Aristotle, Wittgenstein, and Artificial Intelligence.

One of the most challenging projects in Cognitive Science is building pattern recognition devices: Machines that can make judgments about what sort of category a perceived object belongs in. The optical scanners that can read different kinds of type faces are an example of such a device and so are certain security systems that learn how to recognize people’s voices, as well as the voice recognition system that enables you to dictate text into your iPhone. When Cognitive Scientists first attempted to build pattern recognition machines, they tried at first to use something like Aristotle’s category system. They formulated necessary and sufficient definitions for each category the machine was supposed to recognize, and stored those definitions in some kind of logic-based computer code. In some cases, the cognitive scientists who did this were consciously trying to follow Aristotelian principles, but in many cases they were following them because it never occurred to them that any other way of thinking was possible.

When these methods turned out to be unsuccessful, however, Cognitive Scientists took a different approach. Instead of trying to define and identify the properties possessed by every member of the category (i.e. making judgments of the form all S is P), these systems would define category S by a cluster of properties, each of which may or may not be present. They would then make judgments using special statistical processors that make decisions using something like Wittgenstein’s principles of family…