Skip to content

Latest commit

 

History

24 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 

Repository files navigation

DOI PsyArXiv Preprint License: CC BY 4.0

The Confabulation Engine

Why Artificial Intelligence Does Not Lie — It Confabulates

Author: Sergio Cayuqueo (Independent Researcher)
Publication Date: August 2026


Overview

When large language models generate false assertions with pristine confidence, the prevailing public and technical moniker is "hallucination." Borrowed loosely from clinical psychiatry, the term is fundamentally misapplied. Hallucination presupposes a perceiving subject whose sensory channels misfire; artificial neural networks possess no sensory organs, no perceptual continuum, and no subjective awareness to experience a phantom percept.

This repository hosts the manuscript "The Confabulation Engine," which proposes a far more precise and historically grounded conceptual framework: confabulation—the fluent, unhesitating generation of a false narrative by a cognitive system structurally incapable of recognizing its own ignorance.


Key Theoretical Contributions

  • Clinical & Neurological Foundation: Traces confabulation from early psychiatric work (Bonhoeffer, Korsakoff) to Gazzaniga & LeDoux’s split-brain "interpreter theory," showing how systems tasked with narrative synthesis fill knowledge gaps with contextually continuous detail.
  • Rebuttal of Alternative Frameworks: Refutes Emily Bender et al.’s "stochastic parrot" thesis and Hicks et al.’s application of Frankfurtian "bullshit," demonstrating that confabulation is not an intentional choice to ignore truth or a rare glitch, but the default state of an unconstrained narrative generator.
  • Empirical Evidence: Examines real-world failures like Mata v. Avianca alongside empirical studies (Sui et al., 2024) revealing that false model outputs are measurably more narratively coherent and syntactically persuasive than true ones.
  • Institutional & Technical Safeguards: Recommends replacing "hallucination" in regulatory standards with "confabulation," requiring calibrated uncertainty metrics, external verification loops (RAG, logic checkers), and mandated human verification in high-stakes environments.

File Structure

.
├── README.md         # Repository documentation
├── paper.md          # Complete text of the manuscript
├── CITATION.cff      # Citation metadata
└── LICENSE           # CC-BY-4.0 Open Access License

About

Theoretical paper on why large language models produce fluent, ungrounded false narratives by design—arguing for 'confabulation' over 'hallucination' using Gazzaniga's interpreter theory.

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors