 + 1 \cdot 2^3 + 1 \cdot 2^2 + 0 \cdot 2^1 + 1 \cdot 2^0 = 11$. But there’s a problem: You don’t expect the entire computer to just be used to represent one number; you expect it to have other things stored afterwards. So how do you tell where the number ends? If the sequence begins $01101001\dots$ does this represent the number $011_2$ or $01101_2$ or $0110100_2$?

The solution to this problem most commonly used in practice is to declare in advance a fixed number of bits that will be used to represent the number, usually 32 bits or 64 bits. For example, if we fix a 5-bit representation the $01101001\dots$ always means that the number is $01101_2 = 11$ and $001\dots$ represents other stuff in memory. This works well enough in practice, but it has a problem: The number of bits you set aside for storing the number forces an upper limit on how big the number can be. For example, you cannot store any number bigger than $2^64$ in 64 bits. If the computer ever needs to store a bigger number than can be represented with space set aside then the computer fails.

I’ll introduce some terminology. The condition that a system of representing numbers is unambiguous can be phrased formally by saying the this method is a prefix-free code. A prefix-free code consists of a set of codewords, which are sequences of bits, such that no codeword is a prefix of another. A continuing stream of bits can be interpreted as a codeword by taking an initial segment that is a codeword, and by prefix-free property at most one such interpretation is possible. Since we want this code to represent numbers we also want a map from the codewords to the natural numbers (by which I mean including zero, naturally)

Entropic Games

While trying to come up with mathematically interesting variants of the notion of a game, I’ve thought of this: What if each player was trying to maximize the entropy of their moves? You can imagine that each player has some message that she wants to transmit to some third party that can observe the move-history of the game but cannot receive any other information from the players. Then each player must balance the considerations of playing randomly to transmit more entropy at the moment versus playing moves that ensure that she’ll have a higher-entropy space of moves in the future.

For example, here are the rules for the entropic game Hot Potato: The game is played with one potato, that at each moment is possessed by one of the two platers. At a player’s turn, if she hold the potatoe she may either keep the potato or pass the potato on to the other player. If she does not have the potato she is forced to do nothing.

One strategy is for a player to give or keep the potato with even odds every time she has the potato. Then each move when she has the potato has an entropy of 1 bit, the maximum possible entropy for that move. However, this seems like a suboptimal strategy, since it doesn’t account for how keeping the potato is better than passing it: If you keep the potato you are guaranteed at a choice between keeping it and passing it next turn, but if you pass it you risk a zero-entropy move next turn if your opponent doesn’t pass it back. Thus you should keep the potato with better than even odds. However, even if keeping the potato is a better move you shouldn’t keep the potato all the time — that is a zero-entropy strategy.

And ordinary perfect-information two-player game can be made into an equivalent entropic game by giving the winning player a large entropic reward. That is, the winner is given an extra turn where she can recite a large amount of information — more than is needed to describe a game history not counting this move. Then the entropic incentive to reach a winning state is stronger than the entropic incentive to ranomize your moves before that, so players will mostly play as if they’re trying to win.

Here’s a formal model of entropic games: Suppose a player has $n$ possible moves $M_0, \dots, M_{n-1}$. The player has analyzed that after playing $M_i$ the expected entropy from all moves after that is $S_i$. Suppose she plays $M_i$ with probability $p_i$. Then the total expected entropy is

$S = \sum _{k=0} ^{n-1} p_k (\log p_k + S_i)$

Optimizing this value, we get that

$p_k = \frac {1} {Z} e^{-S_k}$
$Z = \sum_{k=0}^{n-1} e^{-S_k}$

This is a Boltzmann distribution, like the probability distribution for a system in thermal equilibrium. Now, the anticipated entropies $S_k$ can be calculated by taking expected values assuming the opponent picks her moves of the same form, and so on recursively.

Unlike ordinary combinatorial games this theory straightforwardly extends to more than two players. In ordinary game theory with three players there’s a indeterminacy with kingmaker positions, where a losing player must choose which of the two other players to make the winner. If a player faces two equally good positions in an entropic game, however, she will surely choose both with equal chances.