Re: How much intelligence?

Tony Orlow <aeo6@xxxxxxxxxxx> wrote:
Curt Welch said:
"chadmaester" <chad.d.johnson@xxxxxxxxx> wrote:
What other approaches are there besides the behaviorist approach to
language?

The Behaviorist approach believes that language is nothing more than a
learned behavior which is learned the same way we learn all behavior -
the same way a dog for example can learn to respond to the sit command.
This implies we have one type of behavior learning hardware that is
used for language as well as everything else we learn (like walking).

The anti-Behaviorist view grows from the idea that humans are special
because of their language powers (which everyone agrees) and since
lower animals don't have the type of language powers humans have, then
there must be something very different about humans which give us this
power. The assumption is that if we learn the same way dogs learn,
then dogs should have the same language skills we have. But they
don't. So the conclusion is that we must have special language
hardware in our brains to explain our advanced language skills (which
is obvious), which must use a different system than reinforcement for
learning (which is not obvious). From there, the theories about how
the language hardware works, are infinite,

Skinner and Chomsky are see as the fathers of these alternate views.
There's been a divide in the field ever since. Skinner was a
Behaviorist, and Chomsky is a Linguist.

http://en.wikipedia.org/wiki/Noam_Chomsky

http://en.wikipedia.org/wiki/B._F._Skinner

The truth, like always, is probably somewhere in the middle.

For AI, it doesn't hurt to study both views because in the end, we do
have to build a language machine that has it's behavior shaped by
reinforcement.

My preference for a reductionist view makes me lean heavily towards the
Skinner side because I don't believe that evolution gave us one type of
learning hardware to learn to walk and a different type of hardware to
learn to talk. I believe we have stronger language powers than the
lower animals because evolution gave us more of the same learning
hardware connected in a way to allowed it to be used for language. So
I think the solution to creating human behavior in a machine will be a
single type of learning technology that has the power to learn
everything from walking, to talking. And I believe it will get it's
direction from pain and pleasure inputs (aka a reinforcement learning
machine). But how do you build, and configure such a machine, to allow
it to learn such a large and complex set of behaviors? That's the
question that needs to be answered.

In general, I think that a majority of people today feel that Skinner's
view of behavior was oversimplified and that human behavior in general
is just too complex and too advanced to be explained by reinforcement
learning alone. Maybe pigeons work that way but not humans is a common
mantra you will hear. Most people I think would tend to say that
Behaviorism failed. The work done in AI with reinforcement learning
machines over the years generally has led to the conclusion that trial
and error learning (another way to look at reinforcement learning) is
just too simple and not a strong enough learning theory. This has
caused the field of AI to branch off in all sorts of different
directions trying to understand the nature of the beast we are dealing
with (studying the very nature of language and knowledge for two
example).

I however am part of a minority that thinks Skinner had it right and
that it's just societies overinflated ego that keeps it from believing
that man is nothing more than a reinforcement learning machine.



If you want to characterize natural language, and all the other behaviors
we exhibit that most other animals don't, as being the result of simple
expansion of the same capabilities as they have, then you have to show
how such abilities rely on some certain critical level of such
capabilities or structures in order to function, without requiring any
addtional structures. Up until our rational level level of thinking,
associative learning dominates, as demonstrated by Pavlov. But,
associative learning is emotional, where an input causes a positive or
negative feeling by virtue of its direct or indirect association with
pleasure or pain, and that feeling causes us to approach or flee the
cause of the input. It's all about what is good or bad, as directly
experienced, and what is associated with those good or bad things, as
directly experienced. We still operate this way, but not exclusively.

When we speak of natural language, there is no emotion necessarily
involved. We can say, "There are seven red bananas on the counter," and
not care, while noting the fact. Language is not emotional persay, and
not concrete. It's an abstraction. Associations are relations between
things, but are not true abstractions of concrete realities into symbolic
format. In order to learn language, the first thing that is necessary is
the ability to maintain an abstraction, a variable that refers to
something, rather than simply associating feelings with its presence or
absence.

Then, as Lester points out, there are considerations of syntax and
grammar which govern the structure of the language itself, regardless of
meaning. There are universal components to natural language which the
comparison of various languages shows can be handled in a variety of
ways, but which have the same underlying roles. We have objects of
various levels of concreteness, for which we need nouns. We have changes
to those objects, things that happen to them, for which we need words
called verbs. We have attributes of both nouns and verbs, which we call
adjectives and adverbs. Then we have qualifiers for nouns and verbs, such
as articles and prepositions in English, which are handled differently in
different languages. It's all about describing objects and what happens
to them in space and time, or in whatever dimensions non-physical objects
operate in, and it depends on there being abstract representations of
such objects and events, and their atttributes and qualifiers, which we
can manipulate using the syntactical rules of the language.

So, Curt, I disagree. To latch on to associative learning as the single
core mechanic of mind is a mistake in my opinion. What separates us from
other animals is not just brute neuronal count. It's a different
configuration that allows us to formulate representations of realities
which can then be manipulated according to logical symbolic rules, which
are represented by the same mechanism. Each such representation acts as a
symbol. Language is strings, not feelings.

Granted, sometimes logic can be simulated by associative learning, like
the horse that stamps out the answers to simple addition problems, which
it has heard before and learned the answer to, but that's fake, and not
an explanation of what logic really is, or how it manifests itelf in all
the ways that separate us from other animals. Sorry for the length.
Enditem.

:)

So, chadmaster, above you will find a typical anti-behaviorism argument.
As I said, I seem to be part of a minority.

I can answer ever objection he raised, but it won't do any good, just like
Skinner wrote an entire book answering the same objections 50 years ago,
and it didn't do any good. People for the most part feel that they are
better than animals and that their power of rational thought elevates them
above the emotional behavior of animals. And as such, they make up
arguments like the above to support their belief that we aren't just
reinforcement learning machines.

The only answer to the type of argument stated above by Tony is to actually
build a reinforcement leaning machine that can learn to use language the
way humans use language. Nobody has done that.

Everyone understands that brains are learning machines of some type but
nobody knows with what type of learning machine we are. Is it lots of
different types of technology's welded together, is it one general
technology, or something between the two? This is what the field of AI has
been trying to get a handle on for 50 years and though the field has
produced a large amount of knowledge and new technologies, we really don't
know anything more today than we did 50 years ago about what is the right
technology because nothing yet produced has managed to act like a human.
There have been lots of promising results, but no answers.

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

Relevant Pages

  • Re: Strong AI Thesis (No Chinese room, I promise)
    ... with the idea of a generic learning system at work. ... make vision work better, and language work better, etc. ... Logic, mathematics, and scientific method ology have ... recognized the limitations which linguistic practices im pose on ...
    (comp.ai.philosophy)
  • Re: Gradual Learning, not Reinforcement Learning
    ... You seem to believe that "learning" means acquiring behaviours the system has never exhibited before. ... You use learning HORSE as an example, and posit HOXEL as an intermediate stage, with HORSE as an "ideal" to be achieved. ... Firstly, the system must be capable of "producing language", else it cannot learn the semantic difference between HORSE and HOTEL, both of which are easy consequences of HOXEL. ...
    (comp.ai.philosophy)
  • The Gaelic Society of Moscow speaks out!
    ... TEN REASONS FOR LEARNING IT ... ©Gaelic Society of Moscow,2003 ... language if you have few opportunities to use it in addition, ...
    (soc.culture.scottish)
  • Re: The Gaelic Society of Moscow speaks out!
    ... TEN REASONS FOR LEARNING IT ... ©Gaelic Society of Moscow,2003 ... language if you have few opportunities to use it in addition, ...
    (soc.culture.scottish)
  • Re: Opinions on intro lisp books
    ... But Lisp is a little different, ... Some languages support one style of programming better than they ... Even if that weren't the case, I'm not sure that a language being a ... I don't believe that learning to program in CL requires more theory ...
    (comp.lang.lisp)
Loading