GLADWELL'S "THIN SLICING", LATENT INHIBITION, AND ARTIFICIAL INTELLIGENCE
(Posted 2006-07-10 08:29:11 by Ray Lopez)

Malcolm Gladwell's book entitled _Blink_ was on the bestseller list for a
long time, and for good reason. It was amazing to most people that humans
can do so much decision-making wholly outside the realm of consciousness,
and outside the realm of what we usually think of as "reason." Yet, these
quick decisions, these types of "thin slicing", to use Gladwell's term,
make up a considerable amount of the daily information processing that
humans engage in. Not only that, but there's the fact that these types of
decisions are more often than not the correct ones to be made in any given
situation.

Gladwell's work is important because it shows us that humans are able to
make very rational and very correct decisions without engaging in any form
of conscious reasoning. Gladwell's work is in an area of experimental
psychology called "social cognition." It is likely that social cognition in
general, and thin slicing in particular, might play important roles in
helping humans to think the way we do.  Indeed, I would argue that this
type of reasoning is a most fundamental component of our reasoning ability,
and would further argue that it should be a fundamental component of any AI
system, whether that system seeks to model human thinking, or derive the
most rational answer to a problem. 

Even our highest forms of thinking, such as theoretical physics, are driven
by an aesthetic component that is beyond our ability to articulate cleanly.
Steven Weinberg, Nobel laureate and professor of physics and cosmology at
The University of Texas, developed most of the current paradigm of
theoretical physics, the so-called "standard model." Dr. Weinberg has
written extensively about the search for better theories to explain our
physical world.  According to Weinberg, the most important aspect of any
physical theory is that it be "beautiful" and "elegant". But what
constitutes beauty and elegance? Weinberg doesn't know, but he knows it
when he sees it, as do all other theoretical physicists. This notion of
beauty and elegance in a world as complex as theoretical physics speaks
volumes about the importance of our abilities to thin slice, to pick out
those cues that are most important, and the importance of those mechanisms
that underlie thin slicing.

Gladwell's thin slicing may become an important area of AI research in
coming years. If it does, there will be a need to define some type of
mechanism by which thin slicing can occur. Thin slicing may be mediated by
a process called latent inhibition. This process has been studied by
experimental psychologists for a very long time, going back as far as
William James. Briefly, latent inhibition is that process by which we learn
the features or characteristics of any given thing that are LEAST important
to helping us reach some sort of decision about that thing. For example, in
Gladwell's book, he writes about the abilities of antiquities experts to
discern between real antiquities (such as a Greek statue) and well-made
fakes. These experts have honed their knowledge over many, many years of
looking at and talking about antiquities. If the process of latent
inhibition underlies their abilities to instantly thin slice a real statue
from a fake one, then we may hypothesize that their many exposures to real
antiquities and fake antiquities have taught them the ability to ignore
those features that have nothing to do with telling a real from a fake. 

We suspect that latent inhibition underlies this ability because it has
been studied extensively by experimental psychologists. In one landmark
study, Japanese "chicken sexers" were studied to understand their ability
to tell the difference between recently hatched male and female chicks.
This is a very difficult thing to do. If a non-trained person were to pick
up a baby chick and hold it the way the chicken sexers do, and try and
guess what the sex of the chick is, they would do no better than chance.
But Japanese chicken sexers are able to tell, very quickly, whether or not
a given chick is a male or female.  The way they are trained is relatively
straightforward. They simply look at lots of baby chicks, under the
guidance of an experienced chicken sexer.  Over time they acquire the skill
necessary to tell the difference between male and female chicks.  The
process of latent inhibition works by actively inhibiting the learning of
those cues which do NOT matter in telling one thing from another. In the
case of chicken sexers, latent inhibition works by inhibiting the
association of specific cues that have little or nothing to do with telling
the difference between male and female chicks.

The particulars of how latent inhibition works are still being researched.
At present, there is no doubt that our brains possess a system specialized
for extracting cues that are relevant for making subtle distinctions, and
for suppressing those cues that are irrelevant. It is likely that this
system underlies our ability to do Gladwellian style thin slicing.

Thin slicing is an extremely powerful tool. It enables us to make very
accurate decisions very quickly. If the process of latent inhibition
underlies this process, it is possible that latent inhibition may prove to
be a fruitful avenue of research in AI.  AI research in feature detection
and pattern recognition already employs tools and techniques that look like
latent inhibition, but other fields such as automated reasoning and machine
learning may gain much by not only building systems that focus on what to
learn, but also focus on what not to learn.

--------

There are no comments on this post.


To submit a comment on this post, email rl@well.com or visit us on the web
[ http://ratthing.com ].
.