Can Game Theory Predict the Future?

cavedave

Talk on this Thursday

Game theory: When people compete with each other they always do what they think is in their own best interest.

Hailed as 'the new Nostradamus', Bruce Bueno de Mesquita a consultant to the CIA and the Department of Defense in the US, Bruce Bueno de Mesquita has built an intricate computer model that can predict the outcomes of international conflicts with bewildering accuracy.

Is anyone going to this talk? Anyone read Predictioneer?

Any opinions in general on game theory being used for practical purposes? I have always found it very theoretical and not particularly applied.

hivizman

This looks like an interesting lecture. I'm a bit worried, however, that the speaker, Bruce Bueno de Mesquita, seems to believe that game theory was "formulated by John Nash" (to quote from the lecture flyer) - game theory is usually attributed to John von Neumann, who developed the basics in the 1920s, and wrote, with Oskar Morgenstern, The Theory of Games and Economic Behavior, published in 1944. Von Neumann taught Nash at Princeton. However, they made a movie (A Beautiful Mind) about Nash, so perhaps he's better known to the general public.

Game theory has been used in many applied contexts. One of the most famous was the auction of 3rd generation mobile telecommunications licences in the UK around 2000. The structure of the auction was developed by a team led by the mathematician/economist Ken Binmore (he recently wrote Game Theory: A Very Short Introduction, published by Oxford University Press in 2008), and was based on the ideas of game theory. The auction had originally been expected to raise around £3 billion, but ended up generating £22 billion.

I'd be interested to know two things about the prediction model of Bruce Bueno de Mesquita. First, he claims "more than 2,000 correct predictions", but how many of these are actually "post-dictions" - applications of the model to events that have already taken place? Second, has he compared his model to any other prediction model - a 90% "success" rate may appear impressive, but if there are rival prediction models that achieve a higher rate, then his model may not be all that interesting.

Sheeps

How exactly do you model an international conflict?

cavedave

Bruce Bueno de Mesquita, seems to believe that game theory was "formulated by John Nash" (to quote from the lecture flyer)

I cannot find that reference. I presume he is refering to non cooperative game theory being kind of founded by Nash as opposed to the cooperative kind of Neumann and Morgenstern. The non cooperative kind would seem to be of more use in conflict settings.

but how many of these are actually "post-dictions" - applications of the model to events that have already taken place?

They seem not to be

Second, has he compared his model to any other prediction model - a 90% "success" rate may appear impressive, but if there are rival prediction models that achieve a higher rate, then his model may not be all that interesting.

Also how is "success" defined?

Sheeps
How exactly do you model an international conflict?

You get a herd of sheep. Each one represents a country. You put them on some grass with a grid system laid out on it. Where they **** is your answer.

hivizman

cavedave wrote: »

I cannot find that reference. I presume he is refering to non cooperative game theory being kind of founded by Nash as opposed to the cooperative kind of Neumann and Morgenstern. The non cooperative kind would seem to be of more use in conflict settings.

It's on the page that you linked to (scroll down to the section headed "Further Information"). You are probably right that they have non-cooperative games in mind, since that's what de Mesquita is modelling.

Here's another quote from the website advertising the lecture:

His systems based on game theory have an astonishing 90%+ ratio of accuracy and are frequently used to shape US foreign-policy decisions.

Are we really sure that the use of game theory is actually improving US foreign policy?

The outputs of some game-theoretic models are not predictions that specific outcomes will certainly occur but rather probability distributions across possible outcomes. For example, given various inputs, the model might suggest that the probability that Iran will develop a nuclear weapon by 2012 is 10%. If in fact Iran does develop a nuclear weapon by 2012, is this counted as a "successful prediction" of the model?

Mellor

cavedave wrote: »

Any opinions in general on game theory being used for practical purposes? I have always found it very theoretical and not particularly applied.

It's used in a practical sense over on the poker forum. Game theory threads come up a lot over there

Fremen

Global thermonuclear war? Strange game. The only way to win is not to play.

cavedave

He has a talk here

https://www.youtube.com/watch?v=ts5MKtXNpMQ

Oddly enough in the book he claims that people dont have intransitive preferences even though in his examples they do and in reality they do

tony 2 tone

hivizman wrote: »

This looks like an interesting lecture. I'm a bit worried, however, that the speaker, Bruce Bueno de Mesquita, seems to believe that game theory was "formulated by John Nash" (to quote from the lecture flyer) - game theory is usually attributed to John von Neumann, who developed the basics in the 1920s, and wrote, with Oskar Morgenstern, The Theory of Games and Economic Behavior, published in 1944. Von Neumann taught Nash at Princeton. However, they made a movie (A Beautiful Mind) about Nash, so perhaps he's better known to the general public.

In his book Predictioneer, he credits von Neumann and Morgenstern with inventing cooperative game theory and Nash with the other type (uncooperative? bluffing?) in the 1950's.
The blurb on the meetforreal page was probably written by the people who originised the talk.