Bounded rationality is an important topic that I see connecting the two disciplines. Computation and worst case scenarios can be captured well through this notion. Some progress has been done, but I think there is still a lot to understand.

thanks

killing games ]]>

Some issues such as worst case analysis are tough to accept even for some CS researchers (for example those who are trying to solve hard problems quickly).

I’ll give two examples for more collaboration that I think will grow in the near future.

Game theorists are probably one bridge between economic theorists and CS theorists. Some of them in fact study “future” questions, and others put emphasis on real modeling (perhaps it depends also on the department they are working in).

Bounded rationality is an important topic that I see connecting the two disciplines. Computation and worst case scenarios can be captured well through this notion. Some progress has been done, but I think there is still a lot to understand.

But not only through game theorists collaboration can take place. A pretty new “type” of economists are market designers. For these economists if the average case is solved efficiently they are happy. CS researchers can contribute to such research. One example for such a contribution is the kidney exchange problem where algorithms that solve this hard problem for large instances have been developed by CS researchers.

My point is that specific researchers from both fields have a lot to work on together. This has been done in recent years, and will continue to take place.

]]>When talking to people who do game theory in both the economics and AI communities, I’ve encountered a great deal of resistance to the worst-case point of view. You make a good argument for worst case bounds in the context of prior-free mechanism design. What I’ve found more problematic though is the objection to worst-case hardness results. I’ve been told that the experience of many people is that Nash equilibria are easy to compute (fictitious play often converges), and NP-hard allocation problems are easy to solve (CPLEX returns solutions in seconds). They concede that there exist instances of these problems that they won’t be able to solve, but claim that worst case results like this don’t say anything about the “real world” difficulty of these problems.

Of course towards the goal of an elegant an internally consistent theory, its great to have tractable solution concepts. But I think many economists still view themselves as social scientists first. I’d love to see someone write down an argument intended for economists for why (worst case) computational tractability should be a central feature of a mechanism or solution concept.

]]>