Interesting People mailing list archives
IP: Re:Another view from Josh Lederberg on First Cells, Then Species, Now the Web
From: Dave Farber <farber () cis upenn edu>
Date: Tue, 26 Dec 2000 17:54:28 -0500
To: farber () cis upenn edu Date: Wed, 27 Dec 2000 09:48:57 +1100 (EST) From: root () suburbia net (Charlie Root)[ Josh is a Nobel Laureate djf]To: Dave Farber <farber () cis upenn edu> From: Joshua Lederberg <jsl () jl10 rockefeller edu>Dave,Conversely: <<<< Microbiology=s World Wide Web by Joshua Lederberg (excerpt from a column syndicated abroad) .fi All the fashionable talk about computer "viruses" is supposed toexplain whatthese culprits do by analogy to their biological namesakes. But it may be equally enlightening to think of the biosphere of the real, livingmicrobesas a world wide web of informational exchange. Indeed, the two havemuch incommon, for living microbes exchange information with each other and their environment, with DNA serving as the packets of data going every whichway.What is different in the world of microbes is that they, unlike computer viruses, can evolve, and do so at a faster pace than theirhosts. Microbesare in fact well designed to exploit this difference to their advantage in the war that occasionally erupts between them and other species.God help us, someday the computer viruses may also be designed to"evolve".Or, unlikely, by happenstance. JoshuaFor archives see: http://www.interesting-people.org/I only wish they could! The fitness population of computer programs are highly discontinuous. Von-Neuman architecture (which all modern computers use) was designed to do what humans could not do, or at least not do well. VN machines are designed to be accurate and predictable. The trade off is that they are woefully brittle and inflexible. Economically this hasn't mattered too much, because we have a ready supply of flexible human beings who adapt code by hand. In VN machines, the genotype is the phenotype and single point mutations almost inevitably cause catastrophic failure. This is not the case with DNA based organisms where there are many layers of feed back and indirection, self-regulation, adaptive embryology etc, that cause the developing organism to adapt to point mutations in such a way that they are almost never catastrophic, are usually only slightly harmful and are rarely, but occasionally beneficial. Even if the chance of a beneficial mutation was the same (and it's not), the discontinuous fitness phenotype of evolving programs means that it is very hard for them to climb even a smooth fitness landscape. However, for something like a computer virus, the situation is harder still. The fitness landscape of the environment is highly discontinuous. System call 185 has no relation to system call 186. It's not an "almost 186". Either a checksum routine is correct, or its incorrect. Biological RNA viruses are simple, have no embryology, and only extremely primitive feeback mechanims, yet even there, the very three dimensional nature of viral building blocks, the electrical inverse square law, smooth gradients and parallelism in chemical reactions, and even the adaptability of the host cell itself all conspire to produce an smoothly evolveable system. This does not mean computer viruses can not evolve. It just means that they can not for the foreseeable future evolve anything truly novel. Evolving priorities from a list of pre-canned strategies is easy. Coming up with an original hole is not. If we can come up with truly adaptable code, we will be able to solve many difficult AI problems. It's a great outstanding research area which so far has achieved only marginal results. -- Julian Assange |If you want to build a ship, don't drum up people |together to collect wood or assign them tasks proff () iq org |and work, but rather teach them to long for the endless proff () gnu ai mit edu |immensity of the sea. -- Antoine de Saint Exupery
For archives see: http://www.interesting-people.org/
Current thread:
- IP: Re:Another view from Josh Lederberg on First Cells, Then Species, Now the Web Dave Farber (Dec 26)