100 YEARS AGO

Captain Dutton's valuable memoir on the Charleston earthquake of 1886 contains many accounts of the effects of this great earthquake on human beings. Nowhere could they be more vivid than in Charleston itself. “On every side,” says one witness, “were hurrying forms of men and women, bare-headed, partially dressed, some almost nude [the earthquake occurred at 9.51 p.m.], and all nearly crazed with fear and excitement. . . A few steps away, under the gas-lamp, a woman lies prone and motionless on the pavement, with upturned face and outstretched limbs, and the crowd which has now gathered in the street passes her by, none pausing to see whether she is alive or dead. . .; many voices are speaking at once, but few heed what is said.” . . . Captain Dutton also gives many records of a feeling of nausea at the time of the earthquake; and, however excitable the observers may have been, these accounts are probably trustworthy, for this is not at all generally known to be a result of earthquake-motion.

From Nature 13 December 1900.

50 YEARS AGO

In the account in Nature of October 14 of machines which carry out strategic sequences of moves, it is stated: “No machine can learn from its mistakes — to improve the play the programme must be improved”. It is, of course, true that a machine cannot learn unless it is provided with a programme or mechanism for learning. But it is quite possible to devise such a mechanism. Machines can be designed to make the best move at each step in a game of noughts-and-crosses or (in theory) draughts or chess. But when playing against a series of human opponents, such a machine may never do much better than draw. A good human player against the same opponents may score more wins by making unsound but more puzzling moves. Can a machine be made to imitate the human player? . . . It can, by the inclusion of an empirical or statistical mechanism, in three units. One unit makes the machine experiment with different alternatives each time certain positions are reached; the second unit counts the results and relates them to the alternatives chosen; and the third steers the machine into the lines of play which have been winning most often.

From Nature 16 December 1950.