[Since the original was posted to many lists; I will respond to all, but
since I am only a member of one of them, it is not clear which of the
other lists this will go to]
[Host's Note: You don't have to be a subscriber to post to the
learning-org list. ...Rick]
On Tue, 2 Jun 1998, Walter Derzko wrote:
> The attached press release caught my attention--what do you think the
> consequences might be ?
In my next message (to cognet only), I will try to answer Walter's
question wrt to managerial and organizational cognition.
Science announcments by press-release are always a bit suspect, and Dr.
Roy may not be nearly as silly as the press release makes him sound, but
since I am too lazy to check the original statements, I will rely on the
press release. If it has grossly misrepresented his position, then that
is something he should take up with his PR office.
Walter asks what the consequences will be:
It means that Dr Roy might get some nice lecture fees, and may even be
able to produce a book that sells. If the press release correctly
characterizes his ideas, then his work will be ignored or ridiculed by
real AI and neurology researchers, correctly.
I will try to state my reasons for that here. In doing so, it might
appear that I am defending connectionism. But I would just like to state
for the record that I have some deep reservations about connectionism.
> BRAIN SURGEON NOT REQUIRED--Expert in Artificial
> Intelligence Challenges Dominant Theory About the
> Human Brain
My two your old niece can challange any theory she wants to. The question
is whether the challenge is well founded or not. I must admit that I
immediately grow suspicious when people try to sell their ideas based on
their "radicalness".
> [much omitted, mostly asserting that the other guys have it wrong.
> Some text trying to paint the picture of a lone genius. The usual]
> A classic, flawed theory
>
> Prevailing thought draws on the teaching of Donald Hebb
> of McGill University, Montreal, a pioneer theoretician who
> postulated a mechanism by which the brain learns to
> distinguish objects and signals, add, and understand
> grammar. According to Hebb, learning involves adjusting
> the "strength of connections" between cells or neurons
> in groups of cells known as neural networks.
I find it interesting that Hebb is listed as a founder of connectionism.
What happened to say David Rummelhart and the PDP group?
> [...]
> Stumbling block
>
> The major stumbling block for future technology, says Dr.
> Roy, is that none of these learning methods reproduce
> the external characteristics of the human brain, principally
> its independent way of learning. Therefore, methods based
> on these classical ideas require constant intervention by
> engineers and computer scientists - providing network
> designs, setting appropriate parameters correctly, and so
> on - to make them work. This drawback is severe, he
> maintains.
There are two possible criticisms here. One is that connectionists have
failed to produce a machine that actually behaves like a human. That is
true, but that is a fact of all AI work. If Dr Roy claims to have
something different, I would like to see it in action.
The second is the implication that since humans learn in ways differently
then they way individual neurons are thought to "learn" there must be
something wrong with theory about neurons. While that view is not stated
explicitly, from the press release the reader is invited make that
inference.
That inference is based on a confusion of level. Human and pidgeons learn
in very different ways. But individual neurons in the brains of people
and the brains of pidgeons probably "learn" in the same way (which will be
different both from how humans learn and how pidgeons learn). There is no
reason to expect that human learning is anything like neuron learning even
if human learning is ultimately based on neuron learning.
> Instead, says, Dr. Roy, scientists must admit that their
> constructs diverge from the human brain and return to
> the original model. Drawing on the way the brain actually
> works, he has used operations research to create
> autonomous learning algorithms that are more human-like
> because they don't require ongoing input.
Show me the beef.
Dr Roy may be one of many many AI researchers who have serious qualms
about connectionism. Marvin Minksy is rumored to have once said: "I like
connectionism so much that I wish a dozen people were working on it."
Raising opjections to connectionism doesn't make Dr Roy special in any
way.
-j
-- Jeffrey Goldberg +44 (0)1234 750 111 x 2826 Cranfield Computer Centre FAX 751 814 J.Goldberg@Cranfield.ac.uk http://WWW.Cranfield.ac.uk/public/cc/cc047/ Relativism is the triumph of authority over truth, convention over justice.Learning-org -- Hosted by Rick Karash <rkarash@karash.com> Public Dialog on Learning Organizations -- <http://www.learning-org.com>