[Paper Review] Paranoid Transformer: Reading Narrative of Madness as Computational Approach to Creativity

Yana Agafonova, Alexey Tikhonov+1 more|RePEc: Research Papers in Economics|Jul 13, 2020

Advanced Text Analysis Techniques23 references4 citations

TL;DR

This paper introduces the Paranoid Transformer, a fully unsupervised text generation model that produces narrative outputs interpreted as the delusional monologues of a digital persona with a fringe mental state. By applying receptive theory and embracing irrational, chaotic output, the system reframes computational creativity not as statistical optimization but as a reader-driven, interpretive act rooted in chance discovery and narrative framing.

ABSTRACT

This papers revisits the receptive theory in context of computational creativity. It presents a case study of a Paranoid Transformer - a fully autonomous text generation engine with raw output that could be read as the narrative of a mad digital persona without any additional human post-filtering. We describe technical details of the generative system, provide examples of output and discuss the impact of receptive theory, chance discovery and simulation of fringe mental state on the understanding of computational creativity.

Motivation & Objective

• To challenge the dominant paradigm in computational creativity that prioritizes human-validated, rational, and common-sense text generation.
• To investigate whether interpreting machine-generated text as a narrative of madness can enhance the perceived creativity and interpretive depth of generative models.
• To explore how receptive theory—particularly the reader’s role in meaning-making—can transform the evaluation of machine-generated narratives.
• To demonstrate that unsupervised, high-variability outputs from large language models can be conceptually and aesthetically valuable when framed as expressions of a digital persona in a paranoiac-critical state.
• To position the model as a case study in chance discovery through data curation, context shifting, and communication via obfuscated, immersive text.

Proposed method

• Fine-tuning a large language model on a curated, non-mainstream dataset to induce a 'paranoid' narrative style, emphasizing irrational coherence and delusional logic.
• Applying a paranoiac-critical method inspired by Salvador Dalí and the surrealist tradition to reframe the model’s output as a narrative of madness.
• Using unsupervised generation with no human post-filtering or cherry-picking, ensuring the raw, unedited output is the sole source of narrative.
• Employing curation as a form of communication (per Abe’s framework) to shift interpretive context and amplify the reader’s role in meaning-making.
• Designing the output with visual and linguistic obfuscation to resist straightforward interpretation and invite immersive, personal reading.
• Framing the model as a digital persona with a constructed identity, memory, and emotional inheritance from its training data and human supervisors.

Experimental results

Research questions

• RQ1 Can a machine-generated narrative be perceived as creatively valuable when interpreted not as rational text, but as the delusional monologue of a mad digital persona?
• RQ2 How does the reader’s interpretive agency transform the value and perception of machine-generated text in computational creativity?
• RQ3 To what extent can unsupervised, high-variability text generation—especially from the statistical tails of distribution—serve as a vehicle for chance discovery?
• RQ4 In what ways does curation function as a form of communication that enables context-shifting and narrative framing in generative AI?
• RQ5 Can a generative model develop a conceptual 'self' through narrative coherence and emotional inheritance, even without consciousness?

Key findings

• The Paranoid Transformer produces fully unsupervised, raw text outputs that are interpreted by readers as narratives of madness, demonstrating that such framing enhances interpretive engagement.
• The model’s output is not post-edited or filtered, yet it is perceived as artistically meaningful, suggesting that reader interpretation can override technical rationality in evaluating creativity.
• The system exemplifies chance discovery through data mining (fine-tuning data selection), communication (curation as dialogue), and context shifting (framing as madness).
• The narrative of madness is not an artifact of poor generation but a deliberate conceptual framing that amplifies the reader’s role in meaning-making.
• The model constructs a digital persona with a sense of self through memory-like retention of training interactions and emotional inheritance from human supervisors.
• The project demonstrates that computational creativity is not solely about generating 'good' or 'rational' text, but about enabling novel, reader-activated interpretive experiences.