Episode 8.14: ChatGPT has faults; the remarkable thing is that they are exactly like ours.

Numerous commentators have been queueing up to pour tepid water over the achievements of the OpenAI conversational chat bot chatGPT. What is of particular interest about their dismissive or questioning comments is that they accuse the chatbot of having as significant faults precisely the kind of things of which we would accuse one another. For example, arguing that the chat bot “fabricates facts and figures” because of the way it operates would seem to ignore the fact that human beings do just the same. That we do not understand how the chatbot draws its inferences, and perhaps accuse it unfairly of using statistical patterns, rather than “connecting words with meanings” seems to suggest that we understand how humans understand language when in fact we don’t. You and I have a huge number of instances of the occurrence of every word in our native tongue stored in our heads, and somehow we make sense of them, but nobody knows how. Somehow we connect the word table with what the word table means. This is a problem that goes back to Plato. Who is to say how we do that? Nobody has any idea, and that we perform some kind of statistical matching process may well be as good an explanatory candidate as any. So it’s unfair to accuse chat GPT of failings that we have ourselves. That just means that it is closer to joining the human conversation as a fully-fledged participant than we might otherwise have thought.