Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!rutgers!aramis.rutgers.edu!athos.rutgers.edu!nanotech From: ems%nanotech@princeton.edu Newsgroups: sci.nanotech Subject: Re: viruses, computer & bio (plus a fix) Message-ID: Date: 28 Jul 89 00:42:02 GMT Sender: nanotech@athos.rutgers.edu Lines: 81 Approved: nanotech@aramis.rutgers.edu >In article ems%nanotech@princeton.edu writes: >>But, I don't think this is quite the entire picture. One promise >>of nanotechnology is the ability to make that nanovirus vastly >>more *selective* in it's targets, hence a better weapon. One might >>build an AI-based nanovirus that would only spare ardent capitalists, >>for instance. > >I find the idea of a nanovirus that could read personalities very >unlikely, due to the inherent complexity of mapping from neural >activity in the brain to even abstract thoughts, much less political >inclinations. Before we have the ability to design such a nanoagent, >we will probably need the ability to design minds from scratch -- and >*that* would have far greater implications (both promises and >problems) than mere biowarfare. > Au contraire, mon frere :-) You're trying to do the job by the most direct route, which is probably also the toughest. There is a much simpler way the nanovirus can get a pretty good picture of how good a capitalist you are, (as well as much else). It simply contacts its compatriots, that have infiltrated your financial records, which are already in convenient electronic form. Now, if you fall outside certain preset parameters, zzzzt! No mind reading is required. In fact, variations of this technique can yield a great deal of selectivity, without requiring "true" AI at all. (Hmmm, should this be meme be discouraged? ) [ Much interesting thought on genocide, HIV elided...] >>This is part of the whole category of questions relating to the >>unethical misuse of technology. Let me now suggest a "fix" that one >>day, just *might* be possible thru nanotechnology. The leading force >>might use their time advantage to design an artificial conscience, >>and apply it to *everyone*, to modify behavior.... >For the reasons I stated above, I find this extremely unlikely short >of a complete solution to both psychology and AI. This would require >knowledge of how extremely high-level concepts such as "other people" >and "harm" are stored in extremely low-level neurological processes. >Furthermore, it requires knowing how to modify neurological structures >to acheive a very complex high-level behavior. If we can do this, I >feel we will be able to design minds to our own specifications, and >when this happens we will need to deal with much more complex issues. I admit that designing an artificial conscience is much tougher than the capitalist nanovirus described earlier. I would attempt it this way: First, map the neuron structure of a human brain using nanotech. Next, use the virtually unlimited amounts of CPU available thru nanotech to simulate the interaction of a computer model of this brain with a (simplified) model of an "outside world". (You can run this simulation very fast, and also run many copies in parallel). Finally, "all" you do is treat this brain model as a black box, and determine the set of outputs you want to avoid. Deciding when the "outside world" has been harmed (by the brain model outputs) could possibly be determined by examining increases in the disorder of the "outside world" part of the simulation. Sounds simple? It's not, but it does outline at least one approach to the job, that is more tractable than creating complete theories of psychology & AI. (I started with CPUs having a whole 8K. After a while, I learned to never make my machines do anything that they don't have to. :-) >Brian Yamauchi University of Rochester >yamauchi@cs.rochester.edu Computer Science Department Ed Strong AT&T, Bell Labs princeton!nanotech!ems And in < 1 month -> Princeton University Computer Science Department [The simulation method for implementing the "conscience" seems unlikely to work, primarily because (remember) the thing we're trying to protect against is mischief done by complex nanosystems. I find it difficult to believe that an nth-generation system could simulate (a) the designer's thoughts, (b) the nth-generation CAD system, (c) the n+1st generation system being built, and (d) enough of the real world, almost necessarily at the molecular level, to detect craftily laid schemes. Though offhand, I can't think of a better one... --JoSH]