Language that Machines Read by John Cayley

Machines are said to ‘read’ text in a variety of ways which now include its being taken in and analyzed by so-called AI, by Machine or Deep Learning algorithms which generate what are usually referred to as ‘language models.’ These models become capable, it is proposed, of ‘automatically’ generating language if not, yet, ‘understanding’ it. But is it actually language, in any significant or affective sense, of which these analyses are models? They are certainly models of once articulated text, text that may well be some kind of record or representation of linguistic behavior, but to understand the relationship between these engines of contemporary computation and human language, we must enquire a little more deeply into the question of what language is.

This essay makes an attempt. In its present form, it takes you through my thinking at a steady pace, but only directly addresses so-called AI as it approaches its conclusion. In this rendition, I've provided an i[n]d[e]x for less linear navigation of its 72 textual spaces, but I hope that most readers will – at least for their first reading – follow the essay’s online flow.