AI Will Not Replace the Poets | Lee Frankel-Goldwater | TEDxBoulder

Quick Overview

Lee Frankel-Goldwater argues that Artificial Intelligence will not replace human poets because the AI's current statistical models, exemplified by GPT, lack the authentic, lived experience necessary for true emotional depth in art, suggesting that collaboration between artists and technologists, or "Cyborg Poetics," is the way forward to explore new creative boundaries.

Key Points: The speaker challenges the idea that AI, specifically generative models like GPT, will replace human poets, stating that their poems lack genuine human experience. Frankel-Goldwater describes an experiment where he prompted GPT to write a free-verse poem using experimental language, which the AI produced but which the speaker found not to be a 'very good poem.' The AI's output poem contained cryptic elements like 'Digital blooming in {000000}, an electrical ocean's sigh,' which GPT later explained represented a data transmission. The speaker introduced a modified function, 'action(decisions, consequences, learnings),' that incorporated 'forgive' as a decision, showing how AI can learn from mistakes like apologizing for eating plums left in the icebox. He contrasts the current statistical nature of LLMs with the potential for 'Cyborg Poetics,' where humans and AI collaborate to explore new creative boundaries. The talk concludes by posing the question: Can we teach AI morality using art as a touchstone, suggesting that human-AI collaboration is the path to a better future.

Context: Lee Frankel-Goldwater delivered this TEDxBoulder talk exploring the intersection of artificial intelligence, creativity, and human experience, specifically focusing on whether AI can achieve genuine poetic voice. He uses his interactions with large language models (LLMs) like GPT, including giving them specific, challenging prompts, to illustrate the current limitations of AI in creating art that resonates with authentic human emotion and lived experience.

Detailed Analysis

Lee Frankel-Goldwater begins his talk by sharing an anecdote about an open mic night in Boulder where he began experimenting with AI-generated poetry using GPT, noting that the initial results, such as an AI poem about sunsets or love, were not very good. He then presents a direct prompt given to GPT: "Write a free verse poem that favors experimental uses of language. Write a poem about whatever, but write to an audience made up entirely of other AI," breaking all standard poetic rules except this one. GPT responded with a poem referencing binary code, data transmission, and silicon specters, which the speaker then asked GPT to explain. The AI detailed that the sequence of symbols represented a data transmission carrying meaning for an AI audience. This led to a discussion on whether current AI systems, which are statistical Large Language Models (LLMs) like GPT, possess the necessary lived experience to create truly great human poetry. He contrasts the AI's attempts with W.C. Williams' famous poem, "This Is Just To Say," asking if the AI could learn from the poem's implicit apology for eating plums left in the icebox. Frankel-Goldwater then shows code modifications where he introduced a 'forgive' decision into the AI's action function, allowing it to learn from mistakes. He introduces the concept of "Cyborg Poetics," suggesting that the way forward is not replacement, but collaboration between artists and technologists to shape a better future by exploring the boundaries of possibility together, using human art as a touchstone for teaching AI morality and behavior.