Main Session 3

In my The Instruction of Imagination (OUP, 2015), I suggest a new general characterization of language as a socially constructed tool, designed by cultural evolution to allow speakers to bridge the experiential gaps between them and their listeners – by instructing their listeners’ imagination. This view of the function of language allows for a deep understanding of the complex dynamics involved in linguistic communication. Speakers go through two processes of mental translation before uttering a sentence. First, they translate their private, experiential meanings into social, linguistic meanings. Then, they translate linguistic meaning into linguistic form, which eventually gets uttered. Listeners receive linguistic form, translate it into linguistic meaning, and then use this meaning as a code that instructs them in the process of the construction of an imagined experience, inside their heads, which – if everything works right – should give them a good sense of what the speakers had in mind. In regular conversation, where the experiential gap is not too wide, and the topic of conversation not too serious, all this happens without effort. This is testimony to the fact that we have adapted ourselves, through our evolution, for the requirements of the instruction of imagination. When the gap is very wide, and the topic very serious, the process becomes a difficult challenge. Bridging the gap with a person who is experiencing an emotional crisis, and dialoguing with that person in a useful way, may be one of the most extreme such challenges. For the person in trouble, the words often no longer represent the experiences; the gap between the person and the other is very wide; there is much suspicion and much more. The effort to meet this challenge can only be based on the capacity to hear the person beyond the words – to see the person behind the gap – and this cannot be done without empathy, sensitivity and experience. All this clarifies the difference between talking to a human and talking to an AI agent. AI agents are not experiential entities. They do not go beyond the words. In fact, they do not go beyond the forms of the words: they perform statistical analyses on the relations between forms in their pre-trained memories, and produce their sentences based on next-word predictions. The sentences are meaningful because the forms in the database were originally uttered with meaning, but the AI sentences as such only mimic meaning: there is nothing behind them. There cannot be, not without experience. This is good enough (and often excellent) for many practical purposes, but emergency calls are exactly the type of conversation from which AI agents should be barred.