Although I use this blog mostly to discuss cancer and evolution (thus the name of the blog) and the role of mathematical tools to study those, some times other topics manage to squeeze in. More often than not , for a reason. This time is not coffee but game theory. The tool that some researchers use to study evolution (and more recently cancer!) started as a mathematical tool to study the politics of the cold war. Von Neumann himself was, for a while, an employee of the DoD affiliated RAND corporation.
Back to those roots, NYU’s Bruce Bueno de Mesquita delivered this TED talk in which he explains GT and how it could be used to understand the future of Iran’s nuclear policy. For those of you who don’t know TED, these are a series of conferences on the topics of Technology, Entertainment and Design. The aim seems to be to make these talks short and as accessible and understandable as possible so they are nothing if not entertaining.
The idea is that international politics are shaped by the interactions of rational players and that we can make predictions if we know who these players are and what is their motivation. In the case of Iranian politics with respect to nuclear policy, key players include the president of the U.S., the president of Iran, the prime minister of iran, the advisors of these three people (who are closer to the subject but haven’t got all the power to implement their will) and the advisors of these advisors (to put a limit somewhere). The more agents and interactions we consider in our game, the more challenging the mathematics (or the politics) get, so the better the computer will have to be to run all the required simulations.
In brief, de Mesquita lists four things that characterise his political game: players (political leaders and different levels of advisers), public motivations (interestingly, no need to second guess what the real motivation could be, just what they publicly declare), the importance of the topic to each of these players (I guess it could be argued that the stakes for Ahmadinejad are higher than for Obama) and how much can the players influence the outcome of the game. Bruce de Mesquita noticed that history does not play a role in determining the outcome of the game: as long as we have a good idea of the motivation and influence of the player (although it could be argued that to have a good understanding of the player, his/her interactions and motivations, a good understanding of her/his history would be crucial).
All in all an interesting talk about a model that is claimed to have a 90% of accuracy, which is a pretty spectacular degree of success when so many assumptions have to be made about the players and their motivations. How much more accurate could a model be to predict cancer evolution? We don’t need to second guess the motivation of the cells: survival and proliferation; or the importance of proliferation for the players: crucial for tumor cells; or the influence of each cell on the overall tumour progression (important at the beginning, a bit less as the tumour grows, important again when it comes to metastasis).
Very interesting idea. I look forward to watching the talk. I recall how accurately fivethirtyeight.com predicted the US election (even down to the number of Congressional seats). It raises several warning flags however. How do you make reliable predictions based on the views of a few actors when they are largely representatives of larger interests? Are the motivations of the oil and gas industries included in the analysis? What about the population of Iran who still remains opposed to American interventionism after Eisenhower overthrew their democratically elected government and reinstated the Shah (setting up the populist anger behind the Iranian Revolution)? But perhaps these are largely background concerns and can be modeled effectively.
Interesting. However, Iran is a theocracy, so the list of key players should include its Supreme leader, as he is more powerful than its president, including the say on defence. Whether this makes things easier or harder to predict, I’m not sure.
Eric: Good point about extrapolating too much from a few players. In the talk de Mesquita talks about how fast computers allow him to run simulations involving manny players and many interactions which would make the model (potentially) more robust. The rest of the issues you rise are in a sense (partially) related to the history of the game which de Mesquita disregards explicitly.
Lee: I guess I forgot to mention the supreme leader but the point is that by including the decision makers and a number of layers of advisors, de Mesquita claims he can give accurate predictions of rather complicated political problems.
I loved this TED talk, but couldn’t help thinking that the act of observing must influence the outcome. The speaker made specific predictions about the actions of Iran over the coming years. If any of the people of influence built into his model sees this TED talk, that may affect the decision-making process. This question was raised at the end, but the discussion was minimal. If these theories are widely adopted, wouldn’t they immediately become useless, as the ‘opposing side’ seeks to do something other than what the algorithm predicts?
I am not de Mesquita but I will give you my take on the issue. In game theory the idea is how, if we know the interactions between the relevant players, we can predict an equilibrium (for instance, a Nash equilibrium) so that all of the players have no incentive in changing their way of playing. In many cases knowing the outcome of the game does not make a difference if you are not willing to change the payoffs, that is, your motivation. In the case of the Iranian game (for want of a better name), a solution that saves the face of the main players is likely to be stable, even if the players know the solution beforehand.
If you are into game theory then consider Tit-for-Tat (by Robert Axelrod studying evolution of cooperation). His research showed that, even when every other player knew how this tit-for-tat player was going to play (play nicely with nice players and retaliate those who behave bad), that didn’t affect its success.