CS Postdoc Profiles

Research Profile

Imagine a designer—call him D—responsible for designing a system in which self-interested agents (people, autonomous software agents) would act. Perhaps he would be designing a new healthcare system, or something much lower key, like an auction.

The designer has some objective in mind, perhaps maximizing revenue that he would collect from an auction, or universal coverage in a healthcare system, and would like to design a system that achieves his stated objective.

In economic parlance, D faces a mechanism design problem, which has now a rich literature featuring numerous theoretical, econometric, and behavioral insights.
Standard assumption in the former two strands of the mechanism design literature is rationality on the part of the agents that ultimately make use of the designed system, an assumption which is often called in question by the behavioral economists. Significantly, even given the rationality assumption, analytic tractability presents severe constraints on the kinds of mechanisms that can be considered theoretically.

For example, mechanism design theory provides little insight on how to design a healthcare system, or even about outcomes of complex auctions mechanisms, such as combinatorial auctions, in cases when these cannot be perfectly implemented due to computational complexity constraints.

Indeed, there is little known even about simple design problems (such as single-item auctions) in the presence of design constraints. As a response to the overwhelming complexity of practical mechanism design, I have set out to pursue a long-term research program that unites theory, simulations, and behavioral experiments. At its core is a recognition that there are today two fundamentally different kinds of autonomous agents: software agents and people.

Software agents are not encumbered by emotion and are the best approximation to rational agents that we are likely to have.

As such, systematic analysis of equilibria—using simulations when the problem is beyond mathematical tractability—is well-suited to systems designed primarily for software agents.  People, on the other hand, are influenced by a plethora of factors outside of economic considerations (e.g., social and cultural factors).

While it is very difficult to have general and mathematically tractable techniques for analyzing humans, we can use behavioral experiments to calibrate simulation-based models of people in specific settings, and use the calibrated models in simulation-based mechanism design.

Contact Information

E-Mail: EMAIL OBFUSCATED

Phone: 773-562-0148

Categories Posted To:

AI / Machine Learning / Robotics / Vision, Information Systems / Information Science, Numerical/Scientific Computing / HPC / Data-Intensive Scalable Computing, Social Computing / Social Informatics