Chimps beat humans at game theory


#1

[Permalink]


Oligopolistic America: anti-competitive, unequal, and deliberate
#2

Well I’ll be a monkey’s student.


#3

Some call it primatology. I call it Viral Movie Marketing.


#4

Like 1999, still a chimp’s party on Wall Street - Paul B. Farrell - MarketWatch

Raven the chimp is now Raven the Champ! In the 1999 race for top performance honors, the chimp made a monkey out of every darn (so-called) “professional” money manager on Wall Street and everywhere in America. His Monkeydex Index beat the best-of-the-best of all mutual funds run by America’s top managers.

#5

Once again, a completely hyperbolic and meaningless headline by doctorow.


#6

Yeah.

I think it’s interesting that people seem to read this experiment as “chimps better than humans at game theory”, and not “game theory fails to describe or explain behaviour of actual humans even in laboratory settings, only succeeds for stupid animals”.


#7

In fact, look at the supplementary material. The thing to note is that this is not a zero sum game. By deviating from the nash equilibrium, the human participants actually achieve a higher monetary payout from the experimenters than the ‘optimal’ choice would give them, far superior to what the chimps manage.

Chimps are not smarter than humans. The humans in the experiment were smarter than the researchers.


#8

I totally disagree. The headline definitely tells us what we can expect to see in the content.

But I agree totally with @Fang that the fact that chimps do better at game theory than humans in these games should make us ask ourselves what game theory is really describing.

The chimps found the better “seflish” solution and the humans found the better “community” solution. Chimps maximize the output for the chimp, humans maximize the total output of the game. We all know that if you want to “win” against your “opponent” that the chimp strategy is better, and that’s what game theory analyzes. However, if you want to become the dominant species on the planet and master almost all aspects of your environments, the human strategy may have some advantages.


#9

Not just chimps.

http://www.theguardian.com/money/2013/jan/13/investments-stock-picking


#10

No, the chimps also failed to maximise individual payouts. Both participants in the human experiment scored more highly (or at least equivalently) than the chimps did, and higher than the Nash equilibrium allows. This isn’t about aggregate utility, the humans beat the chimps hands down.

The point of the very very commonly misunderstood Nash equilibrium is that it’s the optimum stable solution, being stable under the possibility of one or the other participant acting selfishly to betray the other for momentary gain. But things like trust and co-operation allow participants to sustain a theoretically intrinsically unstable strategy that produces benefits for everyone.


#11

Is that video part of the game-theory paper? I can’t tell. (There’s an image of a similar game in the paper, but the rules look different.)

In any case, can someone explain how the chimp is doing what it’s doing in the video? The numbers 1-9 are covered up by squares, and the chimp instantly appears to see all the numbers, remember them, and touch the squares in the right order.

It must being doing something different, though, because such a feat seems almost impossible even for a human, at the rate that he sees and covers the numbers. Half the time he doesn’t even seem to look across the whole screen.

Are these pre-memorized sequences?


#12

Yes, you caught me using the language of a way of thinking that doesn’t work very well in reality. I said “Chimps maximize the output for the chimp” and what I really should say is that chimps assumed maximized their output under the assumption that the whole world was out to get them. Humans did a much better job of actually getting more things for themselves.

I still think that “chimps beat humans at game theory” is pretty much accurate. It’s just that “winning at game theory” and walking away with the greatest reward from the experiment aren’t the same thing at all. To walk away with the best reward you are better off trusting the other player rather than seeing them as an opponent, and taking a longer term view than round-to-round. It requires some psychology, not just math.


#13

It’s just a common mistake that really frustrates me. The naturalistic fallacy use of game theory is all over the place - see also ‘greed is good’. The fact that something is game-theoretic optimal doesn’t mean that it is actually optimal, or worse, that we should desire situations to become more game-theoretic optimal.

All over the internet, this study is being reported in variants of ‘chimps are smarter than humans’, when this is probably the opposite of what should be concluded. I think it’s downright irresponsible to hide the payout differences and reward structure until Page 21 of Supplementary material.


#14

And if you look at chimp society and you look at human society, this is exactly what you should expect.

As I say every time game theory comes up, empirical studies tend to show this pattern when they actually run game theory tests with human subjects. The only test subjects who do what the simple models predict are Economics students.

Human society has acquired formidable defences against “always defect” style behaviour, and this is only broken down by rigorous training in selfish oversimplification. Unfortunately, we are living in the midst of a political project to do just that.


#15

OK, lab chimps are shrewder than psych undergrads.
And your point?


#16

And even then there have been psych experiments on chimps that show that they too have some kind of built in sense of fairness and group responsibility, just maybe not as developed as ours.

Come to think of it, one has to wonder if the chimps really understood they were playing against another chimp, or if they just thought they were trying to beat a game on a tablet. Having no other “person” involved would change the way humans play as well.


#17

This topic was automatically closed after 5 days. New replies are no longer allowed.