Re: Why don't we have a stong AI by now? (To Curt Welch and other)

Doc O'Leary <droleary.usenet@xxxxxxxxxxxxxxxxxx> wrote:
In article <j21kt9$qtp$1@xxxxxxxxxxxxxxxxx>,
Burkart Venzke <bv1@xxxxxx> wrote:

Therefore we should defines aspects that seem to belong to
intelligence, intelligence as goal as the flight was a goal.

Yes, but while that *has* been done, I argue that the aspects commonly
used (e.g., learning, natural language understanding, etc.) are at too
high a level to actually address the underlying mechanism of
intelligence. Just because one aspect of an intelligent system is X,
that doesn't mean that a system with X will necessarily exhibit
intelligence. While one might hope the study of X would shed some light
on the underlying nature of intelligence, that does not seem to be the
way most AI research is going.

I think you have made your point. But in all this, you have put forth no
suggestion as to what "I" might actually be - or even what sort of thing it
is so as how we might describe it. Do you have any notion or rough idea
what it is you think you are looking for?

For example, I think the brain is just a single processing machine that
makes the arms move in reaction to the sensory data. So understanding it
as a signal processing machine is a high level place to start to understand
what we are trying to describe. Do you agree with that, or do you suspect
there might be something even more mysterious than neurons making arms move
at work creating our intelligence?

The wrote about of "understanding of what intelligence is". A problem
is that is no clear definition or borderline to "not intelligent", one
may define one, me or you another one. We can see it in the weak AI:
Chess playing once was seen as intelligent, in the meantime it is not
(by Fritz and other).

This misstep is the heart of my argument. Playing chess *can* be an
intelligent activity, if you actually *do* set out to explore the
boundary between what is and is not intelligence. Instead, people made
the goal of their research to be *winning* chess games, and poured a lot
of hardware, software, and human ingenuity into attacking that problem.
Investigation into actual intelligence fell to the wayside.

Yeah, I think there were high hopes that "intelligence" would be solved in
the early days of AI, but then when that failed to happen, people just
choose to pick projects that were related, and focus on them, with no real
expectation they were solving the intelligence problem. The field of AI
just fragmented into a field of building machines that could perform some
limited domain task that only a human could perform, without any real care
or concern about how well it fit into the bigger picture. After all, then
needed projects they could complete, so they could continue to get funding.
Most the field of AI is sill in that mode.

My work is a self funded hobby, so I never had to worry about taking short
cuts to get results. I've only ever worked on the bigger picture of trying
to understand the correct founding principles of what the machine we call
the brain is doing.

Do you know that there are different types of well defined machine
learning models? Again, we should use them to go forward to a broader
model of learning.

No, we should use them to go backwards to a finer understanding of
intelligence. Again, we are on the wrong path; going forward is not
progress.

Can you imagine something intelligent which can not learn? I cannot!

I can imagine something that can learn and is not intelligent.

Right, that's a valid point.

I don't
even have to imagine it; we're surrounded by those kinds of machines
every day.

--
Curt Welch http://CurtWelch.Com/
curt@xxxxxxxx http://NewsReader.Com/
.

Relevant Pages

  • Re: What are some good beginner books?
    ... learning new things, so I like having books on the subject as well. ... I have read Learn Chess: A Complete Course and am reading Logical Chess: ... The former I found extremely helpful in learning the basics, ... Is it to gain a greater rating, more understanding of the game, or what? ...
    (rec.games.chess.misc)
  • Re: Why dont we have a stong AI by now? (To Curt Welch and other)
    ... The definition of a reinforcement learning machine operating in a high ... I don't need a chess playing ... Dualism is an illusion that many people are tricked by. ...
    (comp.ai.philosophy)
  • Re: close table restore form
    ... better my product and my understanding of Access. ... NHNF and thanks for your advice! ... I'll point out potential issues with learning to use Access & creating ...
    (microsoft.public.access.formscoding)
  • Re: Robotics, AI, and Ethics
    ... weights in the network. ... your agent which shows it's a REALLY BAD learning system. ... I'm glad you think that's how the brain works. ... It's like trying to design an airplane without first understanding the ...
    (comp.ai.philosophy)
  • Re: Ben G on reinforcement-learning and the wirehead problem
    ... Yes, all human behaviors like driving cars, playing chess, ... and so on are learned but it is the innate abilities you can't ... will be of use to the generic learning hardware. ...
    (comp.ai.philosophy)