Evaluation of Automatically Designed Mechanisms

By Anthony Jameson, Christopher Hackl, and Thomas Kleinbauer (2003)

Proceedings of the First Bayesian Modeling Applications Workshop at the Nineteenth Conference on Uncertainty in Artificial Intelligence, Acapulco, Mexico.

Abstract

In their UAI 2002 paper “Complexity of Mechanism Design”, Conitzer and Sandholm presented computationally tractable methods for automated mechanism design (AMD): For a given specific preference aggregation problem, such a method generates a nonmanipulable outcome selection mechanism. The present paper examines some of the decisions that need to be made by a designer who is considering the application of AMD in a particular context. Using examples from the application area of rating aggregation, we discuss several potentially desirable properties of mechanisms: symmetry, determinism, familiarity, equity, and robustness. Our example evaluations illustrate that some of these properties can be straightforwardly achieved with the methods of AMD, whereas others may sometimes require the designer to resort instead to standard mechanisms (such as averaging of ratings) that may not fall within the scope of AMD. We conclude that it is advisable, in any particular application setting, to conduct comparative evaluations of potential mechanisms; and that these comparisons should sometimes include hand-crafted mechanisms.

Note

Download

Full Publication:  

BibTeX entry

@inproceedings{JamesonHK03,
  year = {2003},
  author = {{Jameson}, Anthony and
            {Hackl}, Christopher and
            {Kleinbauer}, Thomas},
  title = {Evaluation of Automatically Designed Mechanisms},
  booktitle = {Proceedings of the First Bayesian Modeling Applications
    Workshop at the Nineteenth Conference on Uncertainty in Artificial
    Intelligence},
  address = {Acapulco, Mexico}}