Ep. 467 Haman and Murphy Discuss the Weaknesses of Game Theory
Adam Haman returns for another crossover. In this episode, they give the brief history and a critique of formal game theory.
Mentioned in the Episode and Other Links of Interest:
- The YouTube version of this conversation.
- This episode’s sponsor, The Swan Brothers.
- Bob’s article on A Beautiful Mind.
- The HamanNature substack.
- Help support the Bob Murphy Show.
About the author, Robert
Christian and economist, Chief Economist at infineo, and Senior Fellow with the Mises Institute.
Interesting episode … I’m behind on my listening but just go to this one today.
Many comments come to mind. I was surprised to hear Adam say, “I just don’t really see how computers fit into this. Game theory is about people making choices, not machines.” Maybe he intended to be deliberately obtuse as an attention getting device, or maybe he has a grudge against computer poker players … but computers are, after all, decision making engines. Sure, computers aren’t drop-in replacements to human beings, and maybe never will be … but that’s not a fundamental distinction … only a difference in advantages and disadvantages. Computers are much faster than humans, at certain kinds of decision making … while humans tend to be better at strategic long-term thinking.
At any rate, the “big deal” about computers in game theory started back around 1980 with Robert Axelrod, running an open tournament which was played out as an agent-based simulation with each game strategy being coded into a program. Bob Murphy makes some allusions to this tournament but doesn’t cite a proper reference, and misses some details. There’s a description here – https://cs.stanford.edu/people/eroberts/courses/soco/projects/1998-99/game-theory/axelrod.html
Axelrod went on to write “The Evolution of Cooperation” published in 1984, which was hugely influential … and has triggered many followup tournaments and all sorts of agent based modelling research. And the thing is, you can’t even attempt this avenue of research without computers. These simple strategy programs are not really intended to replace humans from the perspective of a complete personality, but in this context the computer is a tool to simulate and test strategies at scale. It demonstrates the overall effect of large numbers of individual choices in a system.
The gist of it being … even though coded strategies are over-simplified, the additional nuance and subtlety that a real human brings, probably does not make a difference at the system-wide scale … and in cases where we conclude that it does make a difference, then it ought to be something clear enough to be able to codify and then work back into our simulated agents.
Additional reference relevant to the above, an online free to play simulator, assigning cute character bios to each of the common Prisoner’s Dilemma strategies.
The Trust Arena
It says MIT license but I don’t see where the source code is downloadable, anyway I played for a bit and then gave up. This kind of resource is what we really need for getting kids (and Adam) into the spirit of the whole agent simulation.