Re: Reply to Wolf

On Jul 4, 9:00 pm, Wolf <ElLoboVi...@xxxxxxxxxx> wrote:
JGCASEY wrote:
Wolf:
Yes, there seems to be a prejudice against behaviourism.

JC:
Maybe because they don't say anything that can be translatable
into an actual program? Whereas all the cognitive talk can be
translated into actual working programs that although perhaps
not "real" intelligence has produced some programs that can do
some simple things that if done by a human would have to have
required a degree of intelligence.

Cognitive talk can be translated into a program because the programmer
talks in the same metaphors as the cognitivist. Eg, both think of memory
as data storage. Both think of symbol manipulation as a sign of
intelligence. Both think of thinking as symbols manipulation. Etc.

And so in order for programmers to take behaviourism seriously,
behaviourism would have to provide a way to map what they talk about
to structures that a computer can implement. Of course, it can't do
that, since it's just a set of use cases -- when this happens, the
system has to do this -- and no where near being a design spec.

You can protest this, but then you'd have to show some results that
could be seen as a design spec instead of simply criticizing other
people for using the wrong words ...

Make it clear and precise and it can
be programmed. In the meantime it amounts to nothing useful.
It seems to be nothing more than a play with words.

It's not playing with words, my friend. That's the problem.

"Memory" is a noun, and naive users of language therefore think it must
refer to some person, place, or thing (to quote the throughly misleading
definition from grade school.) "Remembering" is a verb (a participle
when used in a verb phrase, a gerund when used in a noun phrase, to be
precise), and therefore refers to a process, action, or change of state
(which your grade school grammar did not tell you.) See?

In animals memory is _not_ storage of data. There is no place in an
animal's nervous system that you can identify as being data storage.
There is nothing that resembles RAM or ROM, or an HDD or a floppy drive
or magnetic tape. Remembering is a conditioned behaviour, not a playing
back of stored data. That's why it's easy to make people remember things
they never did or that never happened, a fact that experienced or
well-trained interrogators know.

The closest thing to storage of data is chemical change in the synapses,
which amounts to modifying synaptic connections, which amounts to
construction of networks. But the same process goes on when a behaviour
is conditioned. When you engage in some behaviour, certain neural
networks activate (and boy, does that cover an enormous amount of
detail!). When you recall something, certain neural networks activate.
At the neural level, there is almost no difference between remembering a
painting you saw yesterday, and looking at the painting. From what I've
read, you even set up potentials in the motor cortex to move the eyes as
if you were looking at the painting. These potentials may or may not be
repressed; watch a kid remembering what it's seen, and you will its eyes
move as if it's looking at something.

Wow, you've mixed up some many concepts here that it's actually
amazing.

First, you seem to be claiming that because human memory doesn't work
like computer memory, we shouldn't call human memory memory. The
problem is that our notion of human memory came first, and we called
computer memory memory because it was LIKE human memory. In what
way? It generally ACTS like human memory. Yeah, it's generally more
permanent than human memory, but we've built computer memory to act in
most ways like human memory. It does the same things as human
memory. So why would we stop calling human memory memory if the
Johnny-Come-Lately isn't implemented the same way as the thing that
came first? You can claim it's "misleading", but your only comments
for that are about the implementation ... and memory is surely a high-
level design/algorithmic description/artifact, not an implementation
one. To claim that having memory means having a specific type of
physical implementation is not only wrong from a cognitive point-of-
view, it's even wrong from a behaviourist view.

Second, what's the same in all types of memory? Memory IS storage and
retrieval behaviour. When you memorize something, is it not to
"store" it so that you can "retrieve" it later? When I try to fix the
image of that really attractive woman in the short skirt in my memory,
you can BET that it's to retrieve that image later [grin]. So if
memory ISN'T storage and retrieval behaviour, what is it? You can
insist that the memory isn't stored like physical memory in a computer
in humans or other animals, but that's irrelevant to what type of
behaviour it is. And if you can't accept that, then it seems that
even your behavioural analysis has failed since you don't seem to be
classifying memory BEHAVIOUR properly, which is all you claim to be
able to do.

Third, your analysis of how memory is represented in a brain is
lacking. Look at analyzing computer memory at a high implementation
level. What you notice is that certain areas (the RAM sticks, or the
hard drive) change. What do you see when you look at a brain?
Certain areas of the brain change when things are memorized. So
what's so different about that? Yes, the fact that a brain is
biological and a computer is not mean that there are some differences
in the LOW-LEVEL implementation, but so what? Why would we expect the
low-level implementation to be the same in completely different
media? That's the whole point of algorithms; you can implement them
on different systems and media, and they'll look different ... but act
the same, both on an observable behavioural level and on a design
level.

BTW, writing to a hard disk is not commonly called "memory" to a
computer. Storage, yes; memory, no.

Finally, in your last point you conflate retrieving a memory with
experiencing it again. If you retrieve an image from memory, you
experience it again. The experience has certain things that go along
with it.

And even more finally, there is nothing inconsistent with calling
memory storage and retrieval behaviour and have it be malleable or
even inaccurate. That's just a feature of the implementation of
memory in humans.
JC:
Yes I am aware of that. Of course the robot may not have the
filters to make "Sweet Caroline" a different stimuli to a
buzzer sound in the first place so it may not be a big issue.
However if it could recognize the tune and if the tune failed
to usually be present with "food" wouldn't the connection fade?
It is only when a stimuli is not tested, or can't be tested
does it persist.

Buss, Sweet Caroline - what difference does it make? It was sound as a
secondary input that I was referring to. OTOH, if you want the robot to
learn how to tell the difference between a buzzer and a Neil Diamond
song, well, that's a little more difficult. ;-)

Um, but here's the thing: if we had an intelligent robot, it would
refuse to respond to Sweet Caroline because it would reason that that
could have nothing to do with getting power (if I recall the start of
the example correctly). THAT'S an intelligent decision. Getting
conditioned to the wrong stimulus is a prime example of the
unintelligence of conditioned behaviour.

Wolf:
There has to be a second level (at least) of conditioning to
prune out the stimuli you don't want. It's at that point that
training gets complicated, as anyone who has tried to train a
dog knows. You have to ensure that the dog responds to the
cues that matter to you.

JC:
This is why a higher level kind of "training" possible for
humans is desirable in a robot. Not an automatic unconscious
conditioned reflex but a conscious cognitive understanding
of what is required.

Nope. Conditioning is always "unconscious", even operant conditioning,
which is what you actually want a robot to be capable of.) That is, even
"consciously deciding to do something rather than something else" is a
conditioned behaviour (operant, in this case.) It's a behaviour that is
quite difficult to establish, as St Paul knew only too well: "That which
I would, I do not; and which I would not, that I do." Or the Spanish
proverb: "Habits begin as cobwebs and become ropes."

Now, if you insist on this, then you claim that reasoning -- always
conscious if properly called reasoning -- has no impact on learning or
on intelligent behaviour. Which means that you take all of that out
of the picture. But it's the results of that that are considered
intelligent. If you take it out, everyone will agree with you that
it's all conditioning ... but will onder if you have intelligence at
all.


As has been pointed out conditioning is common in even the
most primitive organism but I would also point out that
cognitive processing is not common and seems to have arisen
out of new mechanisms involving the cortex.

There is no reason to believe that "cognitive processing" is somehow not
subject to conditioning.

Subject to? Certainly, in the sense that conditioned responses can
impact it. Nothing but conditioning? You need more evidence before
we'll accept that, and the first statement above is not strong enough
to justify that point.


Cognitive, meaning being self reflective, as to what is happening
rather than it being an uncontrolled automatic process. If you
want to understand how that might work you need to move beyond
your c. elegans as indeed AI researchers have tried to do.

It's clear from the above that your thinking is under strong control
from discriminants such as "automatic", "controlled", "conscious", etc.

Now, is that determinant control, or influential control? This is the
same problem that I pointed out about Skinner in my critique of
"Beyond Freedom and Dignity": control is not necessarily a clear and
unequivocable term, and you seem to use it in two different ways
depending on what you want to claim.


I would also reiterate that AI doesn't have to be about animals.
It is about machines. As you say the question is do you program
something in or do you train it.

Yeah, but the study of animals can establish the features that matter.
In fact, the artificial intelligence enterprise is based on the hope
that certain animal behaviours can be at least emulated by
non-biological systems. If you don't understand this, you've forgotten
that humans are animals, and that "intelligent behaviour" occurs along a
spectrum that certainly includes many behaviours that "lower animals"
are capable of.

Now, let me issue this challenge again: Explain how to tell the
difference between an intelligent and unintelligent behaviour. If
behavioural analysis is to have anything to say about AI, it will need
to be able to provide such a definition, and as if yet I have not seen
such a definition.


.

Relevant Pages

  • Re: Why Humans Are Smarter Than Animals
    ... scan a random series of locations in its memory. ... human memory acts as if it does have instant random access capability. ... we have never heard or used that sequence before. ... What is it with fundies and kinky sex? ...
    (talk.origins)
  • Re: Why Humans Are Smarter Than Animals
    ... scan a random series of locations in its memory. ... human memory acts as if it does have instant random access capability. ... we have never heard or used that sequence before. ... fundies can only be wrong once in that few a words. ...
    (talk.origins)
  • Re: Why Humans Are Smarter Than Animals
    ... scan a random series of locations in its memory. ... physically establishes a synaptic connection. ... human memory acts as if it does have instant random access capability. ... we have never heard or used that sequence before. ...
    (talk.origins)
  • Re: Why Humans Are Smarter Than Animals
    ... scan a random series of locations in its memory. ... human memory acts as if it does have instant random access capability. ... we have never heard or used that sequence before. ... play back a sequence of events is to use a control neuron that is ...
    (talk.origins)
  • Re: Memory modulates color appearance
    ... high-level visual memory." ... Pavlovian conditioning is, of course, to the cognitivist, ... nothing - it doesn't even tell you how to produce the phenomenon. ... Another example of *real* neuroscience researchers doing real science ...
    (comp.ai.philosophy)