DAI-List Digest Tuesday, 12 June 1990 Issue Number 10 Topics: Submission Info for 10th International Workshop on DAI Is Consensus a DAI Problem? Please send submissions to DAI-List@mcc.com. Send other requests, such as changes in your e-mail address, to DAI-List-Request@mcc.com. ------------------------------------------------------------------------ Subject: Submission Info for 10th International Workshop on DAI Date: Tue, 12 June 90 16:07 CDT From: Michael N. Huhns Call for Participation 10th AAAI International Workshop on Distributed Artificial Intelligence The Flying L Ranch Bandera, Texas (near San antonio) October 23-27, 1990 Papers are expected to be extended abstracts, 5-7 pages in length (approximately 2500 words), not counting the bibliography, and hardcopy only (3 copies). If your abstract is accepted, you may submit a more complete version (up to 20 pages) for inclusion in the informal proceedings and for possible publication in a special issue of IEEE Systems, Man, and Cybernetics being prepared by Ed Durfee, or in DAI Volume III. Please submit your papers to Michael N. Huhns (chairman) MCC 3500 West Balcones Center Drive Austin, TX 78759-6509 (512) 338-3651 or DATES: Deadline for submission of papers (3 copies, please): July 2, 1990. Notification of acceptance: August 7, 1990. Final papers due (for distribution at the Workshop): October 1, 1990. ------------------------------------------------------------------------ Subject: Is Consensus a DAI Problem? Date: Mon, 11 Jun 90 16:10 PDT From: Robin Hanson Consider the following "Consensus" PROBLEM: GIVEN a group of agents who 1) can vary radically in style of and capacity for thought 2) have access to different relevant information (which can overlap) 3) do not trust or respect each other (have hidden internal states) 4) are autonomous, and so can't be easily coerced And GIVEN a question where we might expect the agents to (internally) disagree What "social" mechanism can we set up (at a small or negligible cost) SO THAT the agents quickly form a public consensus belief on this question SUCH THAT it is clearly in the self-interest of each agent to participate in the process and ACT as if they agreed with the consensus? (I.e. outsiders could not tell by examining the various agent's actions that they intended to disagree substantially with the consensus.) I believe I have an answer to this problem, at least for the case where, with enough effort and analysis and data, the agents would all eventually come to agree on what the answer to the question is. Note the following are NOT solutions: 1) Big Brother puts a gun to their head, and forces them to act as if they agree. 2) Big Daddy offers them each $1M to act as if agree (costs too much). 3) They vote and majority wins. (Why should they act as if agree with this?) 4) They exchange all relevant information. (They might lie, costs too much.) I don't presently work in DAI--do those of you who do think this problem is relevant for DAI work? Is a result like this publishable? I am willing to post my answer, but would like to hear what other people think first. Robin Hanson hanson@charon.arc.nasa.gov (or hanson@ptolemy.arc.nasa.gov) 415-604-3361 MS244-17, NASA Ames Research Center, Moffett Field, CA 94035 415-651-7483 47164 Male Terrace, Fremont, CA 94539-7921