At 11:58 AM -0700 7/20/97, Robert Hettinga wrote:
The idea of micromoney as processor food, I think, gets us a way to pay for the adaptive evolution with the same, or better, results than we'd get by top-down monolithic project design.
A husband-wife programming team just spent the last day or so at my place, filling me in on their recent successes with their "agent-oriented database" for financial trading. Unlike a lot of the hype about agents, in books and in press releases, they've used their system for data mining of financials, and have made a _lot_ of money with it. (They are not the folks at The Prediction Company, in case anyone is wondering. Nor are they the D.E. Shaw group. I can't say more right now, as the stakes and pool they are running are up in the $10M range, and they hope/expect this to soon get a lot larger. This could turn out to be one of the seminal applications of this set of interrelated ideas.) Bidding of agents is an important part of this approach. (Allocation of resources, "putting their money where their mouth is," genetic programming, evolutionary learning, data mining, neural nets, and all the other major buzzwords. Their agents are somewhat "agnostic" about any particular approach and they "use it if it works, otherwise kill it." The Koza-style genetic programming top layer then makes use of the various buzzword paradigms at lower levels, e.g., using time-series analysis for parts, neural nets for other parts, forward-propagation for other parts, etc.) The "agoric computation" work of Miller, Drexler, Tribble, Huberman and others is relevant. (Huberman edited a book, circa 1987-8, "The Ecology of Computation," which contains several of the important papers. The Miller and Drexler papers are the ones to focus on.) The "Digital Silk Road" work of our own Norm Hardy (and either Dean Tribble or Mark Miller, I forget which right now) is highly relevant. Some of this work is already being used to "market allocate" CPU cycles in distributed systems. Sort of analogous to the "distributed crack" work (except the distributed crack work is a winner take all approach, not counting the confused situation with the DESCHALL crack recently). In computational models where the prize is not so obvious, as in most routine calculations in business, a market allocation model better uses computational resources. Selling spare CPU cycles is another related idea. Several Cypherpunks have ideas along these lines. Coincidentally, my last major project for Intel involved developing a scheme for better automating and streamlining wafer production by having wafer "runs" (manufacturing lots) bid for access to scarce resources. I think I was strongly influenced by thinking about how wafer runs could be seen as "objects" carring their own local state--the complicated time history of the treatments they had received in the fabs, and test results, etc.--and how traditional wafer lot tracking systems failed to intelligently use any of this information. This "Frame-Based Manufacturing System," using the work on frames with slots contained triggered methods (daemons), would have essentially put an economic, agoric layer on what is now mostly a human-run bureaucratic layer. (With various departments and groups clamoring for access to equipment, rather than an auction approach.) My environment at that time was a Symbolics 3670 LISP machine, with a user interface/GUI of unparalleled elegance and power. And KEE, the Knowledge Enngineering Environment, from Intellicorp. I left Intel in '86, and the project limped along for another couple of years. Miller and Drexler visited my group in '87, after I was gone, and the work on using auction methods in wafer fabs entered into their thinking, as they later told me. (Intel did not deploy my vision. I can guess some of the reasons. Last I heard, they're still struggling to adapt conventional relational data bases for tracking wafer runs, but are not integrating in the vast amounts of (expensive) knowledge in any meaningful way.) These ideas are all part of a larger mosaic (TM, the Netscape Corporation) of ideas involving collective computation, intelligent agents, object-oriented data bases, "seas of objects" (a la Gelernter's "Linda" system), auction markets, price discovery, and AI in general. It's sad that so much of Cypherpunks coding has, perforce, been "stalled" at the level of implementing mundane ciphers and showing weaknesses in ciphers known 20 years ago to be weak. The really neat stuff is taking a long time to percolate out. --Tim May There's something wrong when I'm a felon under an increasing number of laws. Only one response to the key grabbers is warranted: "Death to Tyrants!" ---------:---------:---------:---------:---------:---------:---------:---- Timothy C. May | Crypto Anarchy: encryption, digital money, tcmay@got.net 408-728-0152 | anonymous networks, digital pseudonyms, zero W.A.S.T.E.: Corralitos, CA | knowledge, reputations, information markets, Higher Power: 2^1398269 | black markets, collapse of governments. "National borders aren't even speed bumps on the information superhighway."