[extropy-chat] magic johnson, aids, longevity ...

[Martin Striz]

On 4/18/06, Martin Striz <mstriz at gmail.com> wrote:
> On 4/18/06, Alejandro Dubrovsky <alito at organicrobot.com> wrote:
> > On Mon, 2006-04-17 at 05:28 -0500, Robert Bradbury wrote:
> >
> > >
> > >         Because creating stronger selection pressures increases the
> > >         rate of
> > >         evolution, in this case, of resistant strains.
> > >
> > > A nice succinct answer.
> > >
> >
> > And wrong, I think, in this case.  If you want to train a GA to do
> > something very tricky, a standard thing to do is to split the problem
> > into easier problems and you feed it to the population one by one, so
> > that at each step you get a large-sized population having a crack at the
> > next not-so-large problem.  Giving bacteria one anti-biotic problem at a
> > time seems like helping them to me.
> >
> > If you've got a population m of bacteria trying to solve anti-biotics J
> > and K, which need a mutation with probability 1/j and 1/k respectively
> > of arising, where j and k on the order of the product of m * average
> > mutations per bacteria, then the probability of one of the bacteria
> > having a mutation for either drug is quite high, but the probability for
> > any of the bacteria having both is almost zero (as long as bacterial
> > mutations look anything like normal/binomial distributions).
>
> Within a population of bacteria, there is a pantheon of polymorphisms
> already in existence from which to begin selection.  Some are not
> killed off as quickly as others, or remain sickly  in response to a
> particular antibiotic.  At any point many bacterial cells may be
> marginally resistant to a few of the antibiotics, enough so that a
> fully poly-drug resistant strain eventually emerges.
>
> That's why you should always complete your antibiotic prescription.

In other words, GAs are too simple a model of evolution to be
analogous in this situation.  They employ discontinuous fitness
assignments, whereas biological systems have totally smooth ones.

Martin