The obvious comment on this helpful exposition is that it expresses your
philosophy of how AGI's MAY POSSIBLY be grounded. As yet there is no
demonstration that this or any comparable specifically-for-AGI grounding
actually works or can work.
The discussion has been largely about psychology and theories of how the
human brain IS grounded, - and evidence for and against.
(Some others in the discussion have been confusing the two domains too).
Why should you or others care about how the human brain is grounded &
psychology when - as certainly many here believe - you can short-cut the
developoment and evolution of the human and animal brain?
Because the brain didn't evolve in the way it did for arbitrary reasons.
There is , I would argue, a NECESSARY structure about the grounding of
perception and understanding in the living brain (and their development) and
the kinds of sign systems we use. I have tried to argue - although it would
take considerably more exposition to do the argument justice - that there is
a necessary structure to how you represent, generalize about , compare and
draw analogies between objects (and the contents of the world) - using
symbols, graphics and images - as scientifically necessary as geometry is
to perform its particular, relatively limited set of operations in
representing and analysing objects.
I am quite sure that neither you nor Ben - nor indeed anyone or any
science - has yet an adequate understanding of these matters. You are
placing a bet, but it isn't anywhere near informed enough a bet.
I think - correct me - there's an evolutionary developmental robotics or
somesuch which works in close concert with psychologists studying
development. (There is, I'm just not sure of the name). You guys, I
suggest, should do similar.
.
Hi,
The current discussion on symbol grounding, to me, includes several
different (though related) questions. In the following, I'll try to
separate them, and give my opinion on each of them.
*. When is a symbol "grounded"?
A symbol is grounded when its meaning to the system is determined
according to the system's experience on the symbol.
*. What is wrong in traditional "symbolic AI" on this topic?
In those systems, the meaning of a symbol is determined by an
"interpretation", which takes the meaning of the symbol to be an
object/entity in the world that is "referred to" by the symbol. In
this way, the system's experience plays no role, and a symbol can be
interpreted in many different ways.
*. What is the "experience" needed for symbol grounding?
Any input-output activity that happens between a system and its
environment.
*. For the symbols in an AGI to be grounded, should the experience of
the system be the same, or very similar, to human sensory experience?
No. The system can ground its symbols in experience that is very
different from human experience, as far as "intelligence" is
concerned.
*. Is vision necessary for symbol grounding in AGI?
No, for the above reason.
*. Is vision important in deciding the meaning of human concepts?
Yes, since vision is a major channel of human experience, the meaning
of many human concepts include visual components.
*. In that case, if an AGI has no vision, how can it still understand
a human concept?
"Understanding" is a matter of degree. Since the meaning of a symbol
is determined by the system's experience about it, it will have
different meanings in different systems, though as far as the systems'
experience have overlap, the symbol will have common meaning in these
systems. If an AGI's does not have visual experience, it won't
understand a concept exactly as a human, though its other experience
channels may allow the understanding to be close to a human
understanding.
*. Can a blind person to be intelligent?
According to the above opinion, a blind person can be perfectly
intelligent, with symbols grounded in (non-visual) experience.
However, there will always be some difference in what certain concepts
mean to such a person, compared to the "normal" people.
*. How can a sensorless system like NARS have grounded symbol?
In principle, as far as a system has input, it has sensor, though its
sensor can be very different from human sensors. The mistake of
traditional symbolic AI is not that the systems have no sensor (or
have no body), but that their experience play no role in determining
the meaning of the symbols used in the system. Since in NARS the
meaning of symbols (i.e., how they are treated by the system) is
determined by the system's experience, they are grounded. Of course,
since NARS' experience is not human experience, the same symbol
usually have different meaning to it, compared to its meaning to a
human being.
*. If NARS always uses symbols differently from typical human usage,
can we still consider it intelligent?
Yes we can. Even among human beings, the same word often means
different things --- just see what happens in this mailing list! We
should not treat "different understanding" as "no understanding". Very
often, my understanding of English is still different from a native
English speaker, but I guess I can say that I understand English, in
my way. For this reason, when I meet someone who have a different
understanding on a concept, I usually don't conclude that he/she has
no intelligence. ;-)
*. Are you saying that vision has nothing to do with AGI?
Of course not! I'm saying that vision is not a necessary component of
an AGI. Since vision plays an important role in human cognition, there
are practical reasons for certain AGI projects to include it to ground
concepts in a more "human-like" manner, though some other AGI projects
may exclude it, at least at early stage. Again, intelligence can be
achieved without vision, or any other human sensory channel, though it
will have an impact on the meaning of the symbols in the system.
More "academic" treatments of this topic:
http://nars.wang.googlepages.com/wang.semantics.pdf
http://nars.wang.googlepages.com/wang.AI_Misconceptions.pdf
Pei
-----
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/?&;
--
No virus found in this incoming message.
Checked by AVG Free Edition.
Version: 7.5.488 / Virus Database: 269.14.9/1067 - Release Date:
12/10/2007 18:02
-----
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=8660244&id_secret=53257956-37f044
