Limits of human-machine interaction

Abstract

A lot of work has been done to improve the person-machine interaction. A machine must have a language provided with a powerful syntax enabling it to simulate, up to a certain point, a verbal interaction between two persons. The syntax must contain, for instance, illocutionary force indicating devices which determine when an utterance should be interpreted as an assertion, an order, a promise, etc. The machine, to be minimally useful, must also have “at its disposal” a huge data bank about frequent contexts of language use, and about regularities, natural and social, practices and standard way of doing things. I want to emphasize the following difference between regimented languages and natural languages: the language processed by the machine must be highly regimented, while ordinary language has an irreducible “open texture”, and exhibits a strong context-dependency. All the tokens of the same sentence of a regimented language must be understood in the same way by all the members of the same scientific community, in a set of contexts of use well delimitated. This is fundamental for the efficiency of scientific communication. The algorithms processed by a machine works, so to speak, in closed environment unaffected by contextual factors. However, in ordinary language, the tokens of the same sentence must be understood differently, according to the context, to be understood correctly. These features of natural languages are hardly captured by the workings of a machine. This difference, I believe, is anthropologically meaningful.