More on AI and Poetry (2)

In a past Poetry Note “Carmen Starnino” (April 25, 2022), the poet raised the question of how far Artificial Intelligence (AI) could go. Would it one day, even soon, surpass poets’ talents and make human creativity obsolete?

In “More on AI and Poetry” (May 2, 2022), technologist David Brayley rebutted: “The engineers who built a machine that cranks out verse lack a basic understanding of how poetry happens. They mistook the artifact for the art. It may be possible for a poem-shaped blob of words to emerge from an algorithm, but that algorithm cannot have the experience of writing a poem, and will not feel the thrill a poet feels when words condense spontaneously from the vapour of language itself.”

Further reassurance appeared this past week in a Toronto Star article “AI hallucinates” (August 2, 2023), an issue of growing concern in academe, business, and elsewhere. In other words, at some point as AI writes, it begins “just plain making things up.” The reason, explains Daniela Amodei, co-founder and president of Anthropic, maker of the chatbot Claude 2, is: “They’re just sort of designed to predict the next word.” To me, that sounds not only short-sighted and mechanically left-brain, but a process even more nose-to-the-ground linear than for creating expository prose. In contrast, the stunning metaphors of poetry burst forth from right-brain, lateral thinking, discovering a meaningful, not just verbally flashy, new unity in pulling together what had appeared disparate.

In the same article, Emily Bender, a linguistics professor and director of the University of Washington’s Computational Linguistics Laboratory, insists the problem is not fixable. “It’s inherent in the mismatch between the technology and the proposed use cases”—echoing David Brayley’s argument above.

Another point is AI’s backward-looking focus, which Bender describes as “‘modelling the likelihood of different strings of word forms’ given some written data it’s been trained upon.” Applied to a poet, such retro training sounds like a risky, potential path to writing clichés!

I’ll take the “vapour of language” any day!