Path: utzoo!news-server.csri.toronto.edu!rutgers!aramis.rutgers.edu!athos.rutgers.edu!nanotech From: autodesk!peb@uunet.uu.net (Paul Baclaski) Newsgroups: sci.nanotech Subject: Re: Is this stuff for real? Keywords: reality nanotech questions Message-ID: Date: 12 Mar 91 23:26:31 GMT Sender: nanotech@athos.rutgers.edu Lines: 56 Approved: nanotech@aramis.rutgers.edu In article , mmt@client2.DRETOR (Martin Taylor) writes: >...A machine with these desirable abilities > would be in a design space that (at least locally) was rather dense > with viable machines, and the probability of a mutation leading to > a viable design could be appreciably different from zero. Not necessarily. I can see two types of errors that can be controlled using standard engineering: 1. Development errors. This can include errors from reading the genotype, identifying parts and installing parts. The number of degrees of freedom incorrect parts have is directly proportional to the density of the "construction space". Errors such as these can be mitigated through self testing and through validation-- using multiple nanomachines that test each other such that reproduction is not possible if a machine is not validated. Such validation would certainly slow reproduction, but it creates a nice fail safe link such that two or more machines must fail for a failure to be successful. Further, since self-test is difficult, tests using multiple machines would have more flexibility. 2. Genotype transcribing errors. This corresponds to the "dense design space" in the quoted message above. It is tempting that this be a dense space with continuously varying genes-- this is a design for evolvability. On the other hand, the genotype "Turing Machine Tape" does not need to have this characteristic and it can have checksums to validate that it never mutates successfully. Developmental errors must be avoided by using design discipline to test created machines. Genotype errors can be made arbitrarily small by checksums. Genotypes that are designed for evolvability (continuously varying genes that map to continuums in the phenotype) are to be avoided or at least used very carefully (this is what Martin Taylor probably means by "locally dense"). In previous Gray Goo discussions, the conclusion has often been that gray goo will probably not occur accidently--it requires a (malicious) designer. For background, there are two relevant articles in _Artificial Life_, Chris Langston, ed., Addison-Wesley, 1989: The Evolution of Evolvability by Richard Dawkins and Biological and Nanomechanical Systems: Contrasts in Evolutionary Capacity by Eric Drexler. Paul E. Baclaski Autodesk, Inc. peb@autodesk.com