On May 28, 2007, at 8:11 PM, Russell Wallace wrote:
On 5/29/07, Samantha Atkins <[EMAIL PROTECTED]> wrote:
I think you know well enough that most of us who have considered
such things for significant time have done considerable work to get
beyond "metal men".
Yep. So had I. Then I discovered considerable work is nowhere near
enough, alas.
A "will to survive" is probably essentially to any autonomous being
if it is to survive.
Indeed, which is why we observe animals, evolved for autonomous
survival, reliably possess such a thing (or behave as though they
did, which amounts to the same thing here).
Autonomous machines, however, are conspicuous by their
nonexistence, and only partly because of technical difficulty.
It is largely all technical difficulty.
The truth is, the market doesn't want them in the first place.
For some values of autonomous, i.e. unmanned and not directly
monitored, in at least some areas the market does want them.
That's why it's not just that machines don't have a will to
survive, or even that they don't have the rudimentary beginning of
one - it's that they aren't even moving in that direction.
They will move for now in whatever direction we decide for them to
move. I think that has to be toward more autonomy.
Do you believe it is impossible to create an artificial sentient
mind given that an existence proof of sentient minds from a non-
engineered natural development cycle is all around us?
Yes, because the engineered artificial development cycle not only
doesn't have four billion years to spare, it also doesn't have
pressures in the right direction.
Four billion years are not required. We have significant pressures
toward greater intelligent processing and faster responses over ever
larger information sets than humans, even currently or likely soon
augmented ones, can achieve. Some of that intelligence in some
realms may logically require a sense of self and considerable world
model on the part of our software/hardware creations.
The market doesn't reward intermediate steps, and doing the whole
thing in one go is utterly impossible. Maybe given $1 billion/year
for a century, it might be possible to create an artificial
sentient mind, but the market won't support that.
The market certainly does reward intermediate steps if they have
profitable or useful consequences. The history of narrow AI
applications is full of examples.
Now, by "is impossible" I don't mean "will forever be impossible".
Maybe in 10 million AD some kid will hack one up on his Jupiter
brain in between coming up with 65536 nontrivially distinct axiom
sets in which a proof of the continuum hypothesis exists for maths
homework and carrying out abiogenesis in a nonpolar solvent for
chemistry homework, before going out with his family for dinner on
savory beta particles at the Betelgeuse Pulsar Restaurant. I don't
know, none of us can forecast that far into the future. But
impossible for us to accomplish now, this century? Yes.
We won't get to that kind of tech or survive that long without vastly
more intelligence than we have today. There is no 10 million AD
future for us at our current intelligence levels. There is
considerable room between today's AI applications and a Jupiter brain
- considerable intelligent capability that is economically lucrative
and greatly needed.
Meanwhile, real life progress continues to consist of ever more
sophisticated boxes that process data according to keyboard input
and mouse clicks.
Do you think that is the only progress being made or possible to
make in the entire universe of software? Admittedly on a bad day
that can seem like the only stuff being decently funded.
I think it'll stay the cutting edge. Embedded stuff trails a long
way, partly because the reliability requirements are such as to
flush productivity down the drain and partly because hard real time
means our best tools and techniques are limited or completely
unusable.
Still, cake would be nice but we can live on potatoes, so we
shouldn't complain too much if life offers a hope of the latter but
not the former. The stuff we really need - smart CAD, design rules
checking, smart search and data integration, process design and
monitoring etc, to stem the loss of fifty million lives a year, to
get us off this planet - can be done on beige boxes.
Not with today's software or even almost all programming done by
humans. Getting us off this planet is highly unlikely as long as
we are pretty much as we are today. Our beige boxes (hardware and
software) are not up to the task of making it significantly more
likely. Maybe with warp drive or at least very capacious generation
ships and a good supply of earth-like worlds it might make some sense
for a mass diaspora of current design humans. Maybe. But it is
very prohibitively expensive to support a lot of biological embodied
humans in space or on non-terraformed worlds unless they have
extraordinary relevant skills. The market sure as hell is not
moving in that direction.
And it is cool stuff compared to what most of the IT industry works
on, let alone most of the global workforce. Think of it this way: I
bet the alchemists were pretty depressed when they figured out they
weren't going to transmute lead into gold. But looking back, we
know they _shouldn't_ have been - the products of real chemistry
ended up being much, much cooler even if they did take awhile
coming. Maybe someone in 3007 will read these archives and think
the same way about us.
Without AI or such IA to be almost the same thing I don't have much
reason to believe humanity will see 3007.
- samantha
-----
This list is sponsored by AGIRI: http://www.agiri.org/email
To unsubscribe or change your options, please go to:
http://v2.listbox.com/member/?member_id=4007604&user_secret=7d7fb4d8
