While human brains and computers are often considered to be general purpose technologies, like any real technology, brains and computers have physical structures and operating designs that generate constraints and imply trade-offs across tasks. A computer program cannot easily read an image containing distorted text, but a human can. Thus requiring blog commenters to pass a CAPTCHA lessens automated blog comment spam.
Distributing computing between humans and computers is an important aspect of efficient problem-solving. Amazon’s Mechanical Turk provides a general marketplace for distributing computation to persons. Luis von Ahn is designing games for humans that produce both fun play and symbolic work that computers perform relatively badly. Some examples are the ESP Game, which generates image labels, and Peekaboom, which generates descriptions of objects within an image.
Sensory form is an aspect of distributing computing between humans and computers. Consider, for example, women-oriented social drama programming. Prior to the widespread availability of television, such radio programming was highly popular. This type of program shifted almost completely to television when television became widely available. With the exception of sensory modes, the radio and television programs were formally quite similar stimuli. Why did persons prefer the audio-visual mode (television) over the audio-only mode (radio)? A plausible explanation, it seems to me, is that the bodily cost of making sense of conventional drama via the audio-visual experience of television is less than that of making sense of the same drama presented solely through the audio channel of radio. Put differently, image computation was shifted from persons to television studios.
The Internet, by connecting huge numbers of persons and computers, enables a tremendous possibilities for distributing computing among persons and computers. Tim O’Reilly declares:
As the symbiosis between humans and computers becomes deeper, and at a larger scale, we’re going to see problems that were formerly construed as “hard AI” suddenly broken, not because computers themselves have become intelligent, but because humans and computers have gotten better at working together.
Getting humans and computers to work together better requires more thinking about what computers do well, and what humans want to do, and do well.