Measuring Learning LO19775

John W. Gunkler (jgunkler@sprintmail.com)
Mon, 9 Nov 1998 09:53:59 -0600

Replying to LO19762 --

A student asks:
> 1. Is is possible to measure the rate at which we learn?
> 2. Can we measure the rate at which anything learns i.e computers,
dolphins, chimpanzees?
> 3. Could the Hubble constant be a reflection of the expansion of
knowledge as well as the universe?
> 4. What does the IQ test measure and how are values assessed?

Intrepidly, I'll try to answer:

1. Yes.
2. Yes.
3. Yes, but only by coincidence.
4. IQ.

Seriously, I think it useful to remind myself from time to time that
"learning" is not really an intransitive verb. That is, one doesn't just
"learn," one learns something.

With that in mind, if you define what the "something" is that someone (or
a chimp or dolphin or computer) is to learn, and reasonable criteria for
what successful learning of it looks like, it is usually not too difficult
to come up with measures of the rate of that learning.

To take a simple example, in early 20th century psychology laboratories
people often studied paired-associate learning. The "something" to be
learned was a list of paired objects (often nonsense syllables, in order
to try to rule out previous learning effects.) The criterion was the
ability, when given one of the objects of a pair, to respond with the
other (paired) object. The "rate" of learning could be measured either
in terms of actual amount of time spent or, more often, in terms of number
of repetitions of the "learning list." This same kind of learning has
been studied in chimps, dolphins, elephants, rats, pigeons, and other
species.

In computer learning, what is to be "learned" has sometimes been to do
some task for which the computer has not been explicitly programmed.
There are "learning programs" (and much more sophisticated artificial
intelligence programs, such as neural networks) that do exactly this.
Learning rate is, I believe, usually measured by the number of "learning
examples/cases" needed before correct task execution can occur.

I will not get into the IQ controversy here (since I don't wish to make it
my life's work.) The best definition of IQ I ever heard is that is "what
IQ tests measure." I'll only comment that, much like my warning about the
transitive nature of the verb "learn," I believe that an attempt to create
some generalized measure of "intelligence" (or even "ability to learn") is
a waste of time. Intelligence, whatever it is (or, as many are saying
these days, whatever several things it is), can best be measured by the
results of its application to specific tasks. It is common experience
that people have differential abilities at different kinds of tasks. It
has always seemed to me that the more specifically one defines the task,
the more useful a measure of ability to do it becomes.

-- 

"John W. Gunkler" <jgunkler@sprintmail.com>

Learning-org -- Hosted by Rick Karash <rkarash@karash.com> Public Dialog on Learning Organizations -- <http://www.learning-org.com>