Chimps Rock at Game Theory

Economics assumes that people are rational, self-interested, lightning fast calculators. Obviously a bad assumption as we are constantly told. Chimps, on the other hand, are rational, self-interested, lightning fast calculators. That is the surprising conclusion to a great paper by Colin Camerer and co-authors. Camerer had chimps play versions of the matching pennies game also called the cat and mouse game. In the cat and mouse game each player can go left or go right. The cat wins when cat and mouse choose the same strategy. The mouse wins when they choose different strategies. In the simple version the best strategy is 50:50, toss a coin. When the payoffs change, however, the optimal strategies still involve randomization but they change in surprising and nonobvious ways.

Chimps play the cat and mouse game very well. First, the chimps converge on the Nash Equilibrium strategies. In one set of games the Nash equilibrium strategies had randomization frequencies of .5, .75 and .8 and the chimps played .5, .73 and .79. Second, when payoffs change the chimps adapt their strategies very quickly simply by observation of outcomes.

Camerer et al. also tested humans in similar games and they found that humans often deviate from NE play and they adjust their strategies more slowly when payoffs change, i.e. they learn more slowly! The only thing that Camerer didn’t do was to play humans against chimps in the same game. That would have been awesome!

If you want to understand how chimps are able to play these games so well check out this video. When you see what this chimp is doing you will be amazed!


I'd rather rock at theory of mind than at game theory.

No left-leaning documentary about economics is complete these days without one of the famous videos of capuchin monkeys playing a game where one player could take all of the food, but instead shares some of the food with another player (i.e., has an inherent sense of fairness antithetical to strict competition for resources).

Interesting that chimps are apparently more ruthless than lower monkeys.

I, for one, welcome our new capuchin monkey overlords.

It's probably neither a sense of fairness nor innate. Monkeys and people are social animals. The capuchin probably have alliances to maintain and status to preserve.

I know, I'm on left-leaning documentary about economics #4323 this year. Blah blah blah, same old monkeys, same old liberal bullshit.

In the ultimatum game (the game you are referring to), monkeys generally accepts smaller portions of the total than humans.

They have also been documented planning and implementing raids where they kill and eat a member of a neighbouring group (or individual). I couldn't find a reference for this easily because most of the hits are about chimps attacking humans (probably for good reason in a fair few cases).

They share more than the good stuff with us, but I like to think we might be better equipped than them to avoid situations like eating the neighbours when the spat goes too far.

If all I knew was this game, then I would be pretty good at it too.

Heh. That's not a chimp- it's a hedgehog!

Good thing Obama can't run for re-election, because Joe Biden would be in trouble.

Ayumu vs Joe Biden in NL Hold'em: my money is on Ayumu.

The chimp in the video is good. Of course, it has nothing to do with the cat & mouse game discussed in the preceding post.

The paper argues that one of the reasons that chimps are good at the game is that they have excellent working memories.

ray lopez has an excellent working memory

It's true, I do have an excellent working memory. That's a sign of high IQ. Also when you get beat at chess by a six year old, as I have and I'm a strong club player, you understand that 'age is just a number' and you can excel in something by using just a small part of your brain. Try it sometime.

+ many 1's and 0's

Silly title. It should be "Chimps rock at games", which is distinctly less eye-catching.

Even a chimp wouldn't end three straight sentences with exclamation points. This ain't buzzfeed y'all.

Looks to me like the chimp has something like a photographic memory. I wonder if this is a general feature of chimps. Does anyone know?

Without explicitly counting, humans can typically only recognize groups of 0,1,2,3, or "some number bigger than 3". That is, if you see 3 people in the street, you know there were 3. If you see 6 people in the street and don't concentrate on counting how many there are (or they aren't clustered in an obvious way, like three people in one huddle and three people in another) then you will only be sure that there were more than three.

For some reason, the threshold is much higher for chimps. I think around 7 or 8 before they start losing track.

Very interesting. I doubt chimps are "theorizing". It seems more likely they have evolved an instinct for this type of strategy.

The chimp in the video who plays this "concentration" type game better than any human has ever tested also seems to point to chimps having a specific cognitive advantage, though so far as I know the performance hasn't been repeated by other chimps (I don't know how hard anyone has tried to train other chimps at the game).

Both of these apparent chimp advantages over humans make great sense. That kind of memory for maps of objects in space would be a big advantage in group-hunting monkeys and group-battling other groups of chimps in rainforest canopy (besides their occasional interclan wars, chimps have frequent intraclan dominance contests that often split the clan into two groups of allies of alternate alfas). The one-on-one matching/mismatching strategy would be very useful in Chimps' frequent one-on-one dominance contests. Humans simply don't need those abilities as badly as chimps need them,

I don't like the authors' presumption that humans previously had these same abilities but lost them because they needed space in their brains to fit language. Evolution inevitably atrophies all unused abilities, period. Besides, chimp cognitive evolution didn't end at the moment their ancestors divided from ours. Chimp smarts have also been evolving in their own way for the last several million years.

"Humans simply don’t need those abilities as badly as chimps need them"

I think you may have the nail on the head with this one. It plays to the anthropological phrase "use it or you lose it"

I seriously doubt that humans lost this because they needed space in their brains to fit language..Broca's area (speech production in the brain) is not that big anyways. Also, if we needed this level of knowledge and didn't have space, those who had larger heads leading the way to more brain space would have prevailed. But brain size in humans is not linked to more knowledge/ cognitive capacity so the notion that brain size matters at all in terms of "fitting things in" is irrelevant.

The most likely explanation to why humans "dropped" any kind of cognitive advantage that chimps have is simply because humans stopped using that skill. Since we stopped using it, we lost it.

Chimps also do have the ability to communicate quite well, you can teach them sign language. The reason they can't talk is mostly because of their physical anatomy (i.e. vocal chords). They do communicate via screeching where different pitches mean different things. It is likely that if they could talk, they would...

Success at teaching chimps sign language has been pretty limited. Paraphrasing Chomsky, it may be kinda like saying humans can fly, cuz Bob Beamon traveled 30 feet through the air.

Chimps also do have the ability to communicate quite well, you can teach them sign language. The reason they can’t talk is mostly because of their physical anatomy (i.e. vocal chords). They do communicate via screeching where different pitches mean different things. It is likely that if they could talk, they would…

What Brian said. You can't teach them sign language. You can take a very small number of higher primates and grossly over-interpret every hand movement they make as a form of communication. But only as long as you keep everyone except the most credulous journalists away from them.

See, for instance, Koko or Nim Chimpsky.

Although foolishly they let some people test Nim Chimpsky:

Linguistic critics challenged the animal trainers to demonstrate that Washoe was actually using language and not symbols. The null hypothesis was that the Gardners were using conditioning to teach the chimpanzee to use hand formations in certain contexts to create desirable outcomes, and that they had not learned the same linguistic rules that humans innately learn.

In response to this challenge, the chimpanzee Nim Chimpsky was taught to communicate using sign language in studies led by Herbert S. Terrace. In 44 months, Nim Chimpsky learned 125 signs.[17] However, linguistic analysis of Nim's communications demonstrated that Nim's use was symbolic, and lacked grammar, or rules, of the kind that humans use in communicating via language. This constitutes a chimpanzee vocabulary learning rate of roughly 0.1 words per day. This rate is not comparable to the average college-educated English-speaking human who learns roughly 14 words per day between ages 2 and 22.

Nim's use was symbolic, i.e. he wasn't learning language at all. Just that if he pressed the button he would get a reward.

The R^2 from regressing IQ on brain size is around 0.3. That's non-trivial.

Brains are very costly. Not only do they burn a ton of calories, but the large brains of humans make childbearing much more dangerous for mothers.

It makes sense that when cognition for other tasks becomes more valuable (the authors said language, but long-term planning and social cognition are other areas humans excel), this will both drive up brain size and drive down real-estate allocated to these cognitive tasks.

It is certainly more likely that we evolved different features rather than the second group having first evolved in and then lost it, in the absence of additional information that we should believe that this is a common feature among descendants from some common ancestor.

It's all about incentives.

The chimp gets food when he gets the numbers right.
We don't know if he has any other source for food. (What would happen, if he doesn't perform any more? Might he starve to death? We don't know - and maybe he doesn't either but doesn't want to take that risk.)

Imagine how the woman in the video would perform in this number game, if her life depended on it. Surely she wouldn't play it quite so casual.

(I don't want to discard the chimp's performance, of course. It IS quite impressive, no matter what the incentives are.
Obviously, he must have some sort of photographic memory).

I don't think that people fail at these types of games because they aren't trying.

It's not that chimps are so smart; it's that humans are so dumb! :-)

An interesting thing about Chimpanzees is that no chimp is allowed to take food off another. It doesn't matter what the difference in social status is, if the lowest ranked chimp has found something nice to eat then even the dominant male has to beg for some, the same as any other member of the group. Humans, wolves, etc. tend to be much more hierachial when it comes to the distribution of food. Chimps do share food and they keep careful track of who they share food with and who reciprocates. So a chimp that never shares her food will end up with no chimp ever sharing their food with her. This is one possible reason why chimps are so good at these sorts of games. They are constantly playing them in their daily lives while among humans food sharing tends to be more formally arranged.

That's interesting, I'd never heard they had a concept of property rights.

"When you see what this chimp is doing you will be amazed!"

Buzzfeed called, they want their headline back.

Another fail ape study

Humans weigh the chances naturally, this allows them to take advantage of low probability events. This also naturally diversifies the risk. An example might be two food sources, one of which is good (works 70% of the time) and one of which is OK (works 30% of the time). Diversifying here lowers the return, but increases survivability by allowing you to get some return when the 'better one' is doing bad. It lets you quickly change strategy if the probabilities change. Further when you consider competition against others for that same food, it makes more sense. If everyone takes the better one, the returns there will decrease. This behavior also helps explain why some labor is always reserved for gathering, even when hunting is preferred. Gathering is low risk.

This explains every 'humans are bad at probability' experiment ever.

Of course what you are saying is completly irrelevant because here we have chimps who play zero sum games between them, while you are discussing the opportunity of mixing the sources of food , which is completly another thing.

So I don't know this game, and didn't recognize any strategic play when I watched the video. I'll read more about it tomorrow. But was just wondering if someone could generalize a little why they suspect the chimp plays this game so well? Alex mentioned the authors suspect they have an excellent working memory and that that could explain it. But if so, are there other games that depend on working memory to play successfully / correctly (Nash equilibrium)? I'm curious whether they reach the Nash often, or if its an artifact of this particular class of games?

The video is a different game.

I want to see humans of increasing age play that game versus another similar but different game.

The chimp is thinking if she can just master this stupid game, her cage will open and she will finally be able to get at her handlers and rip their arms off and beat them to death with them.

So you haven't beaten the game yet, have you?

It's interesting that our cousins understand this but not (last I heard) utility gains from quantity (i.e. if they like blueberries better than strawberries, they will prefer any number of blueberries to any number of strawberries) which leads to efficiencies from trade that may be a big reason why humans are the dominant species.

This study is interpreted entirely wrongly. Chimps achieve a result closer to the Nash equilibrium. But look in the supplementary material: humans achieve *higher payouts* than the Nash result, for *both players*. In other words, humans find a co-operative strategy that the chimps fail to locate.

Is the result really that chimps are better than humans at game theory, or that simple game theory is a poor descriptor of human behaviour, which can attain results superior to what the theory dictates?

Comments for this post are closed