[Paper Review] From Textual Experiments to Experimental Texts: Expressive Repetition in "Artificial Intelligence Literature"

Zhu Tianhua|arXiv (Cornell University)|Jan 7, 2022

Digital Humanities and Scholarship4 citations

TL;DR

This paper examines the evolution of AI literature from early technical 'textual experiments' to contemporary 'experimental texts' that leverage expressive repetition. It argues that advances in natural language generation enable AI to produce literature with self-verified artistic value, marking a shift from proving technical feasibility to exploring literary expression through recursive, stylistically meaningful repetition.

ABSTRACT

Since the birth of artificial intelligence 70 years ago, attempts at literary "creation" with computers are present in the course of technological development, creating what one might call "artificial intelligence literature" (AI literature). Evolving from "textual experiments" conducted by technologists to "experimental texts" that explore the possibilities of conceptions of literature, AI literature integrates primitive problems including machine thinking, text generation, and machine creativity, which exhibits the two-way interaction between social ideas and technology. In the early stage, the mutual support between technological path and artistic ideas turned out to be a failure, while AI-driven expressive repetitions are made probable in the contemporary technological context, paving the way for the transformation of AI literature from proof for technical possibilities to self-verification of literary value.

Motivation & Objective

• To trace the historical transformation of AI-generated literature from technical demonstrations to artistic experimentation.
• To analyze how expressive repetition functions as a literary device in contemporary AI literature.
• To investigate the conditions under which AI-generated texts achieve self-verification of literary value.
• To examine the interplay between technological development and evolving conceptions of literature in AI-driven writing.
• To position AI literature within broader sociotechnical discourses on machine creativity and artificial intelligence.

Proposed method

• Analyzes historical and contemporary examples of AI-generated literary texts from the past 70 years.
• Identifies key transitions from 'textual experiments'—focused on technical proof-of-concept—to 'experimental texts' emphasizing literary form and expression.
• Examines expressive repetition as a central stylistic and structural device in modern AI literature.
• Uses discourse analysis to interpret the socio-technical dynamics shaping AI literature's development.
• Compares early AI text generation (e.g., ELIZA, Racter) with recent large-scale language models in terms of narrative and rhetorical complexity.
• Evaluates the role of machine learning in enabling recursive, self-referential, and stylistically coherent textual repetition.

Experimental results

Research questions

• RQ1 How has the purpose of AI-generated literature evolved from technical demonstration to artistic experimentation?
• RQ2 In what ways does expressive repetition function as a literary device in contemporary AI literature?
• RQ3 What technological conditions enable AI to produce texts with self-verified literary value?
• RQ4 How do social conceptions of literature interact with technological capabilities in shaping AI literature?
• RQ5 What distinguishes 'textual experiments' from 'experimental texts' in the context of AI-generated writing?

Key findings

• AI literature has transitioned from early 'textual experiments' focused on proving machine text generation to 'experimental texts' that explore literary form and expression.
• Expressive repetition—repetition with stylistic, emotional, or structural variation—has become a central mechanism for AI-generated literary expression.
• Contemporary large language models enable the generation of texts with recursive, self-referential, and thematically coherent repetition, supporting literary self-verification.
• The shift from technical proof-of-concept to artistic exploration is enabled by advances in neural text generation and contextual modeling.
• The integration of social ideas about literature and machine creativity has become increasingly symbiotic with technological progress.
• The paper identifies a critical threshold where AI-generated texts are no longer merely technical outputs but are perceived as having autonomous literary value through expressive repetition.