Skip to main content

Does Evolutionary Game Theory Show That Everything We Believe Is False?


"Nothing can be known; not even this, that nothing can be known."

-- Carneades, (c. 214 – c. 129 BCE)

I refute it thus!

-- Samuel Johnson, (while 'refuting' Berkeley's subjective idealism via the act of kicking a stone)


It is well understood in the evolution of perception that not all creatures perceive the world in the same way. In fact it is a matter of common sense as well. When we play tricks on other organisms we reveal that the 'minimal concept' that they will act on is quite simpler than the concept we have. Does a frog really see a fly or do they just see a 'speedy black speck'? If we can trick them with fake black specs at various speeds, then it maybe seems like the latter. Evolution is efficient and often takes shortcuts where they are available, so it is highly likely that the frog does in fact see something more minimal than what we would conceptualize as a fly. James Gibson called these 'minimal perceptual contents' affordances. I will use the term to help the discussion along.

Regarding the topic of various creatures of natural selection perceiving the world differently, someone with Kantian epistemic intuitions might wonder whether the world really is the way we see it or if some other super-organism might look at us humans and say that we merely have simple affordances. They might think that we don't really see the 'complex objects' they see. They might think of us like we think of a cat responding to a laser pointer as if it is prey. Does this mean that we cannot have real knowledge of the physical world? Is all of our empirical knowledge undermined by the notion that we could be detecting other patterns out there? I think not.

At least one scientist thinks so. Cognitive psychologist and popular science author Donald Hoffman has advanced an argument for perceptual skepticism based on an experimental result from evolutionary game theory. In short, Hoffman argues that if we believe evolutionary biology is true and that we buy his game-theoretical models of evolution and perception, it follows that all of our empirical beliefs, even those involving space and time, are false. Evolution maximizes "fitness-payoffs" not veridicality, argues Hoffman, and so what we perceive should merely be viewed as a useful illusion that tracks reality with "probability zero". Hoffman then goes on, on the basis of this argument, to defend a non-physicalist monism which is a brand of panpsychism, called Conscious Realism. He leverages the Conscious Realism conclusion even further in an attempt to prove that there is some deeper non-physical reality that must consist of only consciousness(es).

In this note I will show that Hoffman's position is obviously self-defeating. If 'everything we believe is false' so is Hoffman's thesis that our beliefs are false because 'fitness beats truth'. Hoffman is actually aware of this objection and attempts an escape maneuver. Hoffman tries to show that Universal Darwinism implies that his result needs only to follow from a priori knowledge and so he can safely claim perceptual skepticism is true without the risk of self-defeat. This attempt ends up also being self-defeating.

First I will explain Hoffman's experiment and his resulting thesis in as simple of terms as possible. Then I will describe how it is merely old skepticism with a new mask and is straightforwardly self-defeating. I will then turn to a powerful argument (advanced by Jeffery Bagwell) against Hoffman's claim that his conclusion can be saved by an appeal to Universal Darwinism.

I will conclude by adding some commentary on how the entire kerfuffle surrounding Hoffman's bizarre views is an excellent example of the interplay between science and philosophy and conclude with some thoughts on Kantian brands of idealism generally.

Interface Games

Hoffman et al's experiments are instances of what he calls 'interface games'. They attempt to show via game-theoretic models of evolutionary competition what strategies beat others in a large number of iterations of the game. I will explain a minimal version of the game as the philosophical arguments proceed the same regardless of the detail of resolution in Hoffman's game/model.

The Rules

  • Players (animals) compete over nn squares in a grid.
  • Each square of the grid is assigned a value for 2 resources "food" and "water".
  • The resources are distributed on the grid in integer values from 0 to 100.
    • Squares are colored to represent their value to the players:
      • Red: 0-24
      • Yellow: 25-50
      • Green: 51-75
      • Blue: 76-100
    • The value of the squares is set as a Gaussian distribution around 50. This is to simulate the marginal value of "breathing more oxygen" or "overeating" etc. There is some optimum with diminishing returns on either side of it.
  • The first player to select a square in the grid reaps its benefits.
  • Once a square is chosen it is out of consideration for the other player.
  • The payoff in this game is equated with fitness.
  • Once a square is 'chosen', the player who chose it gets the resources in it and therefore the fitness.
  • When all squares are chosen, the winner is the player with the most resources.
  • One iteration of the game consists of two players facing off in a race for a grid's resources.
  • Players of a given strategy meet randomly.
  • The winner gets the another player of their strategy in the population, the loser loses one, simulating reproduction.
  • Eventually, one strategy will be driven to fixation and the other will be extinct.

The strategies as as follows:

  • Simple: perceives the total information in the game of nn bits.
    • One bit of information per square, if it is over or under resource value 50
    • 0 for a territory that is under, 1 for a territory that is over
    • In a game of nn = 50 squares, Simple has 50bits50 bits.
  • Truth: perceives the total information in the game of: bits=n×2[resources]×log2101bits = n \times 2 [resources] \times \log_2 101
    • (explanation: The number of bits needed to represent an integer ii is log2(i+1)\log_2 (i+1). We have integers up to 100. So, nn squares, 2 resources, and the above equation to give us the bits needed to represent the resources values per square.)
    • In a game of nn = 50 squares, Truth has to store about 667bits667 bits.
    • Checks all squares exhaustively in a random non-repeating pattern.


So players of either strategy meet each other at random, play the game and whoever wins gets another player, thus simulating reproduction of the winning strategy. Eventually, after enough Monte Carlo simulations a winner is declared.

Representation of bits in the environment costs time and energy. So with the same computing speed, Simple beats Truth. Why? Because according to the rules of the above game if I get to the square first, I win! So while Truth is still computing bits, Simple says "Hey, this square is over 50!" or "nope, under 50, pass!" and before Truth has their shoes laced up Simple takes the game.

In Hoffman's words:

And here is the punchline. Simple drives truth to extinction for most values of the red-green boundary, even when the cost per bit of information is small and the correlation between food and water is small. ... Evolutionary pressures do not select for veridical perception; instead they drive it, should it arise, to extinction.

Again, it should be stressed that the above is a simplified version of the actual experiment which takes other variables into account, but the result and the thrust of the point are the same. "Fitness beats truth."

Our suspicions should be aroused when Hoffman makes the leap from this to perceptual skepticism. Hoffman goes on to make claims like these:

... indeed all perceptions are fundamentally non-veridical

(from The Interface Theory of Perception)

What we call ‘reality,’ consists of an elaborate papier-mâché construction of imagination and theory filled in between a few iron posts of observation.

Conscious realism makes a bold claim: consciousness, not spacetime and its objects, is fundamental reality and is properly described as a network of conscious agents.

Conscious realism contends, to the contrary, that no physical object is conscious. If I see a rock, then that rock is part of my conscious experience, but the rock itself is not conscious. When I see my friend Chris, I experience an icon that I create, but that icon itself is not conscious.

(from The Case Against Reality, 2019)

There are numerous other methodological issues with Hoffman's experiments and conclusions. They are beyond the scope of this post. I will assume Hoffman's empirical conclusions can be granted and that his modeling methodology is sound, for the sake of the philosophical points to made here.

Look here for a good overview of the major methodological issues at play.

Old Skepticism With a New Mask

Even if we trust the leap from Hoffman's experiments to perceptual skepticism generally, there is a huge issue with the argument. If this shows that all of our beliefs are false, then how is the theory itself not false?

General perceptual skepticism has dealt with this issue for thousands of years and still there is not a good answer to the self-defeat objection. Unless Hoffman can show that somehow "all our beliefs are false" is true with the exception of just the beliefs that imply his theory, he is in the same spot.

Universal Darwinism?

Hoffman attempts to show that his position is not straightforwardly self-defeating by appealing to Universal Darwinism as being true a priori. Universal Darwinism is the thesis that:

Any information processing system that has variation, heredity, selection can undergo the process of natural selection.

If it is true a priori, Hoffman believes that this saves his perceptual skepticism from self-defeat because he has used no empirical (and therefore false) information to generate the claim. I think claiming that Universal Darwinism is true a priori is perhaps safe; we can grant Hoffman that.

However, the problem is then that Hoffman is stuck in a series of tricky positions. Jeffery Bagwell points out that either biological evolution by natural selection is an instance of Universal Darwinism or it is not. This poses a dilemma for Hoffman. If it is NOT then the theory is in trouble because we don't have any good reasons to believe that we are the product of the types of processes that lead to non-veridicality. If it IS then this perceptual skepticism is self-defeating again because we are saying biological evolution by natural selection is a true instance of Universal Darwinism and we cannot rationally say that both the biological truths that lead us to skepticism are true AND the very skepticism which is meant to undermine those truths is true.

I believe Bagwell is correct in his assessment of each horn of the dilemma being a dead-end for Hoffman. I will spell them both out here as faithfully to Bagwell's analysis as possible.

Horn 1 of Bagwell's Dilemma

If Hoffman claims that biological evolution by natural selection is not an instance of Universal Darwinism then what reason do we have to believe that Universal Darwinism applies to us? How could we truthfully know we are things which can have variation, heredity etc? If it does not apply to us then it does not follow anymore that 'all of our beliefs are false'. Hoffman can try to claim that we are a kind of thing other than a product of biological evolution by natural selection. Maybe we are just some kind of physical-substrate-free consciousness that has all the ingredients needed for Universal Darwinism? Why would we believe that though when it comes with a host of other problems? Do we have any prior credence that we are a physical-substrate-free consciousness?

Hoffman proposes his Conscious Realism (the thesis that conscious agents are the totality of what is real) and further proposes that these conscious agents have the computational resources needed to model themselves such that they could derive Universal Darwinism a priori (Conscious Agent Thesis). This is a solution looking for a problem. Even if this is the case and we can know that Universal Darwinism applies to us, then, if we lift our heads for a moment to take a breath from all of the assumptions we have made, do we have compelling reasons other than saving Hoffman's perceptual skepticism to buy the ideas of Conscious Realism and the Conscious Agent Thesis? I do not believe so. We can only show we are in fact an instance of Universal Darwinism (if all of our empirical beliefs are false) if we can show that we aren't an empirically-known thing but can still compute Universal Darwinism. The satisfaction of this conditional is not a good reason to make such outlandish claims as Conscious Realism and the Conscious Agent Thesis.

Horn 2 of Bagwell's Dilemma

If Hoffman claims that biological evolution by natural selection is an instance of Universal Darwinism then it looks like we really do have some reason to believe that Universal Darwinism applies to us and perhaps we can bring back the notions of perceptual skepticism that were motivated by the interface game. But wait a minute. If biological evolution by natural selection is a true instance of Universal Darwinism then it is true of the real world. If biological evolution by natural selection is true then our perceptions of it must be true and thus our perceptions must have some veridicality. This is also not a position Hoffman wants to be in.

Science and Philosophy

Hoffman's technical stride is swift, but his abstract reasoning is more like a slow limp. The result is traveling in circles at a dizzying speed. I believe that Hoffman's popularity, the audacity of his claims, and the philosophical nail in the coffin provided by Bagwell as detailed above, all underline important features of the interplay between science and philosophy.

Scientific discoveries can often tempt us toward leaping into crazy philosophical conclusions that may or may not be warranted. The double-slit experiment has led many to believe strange things about human consciousness (see also). Are these conclusions warranted by the scientific findings alone? Is that a conceptual question or one for future experiments? The answers to these questions are not always clear. They demonstrate the utility of philosophical thinking amidst the empirical discoveries we make.

Imagine a world of Hoffmans without philosophical critics. I strongly believe that would be a world of 'too much gas not enough breaks'. The pursuit of knowledge should be far more balanced.


Hoffman's evolutionary-game-theoretic motivations for perceptual skepticism are just a fancy way of ending up in the self-defeating position that perceptual skepticism has danced around for thousands of years. No appeal to Universal Darwinism will save the self-defeating nature of Hoffman's arguments.

There is a grain of truth in all this mess though and with it some tricky philosophical questions. Different organisms really will have different affordances. Reality will look different for a frog, a house cat, and an eagle. There is no doubt about that. Is that any reason to cast doubt on the reality of the patterns of perception or the phenomena they pick out? I think not. Perception compresses reality in a way that has served a fitness-enhancing purpose for us as products of natural selection. Science and human understanding generally use the available compressible-information (defined as information that is not random noise) in reality to cope with sensory perception. Calling it 'real' or 'not real' outside the context of a particular discovery or new way of looking at things is a fruitless endeavor of which not much sense can be made.

What makes a fly the real thing that is really there and not a speedy black dot? Absent the context of further inquiry, how are we supposed to decide the answer to that question?

"Things as they really are" outside of the context of further possible inquiry don't make much sense under the sober light of potential conditions for evaluating claims about them. I challenge all 'things in themselves' defenders to clarify exactly what meaningful challenge has been failed when any 'thing in itself' has been represented perceptually instead of being 'apprehended directly', whatever that means.