[From Bill Powers (970928.0514 MDT)]
Bruce Gregory (970927.2015 EDT)--
Sheri Turkle's latest book, _Life on the Screen_ has two
chapters providing an interesting view of the recent
history of AI and A-Life. As far as the latter is
concerned, she says (page 152):
"After the 1987 conference at Los Alamos, a consensus
began to develop that to qualify as life artificial
organisms most demonstrate four qualities. First, they
must exhibit evolution by natural selection, the
Darwinian aspect of our definition of life. Second, they
must possess a genetic program, the instruction for
their operation and reproduction, the DNA factor in our
definition of life. Third, they must demonstrate a high
level of complexity. A complex system has been
defined as "one whose component parts interact with
with sufficient intricacy that they cannot be predicted
by standard linear equations; so many variables are at
work in the system that its overall behavior can only
be understood as an emergent consequence of the
myriad behaviors embedded within." With complexity,
characteristics and behaviors emerge, in a significant
sense _unbidden_. The organism can self-organize.
This makes life possible. Thus the quality of
complexity would lead to the fourth necessary
quality: self-organization.
Before I encountered PCT, those conclusions might
have seemed sensible. I doubt I would even have
noticed that they omit the most salient feature of
living things...
I tried to open a conversation with a person in Japan named de Garis (a
Westerner) who has somehow persuaded backers to invest tens of millions of
dollars in the production of a genetic-algorithm learning chip that can
"evolve" neural connections at the rate of a billion per second. In his
writings, he shows no evidence of having any model at all of behavioral
organization (other than a rudimentary S-R idea). Instead, he is relying on
this GA approach to provide the model by itself -- but not in any form we
could understand. The model will exist as evolved connections among some
billion artificial neurons, which of course would leave us in the the same
position we now occupy in relation to the real human brain (assuming the
project works).
I think the "complexity" approach as you decribe it above really amounts to
giving up the attempt to understand how organisms work. The principle seems
to be that since we can't model behavior, the best we can do is model a
system that will (one hopes) generate a behaving system from scratch, by
itself. The aims of such projects shift accordingly. Instead of leading to
understanding, what they will lead to is a new breed of life, Silicon Life,
which will surpass the human race and become master of the Galaxy, if not
the universe. de Garis calls himself a "Cosmist," which is some indication
of the way this approach to artificial life can develop.
Take a look at de Garis' web page, at
www.hip.atr.co.jp/~degaris
He has posted a number of interesting documents. One of them is his
Doctoral thesis, which I have downloaded and read. Another is a long paper
intended for reading by the Japanese scientific/industrial establishment,
in which he offers the helpful observation that the Japanese are not really
creative and should rely on Western brains to tell them what science and
engineering should accomplish. He helpfully points out all the social and
individual flaws in the Japanese, saying that they should reform
themselves, and in the meantime fund a large institute in which the staff,
90% from the West, would show the Japanese what to do, providing the ideas
they are incapable of coming up with themselves while the Japanese provide
the money.
In his one post to me, he complained about his difficulties in working with
the "japs."
Best,
Bill P.