Poetry Will Not Optimize; or, What Is Literature to AI? by Michele Elam

Wednesday 21st February, 2024 - Bruce Sterling



Literature, poetry, and other forms of noncommercial creative expression challenge the techno-instrumentalist approaches to language, the predictive language generation, informing NLP (large natural language processing models) such as GPT-3 or -4 as well as, more generally, generative AI (text to image, video, audio). Claims that AI systems automate and expedite creativity reflect industry and research priorities of speed, scale, optimization, and frictionlessness driving much artificial intelligence design and application.

But poetry will not optimize; the creative process cannot be reduced to a prompt. Some have noted that literary creations generated or augmented by artificial intelligence at best can offer form without meaning; using a GPT creation prompted by Maya Angelou’s poem “Still I Rise” as a case study, this essay argues that NLP’s predictive language generation and what I call algorithmic ahistoricity can also, more disturbingly, render meaning senseless. In doing so, GPT-3’s literary experiments are not “failed” because they do not meet some moving target of a literary standard, nor because of technological insufficiency, but because it can make it harder for people to name and navigate their realities. The coda explores an example of AI as literary interlocutor and creative engagement beyond optimization….

Issue Section: Essays
artificial intelligence, literature, art, race, GPT-3


Here I suggest essential challenges posed for AI by the arts (writ large to include literary, visual, performative, theatrical, graphic, musical) and, in turn, how AI might productively challenge the arts. Just as AI has invited debates about what constitutes or performs intelligences far beyond the Turing test, AI revives foundational questions in the arts and humanities about what is or is not literature or art; who or what can make it; how is it credentialed; how compensated; who arbitrates taste, value, valuation, proprietary content, and provenance (especially in the case of AI-generated art); who gets to decide the arbitrators; and who (or what) counts as a maker. These are not abstract questions, and the stakes are high….