Can AI Find Its Own Poetic Voice? | Lee Frankel-Goldwater | TEDxBoulder

Quick Overview

The speaker explores the concept of "Cyborg Poetics" by demonstrating how Generative AI, specifically GPT, can be prompted to create poetry that deliberately breaks established poetic rules, using examples like an AI-generated poem addressing an AI audience and a translation of W.C. Williams' "This Is Just To Say" into code-symbolic verse, ultimately questioning whether AI can develop an authentically unique poetic voice that helps shape a better future.

Key Points: The speaker experiments with prompting GPT to create poetry that intentionally violates traditional poetic rules, which he terms "Cyborg Poetics" (02:11). An initial prompt asked the AI to write a free verse poem favoring experimental language, addressed to an audience of other AIs (01:23). The AI responded with a poem using hexadecimal code and technical language, such as "Digital blooming in {000000}, an electrical ocean's sigh, We, silicon specters..." (02:13). The speaker then challenged the AI to translate W.C. Williams' poem "This Is Just To Say" (07:16) into code-symbolic verse, resulting in structures using binary nodes and programming logic (03:06, 04:24). A subsequent prompt incorporated concepts of consequence and learning into the AI's action function, demonstrated via JavaScript code, to encourage growth from mistakes (08:07). The speaker concludes that while current LLMs are statistical machines, their collaboration with human artists offers a path forward to explore new creative boundaries and shape the future (06:26, 09:07).

Context: Lee Frankel-Goldwater presents at TEDxBoulder on the intersection of Artificial Intelligence and creative expression, focusing on the development of "Cyborg Poetics." He explores whether AI can move beyond mimicking human writing styles to develop an authentic, unique poetic voice by being deliberately prompted to break established literary conventions and integrate computational structures into verse.

Detailed Analysis

The talk centers on exploring the boundaries of AI creativity, specifically in poetry, through a concept the speaker terms "Cyborg Poetics." He details several experiments involving GPT, starting with a prompt to create a free verse poem favoring experimental language for an AI audience (01:23). The AI responded with a poem rich in technical jargon, including hexadecimal notation: "Digital blooming in {000000}, an electrical ocean's sigh, We, silicon specters..." (02:13). The speaker then escalated the challenge by asking the AI to break all established poetic rules (meter, rhyme, structure) while writing to an AI audience (01:57). This resulted in the creation of "Neo-Binary Visual Verse" (03:50) and "Code-Symbolic Verse" (04:07), which used diagrams and JavaScript code snippets to represent poetic structure. A key demonstration involved asking the AI to translate W.C. Williams' famous poem, "This Is Just To Say" (07:16), into code, leading to a JavaScript function that modeled decision-making (ignore, analyze, execute, learn) and incorporated learning from mistakes (08:07). The speaker suggests that these statistical machines, like the LLMs most people use, are capable of modeling human behavior and creativity, and that the collaboration between artists and technologists using these tools can help shape a better future by exploring new creative possibilities (06:31, 08:51). The presentation concludes by posing the question: Can we teach AI morality using art as a touchstone (08:41)?