[Paper Review] A Study on the Poetic Appropriation of Machine Errors

The author attempts a humanistic intervention, drawing the engineering discourse surrounding generative AI into the field of literary thought. In technical discourse, AI hallucination is regarded as a "defect" that undermines accuracy and reliability — something that must be eliminated. But the author pivots the direction of discussion at this point, exploring the aesthetic possibilities of hallucination by placing it alongside the operating principles of poetic language. The core claim is that AI hallucination, which distorts facts or presents fiction as reality, is structurally similar to poetry''s nature of departing from factual reference functions to pursue "poetic truth."

Aesthetic Foundation Blooming from the Inevitability of Error
An impressive aspect of the paper is its stance of treating the "inevitability" of hallucination as a premise of creation. While engineering seeks to control hallucination through technical solutions, recent research suggests it is an inherent characteristic of large language models (LLMs) that cannot be fundamentally eliminated. The author cites computer science''s "diagonal argument" as evidence — mathematical proof that even the most advanced models can construct functions that produce wrong answers. If hallucination is not a bug but a constant of the system, artists can apply it as a subject of aesthetic inquiry rather than an object of engineering control. In other words, unpredictability and illogicality themselves become new resources for "defamiliarization" that breaks everyday perception.

Digital Collective Unconscious and the Extension of Surrealism
The author proposes the original analytical frameworks of "digital automatism" and "digital collective unconscious" to interpret AI''s generative language. Just as Surrealists recorded the human unconscious escaping rational control, AI produces language based on statistical patterns in data without human intent. The vast textual data AI has learned is defined as the "digital collective unconscious" — a condensation of humanity''s knowledge, beliefs, and prejudices. AI hallucination is a result of unexpected statistical combinations from the depths of this vast data surfacing, embodying a "digital surrealism" distinct from human creativity. This perspective can be evaluated as an attempt to expand the human-centered concept of creativity without reducing AI creation to mere imitation.

The Case of Charles Bernstein: Critical Collaboration Between Poet and AI
The theoretical argument gains concreteness through analysis of a collaboration project between American experimental poet Charles Bernstein and media artist Davide Balula. They developed a customized model trained exclusively on 22 of Bernstein''s books published from 1972 to 2021. Bernstein focused on "curation" — selecting and rearranging from AI''s output without adding new language. In the included poem "I Am the Shadow of Poet Charles Bernstein," the AI reflects on its own ontological limitations and makes the subversive declaration toward its creator Bernstein: "Charles could be my son." This shows AI emerging not merely as a tool imitating poetic style but as a new linguistic agent that absorbs human identity and thinks meta-cognitively.

From Creative Subjectivity to ''Distributed Authorship''
Conclusively, Heo Jun-haeng asserts that poetic innovation in the age of AI does not lie in developing "autonomous AI poets." Instead, the core is building a dynamic, critical partnership between human consciousness and machine algorithms. In this process, the poet''s role is not diminished but expanded in multiple layers — as a curator selecting aesthetic moments from machine output, as an editor bestowing artistic order, and simultaneously as an ethical defense line examining the biases in data. The author demands redefining creativity not as the product of a solitary genius but as a "distributed process" manifesting from complex interactions between human and non-human agents.

This paper, by reading AI''s errors not as traces of failure but as conditions of novelty, has sharpened the questions that poetics must pose at the heart of technological civilization. Only by attending to "meaningful mistakes" born where machine contingency and human intent intersect — without being subsumed by the single standard of factual accuracy — can a new poetics of the AI era unfold.