Abstract

Philosophers such as Searle have claimed that full articial intelligence is impossible because the symbols used in AI programs are meaningless. This paper proposes that a key to attributing meaning to symbols in programs is satisfaction of a principle of inductive adequacy, which says that an AI program should contain learning mechanisms sufficient to generate all kinds of knowledge structures that it uses in performances. How this principle would work is illustrated using LISP program PI, and an account is developed of how meaning can emerge in a computational system that satisfies the principle