Path: utzoo!censor!geac!torsqnt!news-server.csri.toronto.edu!cs.utexas.edu!usc!sdd.hp.com!ucsd!ucbvax!ucsfcgl!labmed.ucsf.edu!brianc From: brianc@labmed.ucsf.edu (Brian Colfer) Newsgroups: comp.ai.philosophy Subject: Re: Modelling reinforcement Message-ID: <16562@cgl.ucsf.EDU> Date: 7 Dec 90 17:48:54 GMT References: <25667@uflorida.cis.ufl.EDU> Sender: daemon@cgl.ucsf.edu Organization: Dept. of Lab Medicine, University of California, San Francisco Lines: 54 In article <25667@uflorida.cis.ufl.EDU> turner@webb.psych.ufl.edu (Carl Turner) writes: >In article greenba@gambia.crd.ge.com (ben a green) writes: > >>What more is needed for an artifical person is motivation, or needs, and >>an adaptive mechanism that raises the probability of the behavior that >>satisfies them. This is "reinforcement". One of the challenges of AI is to create both the phylogenetics and the ontogenetics of the artificial organism (AO). I think that part of this may be found in programming the control principles (reinformcement, extinction, deprivation and punishment) into subsystems of the AO. This could be done probably with a hierarchical neural network system (NN) with groups of nodes emphasizing particular stimulus sensitivity and behaviors. [...] >You've hit on the basic idea of the "Hopkins Beast" of the 1950's. [...] > >It was a nice demonstration of the idea that you get complex >behavior from a simple system (the Beast) interacting with a complex >environment (the halls and offices of Hopkins). > >Of course, it didn't "acquire" a taste for outlets but was preprogrammed >to find them, else it would not have "survived" for long. The implication here is to scale up the program to a more complex system. Hmmm an interesting idea. >"Reinforcing" a machine for a behavior is the easiest thing in the world, >all you have to do is increment a counter, or reset a parameter.... I think that a systematic development of Skinner's ideas concerning verbal behavior will result in an effective general model for semantic processing. >Behaviorist principles of behavior and reinforcement seem an appropriate >domain for modelling and simulation, but computational modelling does not >seem to have caught on there. Perhaps someone could explain why. Besides the ideologic hangups people have about Behaviorism.. while the principles are simple the interaction results in an incredibly complex network of behavior patterns. It is similar to designing an ecosystem. I think it is just a huge job with lots of tedious work that will pile up to make an interesting whole. >Carl Turner >turner@webb.psych.ufl.edu -- Brian Colfer | UC San Francisco |------------------------| | Dept. of Lab. Medicine | System Administrator, | brianc@labmed.ucsf.edu | S.F. CA, 94143-0134 USA | Programer/Analyst | BRIANC@UCSFCCA.BITNET | PH. (415) 476-2325 |------------------------|