statistics.bgwoodruff.com > Course Materials > Manifesto > Here

On the Importance of Learning From Your Errors
A Compendium of Quotations

Please email me others.


A summary statement of Harry Harlow's "Error Factor Theory"

An animal learns to make the correct response by learning to not make incorrect responses. That is, the animal learns to stop making incorrect responses. Then, all that is left is the correct response.

Harlow developed his learning theory while working with Rhesus monkeys. He gave the animals a long series of six-trial two-choice discrimination problems.

  • For example, the monkey would have to choose between a toy firetruck and a plastic cup. One of the two stimuli was "correct" --when the animal pushed it aside, the food well underneath the stimulus contained piece of food.
    • The problem was repeated for 6 trials using the same two stimuli. Whichever stimulus was correct on the first of the 6 trials remained correct for the remaining 5 trials.
    • The next problem used two new stimuli presented for 6 trials.

On the first few problems (not trials) the animals learned very little over the six trials. On the sixth trial the animal was correct only about 70% of the time. On the second trial the animal was correct only about 50% of the time. AND

He discovered after 200 problems (NOT trials), the monkeys were always correct on the second trial.

By looking at the trial-by-trial and problem-by-problem data records, Harlow surmised that the animals had to learn not to make certain kinds of errors.

  • Some animals had a position preference. Such animals tended to choose the stimulus on the right (or left). Since the position of the correct stimulus was randomized trial by trial (always on the left for 3 of the 6 trials, but a randomly determined three trials), such a monkey would get reinforced on half the trials. To get rewarded on every trial, these monkeys had to learn that responding to the position was an error.
  • Other animals had a color preference. If one of the stimuli was red they would always choose it. That worked about half the time one of the objects was red. These animals had to learn to quit responding to the color per se.
  • Harlow discovered a variety of error patterns in the monkey's performance: win stay, lose shift, alternate, etc.

As an animal eliminated these error factors (error tendencies), performance would get better as now the animal was attending to the object being chosen. After abut 200 problems, the animals had solved the problem: on trial 2 stay with the winning stimulus; choose the other one if you lost.

I think of this principle in terms of an story told about Michelangelo. He was asked how he could ever carve such a masterpiece of sculpture such as the David. He said it was pretty simple. All you had to do was chisel off all the rock that wasn't David, and there it was.

All you have to do to learn is to learn what not to do, and what remains is the correct performance.

Incidently, you may remember that Harry Harlow was the scientist that conducted the research with baby monkeys relating to cloth and wire "mothers."

We make progress if, and only if, we are prepared to learn from our mistakes.

Karl R. Popper

And if we are not afraid to make mistakes from which we can learn.

Kelvin Throop

Mistakes are seldom serious unless repeated.

Anonymous

No [person] ever became great or good except though many and great mistakes.

William E. Gladstone

An expert is a man who has made all the mistakes, which can be made, in a very narrow field.

Niels Henrik David Bohr (1885-1962)

Anyone who has never made a mistake has never tried anything new.

Albert Einstein

Good judgement comes from experience, and a lot of experience comes from bad judgement.

Anonymous

Experience is the name everyone gives to their mistakes.

Oscar Wilde

Experience is that marvelous thing that enables you to recognize a mistake when you make it again.

F.P. Jones

Modified Sunday, March 25, 2007