Overcoming a Lack of Intrinsic Trust
|Dan Oblinger||DARPA/IPTO||[Home Page]|
Date: Wednesday February 14, 2007 at 10:00
Location: EJ228 (Directions)
Each civilization is defined by the set of legal, social, and financial mechanisms that guide and regulate the behavior of its citizenry. By necessity these mechanisms are rooted in the technology of the day. The advent of ubiquitous, programmable, networked computation (the Internet revolution) provides a new base technology that in several interesting ways is very different from the pencil/paper, face-to-face, phone, and postal mechanisms that it replaces. As such, this new technology holds the possibility that age old problems might be attacked in novel and improved ways using classes of mechanisms that would simply have been unworkable using pencil and paper.
In this talk we discuss two novel approaches to very long standing problems for humanity: how to allocate resources, and how to achieve trust (and honest behavior) among groups of individuals with little intrinsic trust.
NEW FINANCIAL MODELS FOR KNOWLEDGE-BASED GOODS A simple free market pricing model emerged from, and works well for many physical goods where most costs are tied to the production of the good, rather than a goods with large startup cost. Knowledge-based goods, on the other hand, do not have this cost model; they have a very high initial cost and low (near zero) cost per additional good. Furthermore, purchasing regarding one IP product (especially software) often constrains or modifies subsequent buying in a way that can degrade the optimality of a traditional market. We argue, in this talk, that such non-efficiencies result in a drastic (order of magnitude) additional burden to the whole knowledge based economy. We explore the possibility of distributed pricing models that mix aspects of open source software, and its ability to avoid lock in, with models that compensate contributors. One possibility afforded by the new networked usage model, is that contributors could be compensated a posteriori based on actual usage, rather than a priori based on agreed upon contracts. This, coupled with a different ownership model, could allow contributors to contribute based on their personal beliefs about the utility of their efforts, and allow buyers to pay only for the a posteriori useful increments to the growing body of open source material.
GHOST - GUARANTEED HONEST ORGANIZATIONAL SELF TESTING This concept is a rather radical approach to creating functioning organizations in situations of low trust. We propose a provably self validating system in order to guarantee that there is no way for a given organization to pass its own self testing if it is systematically breaking the rules published for that organization. I have worked out some details of this approach for use in monitoring governmental elections, and have considered its use in organizations that give our aid money, and also organizations that are involved in public policing. Surprisingly, it seems possible to design a computation + human processes which has provable properties, even if one makes very weak assumptions about the honesty of the members of the organization, and even if *all* testing of the organization is done those same potentially dishonest members. This seemingly magical self-testing approach is rooted in groups of humans collaborating through a novel integration of existing technologies(E.g. public key encryption, peer-to-peer networked algorithms)
Dan Oblinger is serving as an IPTO Program Manager on leave from the University of Maryland where he is a Research Scientist at UMIACS. Dan has 15 years of experience in Artificial Intelligence. His focus is knowledge-based machine learning and it development across a number of disciplines including: text analytics (email-mining), bio-informatics (protein structure prediction, gene array classification), data mining (large scale clustering, speech disabilities analysis, and others). His responsibilities at DAPRA include the Bootstrapped Learning and Mobius programs.
Prior to joining U. Maryland in 2005 he was a research staff member at IBM T.J. Watson Research Center in New York. There he was one of the initiators of an ongoing effort to invent technologies that can be re-programmed in the field by novice users through mimicking behaviors demonstrated by those users. More broadly he is interested in practical learning solutions which succeed because they employ richer input knowledge structures than todays machine learning can.
Dan holds a B.S. (mathematics and computer science) from Northern Kentucky University, an M.S. (computer science) from the Ohio State University, and a Ph.D. (computer science) from the University of Illinois. He currently resides in the District of Columbia.
Please arrive at least 10 minutes early in order to sign in and be escorted to the conference room. SRI is located at 333 Ravenswood Avenue in Menlo Park. Visitors may park in the visitors lot in front of Building E, and should follow the instructions by the lobby phone to be escorted to the meeting room. Detailed directions to SRI, as well as maps, are available from the Visiting AIC web page.