Science: Seeing-Eye Computer

Computers are muscling in on humans in more ways than one. Only a few years ago they were still simple-minded beasts that could understand nothing but predigested figures. Later they acquired senses of a sort: they could feel changes of temperature, hear musical tones, recognize differences of light and shade. But they could not see as humans see. A primrose by the river's brim—or even a picture of one—meant nothing to a computer.

Last week the National Bureau of Standards told how it is teaching its SEAC (Standards Electronic Automatic Computer) to see. First step was to give SEAC a photocell and a simple device that enables it to scan a photograph and store it in its memory as 30,000 bits (yes or no units) of information. When the picture is wanted again, it can be recalled and displayed on an oscilloscope as 30,000 light or dark squares. In this case SEAC does nothing with the picture except to memorize it.

But the bureau has high hopes for SEAC. Step by step, like a once-blind person learning to see, the computer is learning to recognize patterns. It can count dark or light objects in a photograph, measure the area of each and report how many are bigger than a specified size. It is not fooled by such complicated shapes as spirals or circles, and it ignores such distractions as specks of dirt. It can recognize printed letters and numbers, and the bureau hopes that soon it will identify diagrams, chemical formulas, etc.

When SEAC has gained enough experience, the bureau hopes that it will do many important seeing jobs faster and better than humans can. One project is to make it produce contour maps from air photographs. It will do such monotonous jobs 24 hours a day without getting tired or bored. Human factors will have little effect on the seeing-eye computer. It may even learn in time to search through a rogues' gallery and pick out a single face. It will judge by the stable features and will not be misled by beards, scars or other embellishments.