>How could the mechanisms of evolution be made more efficient?
See the comment below on Lamarkian evolution. I enclose all 'cause
the RNA stuff is relevant.
Really-Really-From: degaris at etl.go.jp (Hugo de Garis)
Really-From: alife at cognet.ucla.edu (Artificial Life Digest)
[ The following is a cross-post from Artificial Life Research List
Digest Number 081 Monday, August 3rd 1992. You are encouraged to
subscribe by sending mail to alife-request at cognet.ucla.edu - Moderator ]
Here is a quick report on the highlights of the ALife III
Conference which was held in Santa Fe, New Mexico in June 1992. It reflects the
personal biases and interests of the author.
Far and away the best paper was by Gerald Joyce, who spoke on his "evolution of
molecules" work. He takes RNA, and can mutate one, two or more specific bases,
and then clone the mutants in huge numbers. These mutants are then subject to
a selectionist test, so that only the more successful mutants survive.
The survivors are then further mutated etc, until molecules are evolved which
perform some desired function. This is probable future Nobel work. Good luck
John Koza used his Genetic Programming technique (evolution of Lisp programs)
to evolve self reproducing systems, and told his audience that the size of the
search space (with his primitives) was only of the order of a billion or so.
This is exciting, because its a lot less than earlier estimates (eg von Neuman's
29 state reproductive cellular automata Turing machine). It means that
it will probably be possible in a year or so to evolve selfreproductive
systems which can also do something useful. This will be essential when nano
tech is finally with us. Nanoscale machines will have to self reproduce in
order to build macroscale systems.
The biological robot (biot) community went home with the message that an
evolutionary approach to building biot nervous systems is the way to go.
Brooks-style handcrafting has become too complex and needs an evolutionary
approach. Even Brooks was talking about evolving gnat robots.
Dave Ackley showed that Lamarkian evolution can "blow the doors off" standard
GA evolutionary learning. An amusing and engaging talk.
L. Buss and W. Fontana teamed up to present a mathematical theory of the
development of systems towards life like behavior. They presented levels
0, 1, and 2, with increasing sophistication, e.g. self replication, then
genomes, etc. It was hard to follow but felt important. At the end of the
talk they claimed they had implemented it all in a computer program but
they failed to present results (and this was to an audience of
70% computer types!).
A fast-talking half-crazy Canadian (M. Tilden) upstaged Rod Brooks by showing
that he could do a lot of what Brooks does, but for a thousandth of the
price. Tilden makes ultra cheap little robots with an amazing functionality.
I hope Rod Brooks went home with the lesson. Funders take note!
Tilden will go down in ALife history for his reply to a non native english
speaker's question, "Why do you talk so fast?". Tilden's
reply was, "Whydoyoulistensoslowly?!"
J. Smits made tiny bilayer silicon strips curl up when a current is applied.
He intends to use these strips as mechanical movers for "silicon ants".
Randy Beer now uses GAs to build his insect circuits. Darwinian Robots was
one of the themes of the conference.
Maja Mataric (Brooks student) presented a video on Brooks version of swarm
intelligence. 20 robots were supposed to perform behaviors by emergence.
The video angles were too low to see effectively what was happening. Maybe
she should have done the filming in a basketball court and taken the shots
from a ladder.
Personally, I went away all fired up that "evolvable hardware" is possible.
I learned about FPGAs (software configurable hardware), so it will be possible
to treat the configuration bitstrings of FGPAs as chromosomes in a GA. Thus the
technology may exist today to fulfill my dream of building "Darwin Machines".
I just got back to Japan, so this report probably reflects my jet lag.
Hugo de Garis,
Electro Technical Lab (ETL), Japan.