Behavioral Game Theory and AI

Ever sense the Ron Howard film, A Beautiful Mind, there has been a fascination with game theory by the general public. It has been used in business seminars, television shows, and even sales meetings — often incorrectly.

The math behind game theory has proven itself within machine learning algorithms and even to an extent for financial theory. However, game theory is a mathematical representation of human behavior. As such, it has failed to fully comprehend the illogical nature of real-life human decision making.

Colin F Camerer writes in his book entitled Behavioral Game Theory, “Game theory, the formalized study of strategy, began in the 1940s by asking how emotionless geniuses should play games, but ignored until recently how average people with emotions and limited foresight actually play games.”

The basic premise of behavioral game theory is that choices made by participants in a game do not reflect the benefit those players expect to achieve from making those choices. For example, people use low-interest savings accounts from a fear of loss. Yet, many of the same people engage in high risk financial activities such as gambling and lottery play.

People are the chief complexity for modeling financial markets. Economics began by assuming everyone was rational. Game theory picked up the idea and created models of strategy based on this assumption. It was only recently that researchers caved to the truth. That people are irrational and often make decisions contrary to their own best interest.

Artificial Intelligence and Emotion

So far, machine learning has been successful in making strategic decisions when competing with humans. While impressive, this success has been limited to strategy games such as chess, trivia, and Go. These games are laboratories for rational decision making. However, they do not represent how people behave in life.

Giuseppe Cuccu of eXascale Infolab wrote, “Games are useful as AI benchmarks as they are often designed to challenge human cognitive capacities.” During game play, people concentrate on one goal and carefully alter their plans to maximize the probability of winning. In life, things are not that simple. There are often multiple, sometimes competing goals. For example, saving money for retirement versus having a nice home today.

Machines are not capable of making such irrational decisions. They maximize strategy for a specific outcome. If that is retirement savings, the machine will do everything possible to ensure the largest amount possible over n years. Meaning that a computer would not buy a home or a car or have children. This is because machines are not emotional.

A lack of emotion makes it very difficult for a machine to compete in financial trading. For example, a machine could understand all the factors involved in a stock price and make a sound decision accordingly. However, the human traders act upon varying degrees of emotion that could cause that stock to move in a direction unrelated to logical decision making. Resulting in the machine making a bad trade.

Fairness and Competition

In 1982, the Journal of Economic Behavior and Organization reported the findings of Guth, Schmittberger, and Schwarz. They concluded that people are willing to make a sacrifice when they feel an offer is unfair. In 1999, The Quarterly Journal of Economics reported, “people are willing to forego a gain, in order to prevent another person from receiving a superior reward.”

Mothers of multiple children probably already understood these phenomenons. However, the application to adults is surprising. Yet, it happens all of the time.

Consider a home priced below market value that is a perfect fit for the buyer in question. This home is exactly what the buyer wants and is already a good deal. However, the buyer will forgo the purchase due to the seller refusing to make a few hundred-dollar repair. Real Estate agents see this kind of behavior everyday.

In another scenario, consider two competing firms who are bidding on the same contract. The client likes them both and determines it is best to allow Firm A to manage the manufacturing and Firm B to perform quality control. Firm A is guaranteed millions while Firm B will only generate a few hundred thousand. Even if Firm B needs the business, they may decline the offer since it gives their competitor an advantage over them.

However, a machine is incapable of such irrational behavior. Even in deep learning where the computer is left to develop it’s own conclusions, there is no irrationality. Something great for logical calculation but troubling for a machine needing to understand a market based upon human behavior.

Group Decisions and Irrational Behavior

Game theory focuses on individual behavior. In contrast, behavioral game theory seeks to utilize both rational and irrational group decision making to describe the outcomes of strategic interactions. Therefore, it is currently the most realistic means of modeling human decision making.

In life, many decisions are made by teams over individuals. This is especially true in markets as they are the result of multiple teams both competing and cooperating with one another. Add to that, the fact that humans make 95% of decisions from mental shortcuts and intuition, then the possibility of understanding a market becomes all but impossible.

Perhaps there is hope for understanding by AI. In a 2005 study by Kocher and Sutter, it was discovered that groups did not perform more rationally than individuals at the start of a game. However, groups were able to perform more rationally in subsequent rounds. This appears to mimic a deep learning system as it is building a strategy. The question is can a machine learning system adapt to compensate for irrational human interactions.

The Effects of AI on Markets

The 2017 book, Artificial Intelligence and Economic Theory: Skynet in the Market., concluded that the use of AI machines in online trading and decision making has changed major economic theories. In particular, the authors discovered that AI has resulted in markets becoming more rational.

As machines take over the human aspects of market participation, things should become more rational. That may or may not be a good thing. For example, a company posting one bad quarter may suffer with significant loss of stock value. One that could not be overcome for years.

Such rational behavior could cause some assets to loose significant value at an extremely fast rate while others could skyrocket based upon factors never realized. Who knows? The point is that as machines get more involved things will change.

Conclusion

People are irrational and machines exhibit rational behavior to a fault. As both simultaneously work together and compete within financial markets, the results will be interesting if not horrifying. Perhaps even classical economic models will finally become relevant.

Thank you for reading.