Online Markets Workshop: Predictive Game Theory
About the Series
- Date: Friday, May 11, 2018.
- Location: Kellogg Global Hub L070, (map), Northwestern U, 2211 Campus Dr, Evanston, IL 60208.
- Transit: Noyes St. Purple Line (map).
- Parking: Validation for North Campus Parking Garage (map) available at workshop.
- Registration: none necessary, bring your own name badge from past conference.
- 8:30-9:00: Continental Breakfast
- 9:00-9:05: Opening Remarks
- 9:05-9:45: James Wright:
Predicting Human Strategic Behavior: From Behavioral Economics to Deep Learning
- 9:45-9:50: James Wright Q/A
- 9:50-10:30: Sevgi Yuksel:
Biases Over Biased Information Structures: Confirmation, Contradiction and Certainty Seeking Behavior in the Laboratory
- 10:30-10:35: Sevgi Yuksel Q/A
- 10:35-10:55: Coffee Break
- 10:55-11:35: Annie Liang:
Predicting and Understanding Initial Play
- 11:35-11:40: Annie Liang Q/A
- 11:40-12:20: Colin Camerer:
Frontiers of Behavioral Evidence for Predictive Game Theory: Neural Activity and Visual Salience
- 12:20-12:25: Colin Camerer Q/A
- 12:25-1:30: Lunch
Titles and Abstracts
Speaker: James Wright
Title: Predicting Human Strategic Behavior: From Behavioral Economics to Deep Learning
Abstract: In order to do a good job of interacting with people, a system must have an adequate model of how people will react to its actions. This is particularly true in strategic settings: settings that contain multiple agents, each with their own goals and priorities, in which each agent’s ability to accomplish their goals depends partly on the actions of the other agents. Standard game theoretic models of strategic behavior assume that the participants are perfectly rational. However, a wealth of experimental evidence shows that not only do human agents fail to behave according to these models, but that they frequently deviate from these models’ predictions in a predictable, systematic way.
In this talk, I will survey my work on modeling human behavior in unrepeated, simultaneous-move games. These games can be used to analyze a surprising number of application domains, such as the advertising auctions that fund the major search engines, or the algorithms that are used to optimize the allocation of security personnel in ports and airports.
Title: Biases Over Biased Information Structures: Confirmation, Contradiction and Certainty Seeking Behavior in the Laboratory
Abstract: We study choices among information structures that are characterized by different biases. Bias is introduced via either distortion, through possibility of false reports as in cheap talk games of Crawford and Sobel (1982), or via filtering, through possibility of strategic omission of information as in disclosure games of Milgrom and Roberts (1986). The experimental design exploits how the optimal information structure depends on one’’s prior and the form of the bias- filtering or distortion. Typing subjects based on their choices in a series of questions spanning these cases, we find strong evidence for confirmation, contradiction and certainty seeking behavior. This is particularly surprising given that traditional explanations for confirmation or contradiction seeking behavior are shut down in our design. We discuss implications of our results in the context of political information and the role of media bias.
Joint with Gary Charness and Ryan Oprea.
Speaker: Annie Liang
Title: Predicting and Understanding Initial Play
Abstract: We take a machine learning approach to the problem of predicting initial play in strategic-form games, with the goal of uncovering new regularities in play and improving the predictions of existing theories. The analysis is implemented on data from previous laboratory experiments, and also a new data set of 200 games played on Mechanical Turk. We first use machine learning algorithms to train prediction rules based on a large set of game features. Examination of the games where our algorithm predicts play correctly, but the existing models do not, leads us to introduce a risk aversion parameter that we find significantly improves predictive accuracy. Second, we augment existing empirical models by using play in a set of training games to predict how the models’ parameters vary across new games. This modified approach generates better out-of-sample predictions, and provides insight into how and why the parameters vary. These methodologies are not special to the problem of predicting play in games, and may be useful in other contexts.
Joint with Drew Fudenberg.
Speaker: Colin Camerer
Title: Frontiers of Behavioral Evidence for Predictive Game Theory: Neural Activity and Visual Salience
Abstract: This talk will begin with a rapid tour of how level-k/cognitive hierarchy approaches conceive of predictive game theory. Then I will describe how eye tracking and neural evidence could be helpful. An example which uses computational models from vision science is applied to perceptual hide-and-seek games.