Mechanism design and deliberative agents

The central mechanism design problem is to develop incentives for agents to truthfully reveal their preferences over different out-comes, so that the system-wide outcome chosen by the mechanism appropriately re¤ects these preferences. However, in many set-tings, agents ’ do not know their actual preferences a priori. Instead, an agent may need to compute or gather information to determine whether they prefer one possible outcome over another. Due to time constraints or the cost of acquiring information, agents must be deliberative in that they need to carefully decide how to allocate their computational or information gathering resources when de-termining their preferences. In this paper we study the problem of designing mechanisms explicitly for deliberative agents. We pro-pose a set of intuitive properties which we argue are desirable in deliberative-agent settings. We show that these properties are mu-tually incompatible, and that many approaches to mechanism de-sign are not robust against undesirable behavior from deliberative agents