Game Theory in Practice
Some students have expressed skepticism over how well game theory can be applied to real situations, in which the players are real people who do not necessarily act rationally, and pointed out specific examples of where game theory fails to predict the outcome of games. This article, titled “Game Theory in Practice,” illustrates how game theory can be and is applied to real and difficult problems to predict outcomes or choose the best action to take.
The first examples are political: computer models based on game theory were able to predict the fall of Hosni Mubarak nine months before he fled Cairo. Five years before Ayatollah Khomeini of Iran died, game theory correctly predicted his successor (counter to the popular thinking at the time). Other notable examples from the handful of companies that offer game theory-based computer simulations of specific problems fall in the realm of auctions. When the FCC sold radio-spectrum licenses to broadcasters in an online auction, specially-designed software tracked the bids from all the competing companies and used that data to estimate the values that each company placed on each license. The software found that certain licenses were being overvalued, and advised buying different sets of licenses. Time Warner and Comcast, two companies who used game theory models to make decisions, saved over $1.1 billion. In another example, computer models based on game theory showed that in an auction with a $12 million prize for the second-highest bidder for oil-refineries in Israel, the prize actually caused the bidders to bid lower, because no one wanted to be the highest bidder and miss out on the prize.
The way that this software works is to distill the critical attributes of each entity (individuals, corporations, governments) into a few parameters that will govern their behavior in the simulation. In one simulation, these parameters were the desired goal, flexibility, attention to the problem, and influence. Once the structure of the game is set up, and each of the players has well-defined parameters, an iterative simulation is run to show how each player’s stance will evolve over time. The difficult part of applying the software to each specific problem is understanding the players and the structure well enough to have meaningful models of behavior. That’s why companies that sell their simulation services charge so much (up to $100,000); they have to conduct lengthy interviews with those familiar with the problem in order to create their model.
Some who study game theory believe that in the future, wars could be settled before they ever start by modeling the outcome. This is based on the idea that war is about testing the military strength and political resolve of one’s opponent. Once the players have determined the necessary information about their opponent, then they can agree on the outcome of the conflict (surrender or peace). With accurate enough game theory models, the outcome of a war could be determined before it was ever fought, and the countries could reach an agreement without anyone dying. Some game theorists also believe that in the future, simple, portable game theory simulation software could be run on smartphones and used to make everyday decisions when the best move is unclear.
Individuals may not always act rationally, especially when it comes to personal matters, but when money is involved, when the players are not individual decision makers, and especially when there are multiple players, game theory gives invaluable insight into human behavior. It is fascinating how people have gotten so good at distilling the motivations and dynamics of complicated interactions down to just a few parameters in a simulation that our government and some of our biggest corporations rely on them to make decisions which involve billions of dollars and potentially thousands of lives. As time goes on, I expect our models will get better and involve more parameters. One step towards artificial intelligence would be to write a program that could take information and intelligently assign the parameters to a simulation. Then game theory could be applied to everyday situations with ease.
http://www.economist.com/node/21527025