Hacker News Reader

Comment by rubidium | Hacker News Reader

"Given a longer sequence of repeated interactions, it might be possible that members of a hierarchically organized group would partially overcome mistrust of higher ranked individuals in those that establish generous reputations and these individuals might be able to elicit more contributions from their lower ranked partners. "

To be clear about the implications of this study: Society = longer sequence of repeated interactions. This study is not relevant to thinking about organizing society. It acknowledges it has zero relevance to whether hierarchy is good or bad for human society.

View on HN

My research group is currently writing up the results for a study where we experimentally studied cooperation that took place over an entire month. We found the very interesting result that cooperation was basically maintained at a high level instead of declining over the course of a 1-hour lab session, as predicted by much previous work.

On a related point, this experiment is a little worrisome because each player only had "9 rounds" to interact with others due to the concern of extended reputation effects. In my experience running this type of experiment, that is not even enough to understand the rules within which people play. (In our case, we had people play 400 ten-round games, 4000 rounds total, and a steady state was reached after about 60-80 games.)

While not completely an "organized society", it does address the generalization limitations of behavioral labs in one important way: the time and span of participation. As you mentioned, this is one fundamental respect where society differs from experiments.

That's why I'm personally fine with doing surveillance without consent for research purposes - observing people behaving naturally seems to be the only way to get results that are not totally messed up by constraints of experiment design. People behave differently when they know they're being studied, doubly so when they know what they're being studied for and triply so if they're being told to "act normal".

You didn't see any horizon effect? People didn't become more likely to "cheat" as the game and the value of their reputation approached an end?

We did see that effect! But, it was pretty minor. In fact, we saw all sorts of effects that would have been impossible to observe in a traditional lab experiment.

One other interesting effect we saw was a day-to-day restart effect: just leaving and coming back the next day was enough to raise the amount of cooperation a significant amount. It's not detectable at the individual level, but in aggregate it's as if going everyone going about their life for a day was enough to instill some new optimism about cooperation at the beginning of a new session than at the end of the day before.

Just that last observation returns to the original point that was made: is it even reasonable to draw conclusions about real-life cooperation from these highly clustered, short-term studies in a lab? We're trying to show that the "virtual lab" on the Internet is a way to get around these constraints.

If you're doing research in this area, check out our OSS projects: https://github.com/VirtualLab

That's a very interesting project. The field desperately needs an alternative to z-tree and psychtoolbox. JS web app seems like a great solution!

I' wrapping up my PhD in neuroEcon at USC. I've started a company to help developers use neuro/econ/psych findings to improve their apps. Let me know when you're done with your post-doc. I might have some applied research questions you'd find interesting.

Interestingly, I think "the field" now refers to an intersection of not just economics and psychology, but also sociology, computer science, and others. More and more behavioral research is going to be done on the Internet and this is going to play a role in designing socio-technical systems.

Although it might sound like an oxymoron, I think computer scientists are going to be both best equipped for and have the most reason to study digital human behavior.

I totally agree. Behavioral science needs an applied branch, and computer scientists are going to have a big role to play there.

Exactly, the paper also expresses hopes that people continue to study of human social interaction:

>> Moving forward, experiments in artificial social contexts like ours appear to be a very powerful tool to examine strategic behavior in socially relevant situations. We hope that our work will stimulate further work along these lines.

It is clear it will be nearly impossible to account for all the factors in one controlled experiment. Note that such "total experiments" in the other scientific fields is challenging as well. Usually one first tries to understand certain fundamental processes, then they move on to the interactions between them.

Reply on news.ycombinator.com