Re: Curt's AI epiphany

"Curt Welch" <curt@xxxxxxxx> escreveu na mensagem
news:20081209132032.725$w9@xxxxxxxxxxxxxxxxx
"PeskyBee" <peskybee2@xxxxxxxxx> wrote:
Among many properties of consciousness, being conscious
requires that the organism distinguishes between itself
and the environment that surrounds it. Computers don't have
such capability (so far). Other animals have very limited
abilities. Chimpanzees can recognize themselves on a mirror,
so they must be assumed to have partial consciousness. But
they fail to understand many of other animal's intentions
(such as when one points to a box having a candy behind it).
Interestingly, dogs fail on the mirror test but succeed
in the pointing task. These are all well documented by
ethologists, which can be used as criteria for a scientific
definition of consciousness.

Yes, real physical tests like those could be used to define consciousness.
If only people actually based their ideas of consciousness on such things
there would be no hard problem.

However, I disagree completely with your conclusions about what they are in
fact testing for.

In fact, I must mention again that I don't give a minuscle
rat's ass for this consciousness stuff. I'll be delighted to
meet a machine that is as intelligent as a mouse and as conscious
as a rock. At least, I would have no trouble shutting the thing
down when it comes to pee in my shoes.


Current computers, on the other
hand, fail in all these accounts. They have no perception of
their environment,

Not true. Every computer has some percpetion of its environment. The
computer sitting in front of me has a percpetion of when it's keys are
being pressed for example.

In such a broad use of "perception of environment", my
digital clock (the one that I try to hammer every
morning when it tries to wake me up) is a "perceptive machine".
I don't think this compares well even with those pesky
mosquitoes. And of pesky insects I'm a real connosseur.


they don't model what they receive from
any kind of input,

A typical office PCs don't do much modeling. But our AI robots sure do.

Let's say that AI robots will, one day, in the far future, do.
I'm still waiting to be amazed by any of such things.


they don't know "who they are".

Most AI robots model themselves as part of the environment they model.

Well, then these robots would be high on my scale. I still
am waiting to see such a thing.


It is not
enough to potentially have the "brain machinery" to be conscious
(as any comatose patient will demonstrate).

Yeah, the power switch needs to be in the ON position as well or else the
AI robot will be comatose.

But the problem is that even with the switch in the ON position,
without adequate software these AI robots would be less than
a firefly.


All this talk of
consciousness is just an incredible loss of time: we've got to
make computers behave like a pigeon (but not like the ones in
behaviorist labs!).

I came to comp.ai to see if I could find people who wanted to talk about
how to build robots that acted like pigeons. What I found was that few
people had a clue what to work on, and the few that had some clue, tended
to be working on the wrong problems or approaching the problem of AI from
the wrong direction. What I've ended up doing, is mostly trying to get
people on track which turns into these endless debates about the philosophy
of AI.

I enjoy these debates, and they help me understand things I didn't
understand before I came here. But most of what it helps me understand is
how confused everyone else is, instead of helping me make progress on the
real problem. Still, I've learned a lot of things that are in fact helpful
to my AI work in these debates so it's not all a waste of time. In
addition, it helps motivates me to work on AI, and that's a good thing.
There's nothing like having people like Alpha call me an idiot for the
200th time to remind be I've got work to do! :)


I can also list a lot of confused approaches to AI that have
stalled the development of the whole thing, IMHO. Logic and Cyc-like
things, for instance, are still the hopes of many people, and
in my vision these are severely problematic.

I would say that the main problem of AI is not much to select
a "knowledge representation", or to design a proper "neural net".
The main problem is how to decide about the most adequate level
of analysis. Logic, for instance, is too "high level". Neural
nets are too "low level". But no matter what the level, we've
got to think in terms of inner actions and functions. And hope
that the approach we develop will give the correct "emergent
behavior".

*PB*






.

Relevant Pages

  • Re: Curts AI epiphany
    ... and the environment that surrounds it. ... Computers don't have ... so they must be assumed to have partial consciousness. ... Most AI robots model themselves as part of the environment they model. ...
    (comp.ai.philosophy)
  • Re: 50 years later, Marvin Minsky still doesnt get it
    ... Ethics and rights have nothing to do with consciousness. ... humans react to their environment. ... you have a robot with a computer of similar power to a human brain then ... blenders and rocks and computers, ...
    (comp.ai.philosophy)
  • Re: Construction of an Edible Thing
    ... Each robot is pre-programmed to collect things from the environment at ... Now let the robots free to interact ... are easy to construct from non-edible things, so the construction ... construct consciousness, starting with a non-conscious device, which ...
    (talk.origins)
  • Re: Why human colonization may be irrelevant
    ... Consciousness does not require awareness of self state, ... identification of the environment. ... Nevertheless, computers are still ... We've got computers that play chess ...
    (sci.space.policy)
  • Re: Construction of a Conscious Thing
    ... Construction of a Conscious Thing ... It shows how consciousness might be produced from non- ... Start off with millions of identical ordinary robots. ... Each robot is pre-programmed to collect things from the environment at ...
    (talk.origins)
Loading