CORTEXA
← Browse
openalexZenodo (CERN European Organization for Nuclear Research)2026-07-24Cited by 0

The Genetic Fallacy at Scale

Daniel Toupin

The rapid normalization of generative artificial intelligence has produced two distinct epistemic problems that are routinely conflated. The first is a problem of quality: language models can generate false, derivative, vacuous, or stylistically polished but epistemically poor material at enormous scale. The second is a problem of judgment: the causal provenance of a text is increasingly treated as a substitute for evaluating its propositional content. This paper argues that the latter practice, when the label “AI-generated” or “AI-assisted” is used to reject a claim, argument, or intellectual artifact without content-responsive evaluation, instantiates the genetic fallacy at institutional scale. I first distinguish truth, warranted credence, authorship, and policy compliance, showing that provenance may rationally affect prior confidence or trigger additional scrutiny without thereby refuting a proposition. I then analyze AI-text detection and automated content moderation as forms of probabilistic classification that become epistemically illegitimate when converted directly into adverse adjudication. Empirical work on machine-generated-text detectors demonstrates substantial vulnerability to false positives, distribution shift, linguistic bias, and adversarial evasion, while the literature on automation bias and algorithmic accountability shows why nominal “human oversight” is inadequate when humans merely ratify machine outputs. I argue for a principle of human answerability: tools may be causal contributors to error, but responsibility for consequential epistemic decisions remains with the persons and institutions that design, deploy, supervise, and rely upon them. Finally, I propose six norms for legitimate epistemic governance in the age of generative AI: provenance modesty, content-responsive refutation, proportionality, contestability, auditability, and human answerability. The central danger is not that machines can produce bad arguments. Bad arguments can be answered. The deeper danger is that automated systems can acquire authority over which arguments are permitted to receive an answer at all.

View free PDFSource page

Related papers

openalexZenodo (CERN European Organization for Nuclear Research)2026-07-24

Dataset: SOD1 Research July 2026 - PathMap Experiment #000084

Joshua Dungan

Interactive Data Viewer: Read, View, and Print from Day 1 Use our fully interactive viewer to view, read, and print this research data right from Day 1: https://pathmap.org/viewer.php?id=84 Artificial General Intelligence LLC Claim Evaluated: SOD1 Research July 2026 This dataset…

View free PDFSource page
openalexZenodo (CERN European Organization for Nuclear Research)2026-07-23

SPADE OC2 Mediterranean Trefoil crop dataset (T1) - VITAFields – Visual Intelligence for Targeted Agriculture #4

George Kougianos, Athanasios Kerasiotis, Maria Petsa, Emmanouil Taoulai

This dataset release represents Part 4 of the comprehensive young trefoil crop agricultural analysis project. While Part 1, Part 2 and Part 3 provided the raw image captures and camera parameters and machine-learning-ready image tiles. Part 4 delivers fully processed, georeferenc…

View free PDFSource page
openalexZenodo (CERN European Organization for Nuclear Research)2026-07-25

QSM-CI method: BFRnet (v1)

Xuanyu Zhu, Yang Gao, Feng Liu, Stuart Crozier, Hongfu Sun

Deep-learning background field removal (BFRnet): a 3D dual-frequency octave-convolution U-net trained to predict the background field of the brain — including brains with significant pathological susceptibility sources (haemorrhage, calcification). Consumes the total field (ppm)…

View free PDFSource page
openalexZenodo (CERN European Organization for Nuclear Research)2026-07-25

QSM-CI method: INR-QSM (v1)

M. Zhang, Feng R., Li Z., Feng J., Wu Q, Zhang Z., et al.

INR-QSM — a subject-specific UNSUPERVISED deep-learning dipole inversion using an implicit neural representation. No pretrained weights: a sine-activated coordinate MLP (SIREN) is OPTIMIZED per-subject so that the susceptibility it represents, pushed through the QSM dipole forwar…

View free PDFSource page
openalexZenodo (CERN European Organization for Nuclear Research)2026-07-23

Data from: Automated Ethogram Elaboration: A Cross-species Deep-learning Model Deployed On-board Enables Acceleration-triggered Capture of Diverse Behavioural Video

Tsuneari Kuroiwa

This record contains field-deployment datasets for two seabird species, streaked shearwaters and black-tailed gulls, used in the paper “Automated Ethogram Elaboration: A Cross-species Deep-learning Model Deployed On-board Enables Acceleration-triggered Capture of Diverse Behaviou…

View free PDFSource page
openalexZenodo (CERN European Organization for Nuclear Research)2026-07-25

OpenAI — An Independent Governance and Behavioural Assessment.

Ankit Musaddi

This paper presents an independent external assessment of OpenAI — the OpenAI Group PBC, the controlling OpenAI Foundation, and its publicly released GPT-series and o-series models — benchmarked against four governance frameworks: the EU AI Act (including its general-purpose AI a…

View free PDFSource page