Study Compares Brain Activity In Games Against Humans and AI

Ars Technica covers research done using an fMRI machine to map brain activity game players. The study compares brain patterns in players competing against what they think are other humans against what they think is AI. It also goes into the differences in how games affect the male and female brain. "The human brain appears to try to parse the intentions of others by engaging its own decision-making process; in short, it appears to model another person's mind by seeing what it would do if it were in that other person's skull. The three areas of the brain that the authors identify are involved, in part, in making executive decisions for that brain's owner, in addition to evaluating other people's executive decisions. So, the fact that they're busier when a person thinks they're playing another human could also be interpreted as them focusing harder on an identical decision making process."

Comments

Could have chosen a better game

[falckon]

I think it's interesting that they chose the Prisoner's Dilemma for the game - a 2 choice discrete turn game. While not everyone knows the algorithms computers use for such games, people generally consider computers to be quite good at turn-by-turn games (like chess) and should be regarded as more formidable opponents. Not to mention playing a computer at this game should provoke our minds to attempt to decipher the pattern if we believe we are playing a computer so that we can beat it.

But the main reason I find it interesting is that it is very easy to get into an always defect loop. If you opponent has been defecting every turn, what incentive is there for you to defect? In this sense playing a human is an almost random process as to when to stop defecting, and when you do you will most likely lose the turn anyways. If I were playing a human I would think less about my opponents thoughts and fall into a tit-for-tat play style (repeating the last move), starting cooperatively.

I think it would be more interesting to see the effect of thinking you are playing a computer vs a human in a game with more information. For example, in chess you may leave a piece open when playing a human if you believe your opponent will not see it given the large number of possible moves, whereas with a computer you know at least every immediate move will be considered.

Re:Could have chosen a better game

[Anonymous Coward]

The article makes it sound like their computer's strategy was just, "random." Bad procedure to begin with; consider the Milgram shock experiments; the responses were kept standard, even though they were simulated.

Further, if I knew I was playing a computer, I'd have the "always random" possibility in mind right off the bat; after 10-20 shots with no discernible pattern, I MIGHT go into overdrive attempting to analyze it, but more likely, I'd just start giving random answers myself -- or, now that I'm actually thinking about it, I'd probably just start repeating myself ad infinitum to see if the responses normalized. If I thought I was playing a human, my (in this case mistaken) assumption would be that there was a motivation at work, and that there would be a human-resolvable pattern to discover; it would take much longer for me to "give up" on the simulation.

Re:Yup, gender bias

[plasmacutter]

Where do the sex differences come in? It turns out that, if you bin the data based on sex and then perform a similar subtraction, male brains seem to be more active than their female peers, which the authors interpret in light of the female gender's reputation for being more empathetic.

.

If females were more empathic, you'd think their brain might be more focused on what the other person was thinking.

Sounds to me more like the female subjects didn't care and just weren't trying.

I remember seeing documentaries on problem solving, and people who tend to solve problems more efficiently tend to show less neural activity when faced with a problem.

The increased activity in male brains indicates more second-guessing, and possibly more error.

Empathy is a very powerful tool in guessing the potential responses of others correctly.

Computers model behaviour too

[Shrike82]

Modelling your opponents behaviour is something game AI programmers have done for some time. I studied AI for Games at University, and artificial players designed to be truely competitive (rather than for commercial games where they have to merely be challenging) usually attempt to model their opponents behaviour. In turn-based games many AI players choose moves through forecasting the future based on what they guess their oppponent will do. So in relation to the article I guess that means humans modelling AI modelling humans, and so on...

Re:With a notable exception.

[Ceriel Nosforit]

You get used to embarrassing losses. Heck, I've even lost at arm wrestling to a girl after three shots of the Sauce and... Well the rest is a blur.

Also, the conclusion of this study applies to every situation where you try to guess the motivations of others. And I would like to add, it reveals a lot about the 'accuser' when they confront you on something they think motivated you. Say for example Bob consoles Alice after her breakup with her BFF Jill, at which point Cockead C accuses Bob of trying to get into Alice's pants. This shows that Cockhead C has assumed that is what motivates B, since that is what C would have done. Provided that B isn't shocked for having been found out, his motivation being something else, he now knows that C is a terrible person.