I personally learned a lot from both projects; the first authors have put a lot of work into them (of course, all other authors too 😀). End of 🧵

However, for nonlinear games, we find that any form of (endowment) inequality is detrimental, because it renders successful coordination more difficult.

For linear games, we find that certain forms of inequality can be advantageous: the largest surplus is achieved when more productive participants receive larger endowments. This confirms previous results among western online participants, doi.org/10.1038/s415...

To explore this question, Xiaomin studied a large sample of Chinese lab participants (N>1,500) to explore contributions in public good games. We varied group size, endowments, productivities, and whether the public goods game leads to linear or non-linear rewards (using a threshold function).

The second paper, with Xiaomin Wang and Boyu Zhang focuses on an empirical question: to which extent do different asymmetries between players affect their ability to cooperate and coordinate?

In the paper, Philip gives an elegant (sufficient) condition for strategy spaces to be both best reply complete and payoff complete. His result builds on important previous work by Levinski et al, link.springer.com/article/10.1...

A space S is "best reply complete" if any strategy p in S has a best reply in S. The space is "payoff complete" if any payoff achievable against p (with an arbitrary strategy) can be realized with a strategy in S. Both notions address whether strategies outside S can outperform strategies within S.

The question is this: if some strategy p is superior in a restricted space S, how would we know whether this strategy would still perform well if we allowed for more complex strategies than those in S? To address this question, Philip introduces two notions of "complete strategy spaces".

The paper with @plaporte.bsky.social, @nikoletaglyn.bsky.social and Martin Nowak asks an important theoretical question. Repeated games allow for (uncountably) many strategies. To facilitate an analysis, researchers often study simplified subspaces. To which extent are such results reliable?

Although I'm formally a co-author of this paper, I actually learned quite a bit myself about both complex systems and multiagent learning while working on this project. Thanks Wolfram for leading this really nice effort to bring fields closer together!

Awesome news, congrats! 🙂

Thanks for making me aware of the paper! I'm wondering whether the choice of game makes the difference here (volunteer's dilemma versus prisoner's dilemma). Either way, I'll have to read the paper more closely! :-)

Thanks for making me aware! It seems to me the two studies have slightly different setups (in ours, participants keep their co-player for many rounds, and they "only" engage in two different versions of a prisoner's dilemma). But I like your setup and your results a lot!