<html>
<body>
Hi, Howard!<br><br>
I was a bit too brief on some important ideas you raised. Let me try to
correct that.<br><br>
You said:<blockquote type=cite class=cite cite>
<dl>
<dd>Paul's words suggest that restlessness and boredom have been a key
part of this learning system. I've been calling this a restless
cosmos, a driven cosmos, an obsessive compulsive cosmos for a very long
time. But Paul is suggesting that we make computer-based learning
machines restless too. That we make them try out new possibilities
just for the hell of it, just to evade the pain of boredom, the pain of
staying precisely the same, the pain of ennui. Paul is on the brink
of suggesting that we make computational programs hunger for pop culture,
for music and games that test and expand the silicon brain in new ways.
<dd>
<dd>Paul is suggesting that genes may be as restless and boredom-prone as
Baudelaire, who painted ennui as the ultimate pain. He's suggesting
that on the sly, when they're not working, genes play around and dance in
leisure time. Or at least that's what Paul's ideas inspire in me.
<dd> </blockquote>
</dl>Basically right. Exploration is an unavoidable issue in designing
true intelligent systems.<br><br><blockquote type=cite class=cite cite>
<dl>
<dd>I know that leisure, entertainment, pop culture, art, and play are
not useless. I've known it since I began my 20 years of fieldwork
in these fields--poetry, art, magazine publishing, and finally popular
music. Paul seems to be whispering to me that these cultural
expressions may be a stochastic search for new possibilities. And
his words suggest to me that genes play games too. They play the
sort of musical games--establishment of a theme, then variation on
it--that Greg's mechanisms make possible.
<dd>
<dd>Paul, my apologies if I've bent your words, but they're
extraordinarily evocative.</blockquote>
</dl>======================================================<br><br>
The word "intelligent" has many definitions -- especially in
marketing by Beltway Bandits.<br>
I have explained at length why I would say that TRUE intelligent systems
all involve<br>
some kind of optimization... some kind of learning to "best possible
results" by some<br>
kind of measure of success. Let me not revisit all that just now. (There
is a book<br>
edited by Dan Levine, Optimality?... and lots of stuff I cite in my
papers at arXiv.org,<br>
in the quantitative biology part.) <br><br>
Now... in really complex environments, the optimization always comes down
to<br>
"nonconvex optimization" -- optimization in a situation where
there are lots of<br>
LOCAL optima, like deceptive foothills on the way to the highest mountain
peak.<br><br>
There are actually some very sophisticated and successful engineering
systems, which are<br>
worthy of being called "intelligent" (I feel, for complex
reasons), which do not have a real<br>
systematic exploration component. But to be truly brain-like, they need
that, in order<br>
to perform well in complex nonconvex environments. Conversely, there are
some very simple<br>
genetic algorithms, not at all brain-like, which do very good nonconvex
optimization<br>
AT A GIVEN TIME, SUBJECT TO USING NO PRIOR INFORMATION -- but have many
limitations.<br>
They are very useful in engineering today, in part because no one has
implemented anything more <br>
truly brain-like for these kinds of applications. (Though I know some
people who would<br>
say that is changing -- people like Thaler or Wunsch or Serpen.) They are
the best<br>
state of the art, on the whole, for problems like exploring the space of
possible designs for<br>
antennas to do specialized tasks as well as possible, and things like
that.<br>
They are used in the best real-world Optimal Power Flow packages used
by<br>
electric utility companies. <br><br>
------------<br><br>
OK, so exploration is necessary for the higher capabilities...<br><br>
BUT<br><br>
This does not mean that intelligent people have to try drugs or color
their hair pink.<br>
(Though I have known some who went through that, as most of you
have.)<br><br>
In fact -- in my own theory of how the brain works (the part summarized
in just a few places, like<br>
chapter 10 of my book Roots)..<br><br>
I would say that the essential difference between the wiring of the
original <br>
basic mouse brain and the wiring of the human brain<br>
is that humans don't have to learn form their own mistakes. <br>
They have an inborn ability to learn from the mistakes of others<br>
instead.<br><br>
And that's where we started our discussions, Howard. I was really hoping
that you<br>
and David Smith and I could joint author a more popular (or humanistic
paper)<br>
on the fundamental revision of Freud's theory of dreams implied by this
theory of how<br>
the wiring works. The claim that our human brains are hard-wired to
commonly<br>
give us dreams presenting the viewpoints of OTHER humans. The
implications<br>
are extremely far-reaching in my view, not only for theory but for
improving<br>
our ability foster human growth, one of the very most fundamental
issues<br>
on the table in the world.<br><br>
All for now.<br><br>
Best,<br><br>
Paul</body>
</html>