Learning by SuccessLearning by Success

The considerable sums being spent by the fathers of electronic computers for their young hopefuls are entered in the ledgers under the heading of "school fees." Those automatic calculators which don't or won't learn will not be moved up to the next class. And those programmers who don't succeed in teaching their machines how to learn will over the short or the long run - perhaps - be left with them on their hands.

It is indisputable that computers can be taught to learn. Large numbers of scientific treatises are already in existence on the subject, and taken all in all they undoubtedly make a great deal of sense. But there is still no computer in existence capable of learning so efficiently that it is very useful to us because of its learning ability. The reason is simple: human beings learn throughout their lives, and successive generations of people have been adding to their knowledge as long as there have been human beings - but what goes on in their heads when they are learning is a thing about which, unfortunately, we are still inadequately informed. So, we cannot use what little we do know about human learning processes as the basis of a learning program for an electronic computer. The capacity of a computer for learning at all is therefore still a primitive and undeveloped talent, and it does not exactly rejoice the hearts of its instructors.

But can there be such things as machines that learn? The question once again is: is learning a process that requires intelligence?

The answer is that there are very simple learning processes which need no intelligence at all. For example, a man may memorize the contents of an encyclopaedia by heart from A to Z and still be a fool. (Would he take the trouble to learn all the volumes by heart if he were not? Every theatrical agent knows of people with photographic memories who are as stupid as they come. It is not hard either for electronic computers to learn thirty thousand irregular verbs in various languages and their meanings in French and German in the bargain.

The expression "learning power," however, generally conveys more to us than the mere absorption and retention of the contents of dictionaries. What we primarily understand by the term is the ability to learn by experience, to memorize certain situations and actions and their consequences, so that when we encounter similar situations in the future we shall be able to draw the right conclusions and act accordingly. We talk of 'learning by success." Computers are already quite capable of doing that today, though only within the narrow limits set by the programmer.

The Mouse and the Cheese >>>

© by PhiloPhax & Lauftext & Redaktion Lohberg
Kybernetik - Was ist das?

First printed in Germany: 1963


Cybernetic Computer and Electronic Brain

The fascinating story of how computers work in clear, non-mathematical language