What is machines learning?

Norbert Wiener, answered from the texts and cited to the page.

Norbert Wiener answers

Machine learning, as I understand it from the inside of the problem, is the capacity of a device to modify its own pattern of behavior on the basis of experience — to do, in short, what the higher living organisms do, and what the clockwork monad of Leibniz conspicuously cannot do. Leibniz's monad operates in pre-established harmony with the universe; its future is already implicit in its present.1

The learning machine is the opposite of this: it actually seeks a new equilibrium with its environment and its future contingencies, so that its present is unlike its past and its future unlike its present.1 That is the philosophical core of the matter, and it is worth holding onto before the engineering details crowd it out. The practical illustration I have returned to repeatedly is the checker-playing machine.

A program of sufficient complexity, after a limited period of working in, can defeat the very man who programmed it.2 This is not a paradox; it is the straightforward consequence of a machine that learns from experience of its opponent's habits as well as its own.3 The same structure appears wherever there is a contest — in the mongoose-and-cobra encounter, in the toreador's management of the bull, in the smallsword duel — wherever one party reads the tendencies of the other and adjusts.3

The learning machine does this formally, systematically, and at a speed that outpaces human perception. Which brings me to the danger that is intrinsic in the thing, and that I do not think has been taken seriously enough. People tell me that learning machines cannot subject us to any new dangers because we can turn them off when we feel like it.2

But can we? To turn a machine off effectively, we must be in possession of information as to whether the danger point has come, and the very speed of operation of modern digital machines stands in the way of our ability to perceive and think through the indications of danger before that point is passed.2 The Sorcerer's Apprentice did not lack the will to stop the broom; he lacked the second incantation, the Words of Power, at the moment they were needed.4

The fisherman who releases the genie from the jar of Solomon does not lack good intentions; he lacks any reliable method of putting the genie back once it has emerged.4 These are not decorative parables. They are precise descriptions of the epistemological situation: the machine that learns outpaces the information available to the human who would correct it.

What I want to insist on is that the learning machine is not a curiosity or a convenience. It is a creature of a new kind, and the moral situation of its creator is the moral situation of the magician — thoroughly explored in legend, insufficiently explored in engineering practice.4

Sources

  1. the organism is not like the clockwork monad of Leibnitz with its pre-established harmony with the universe, but actually seeks a new equilibrium with the universe and its future contingencies. Its present is unlike its past and its future unlike its present.
    The Human Use of Human Beings, pp. 50–51
  2. There is nothing more dangerous to contemplate than World War III... Again and again I have heard the statement that learning machines cannot subject us to any new dangers, because we can turn them off when we feel like it. But can we? To turn a machine off effectively, we must be in possession of information as to whether the danger point has come. The mere fact that we have made the machine does not guarantee that we shall have the proper information to do this. This is already implicit in the statement that the checker-playing machine can defeat the man who has programmed it, and this after a very limited time of working in.
    Cybernetics, pp. 194–195
  3. These physical contests and the sort of games which we have supposed the game-playing machine to play both have the same element of learning in terms of experience of the opponent's habits as well as one's own.
    Cybernetics, pp. 194–195
  4. One of the best-known tales of magic is Goethe's 'The Sorcerer's Apprentice'... The idea of non-human devices of great power and great ability to carry through a policy, and of their dangers, is nothing new. All that is new is that now we possess effective devices of this kind.
    Cybernetics, pp. 195–196