From Oracle to Mapmaker: Protecting Human Agency via Structured Anti-Bias Prompting

Fuente: Zenodo
Salvato in:
Dettagli Bibliografici
Autore principale: Ishibashi, Ryuhei
Natura: Recurso digital
Pubblicazione: Zenodo 2025
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866902007728570368
author Ishibashi, Ryuhei
author_facet Ishibashi, Ryuhei
contents <p>As Large Language Models (LLMs) become ubiq-<br>uitous in decision-making, a critical sociotechnical<br>challenge has emerged: Automation Bias, where<br>users surrender agency to authoritative-sounding “Or-<br>acles.” We attribute this interaction failure to the<br>Certainty Bias Trap, a structural mechanism where<br>autoregressive models prematurely converge on a sin-<br>gle narrative.<br>To reshape this interaction, we introduce Struc-<br>tured Anti-Bias Prompting (SAP). Originating<br>from engineering root cause analysis in Sep 2025, SAP<br>transforms the LLM from an “Answer Engine” into a<br>“Mapmaker.” By mathematically enforcing diver-<br>sity via Multi-Perspective Validation, SAP acts<br>as a Cognitive Forcing Function. It prevents the<br>model from making the decision for the user, instead<br>providing the cognitive scaffolding necessary for the<br>user to act as the “Navigator.” This paper proposes<br>SAP not merely as a technical fix, but as a philosoph-<br>ical framework to restore Epistemic Agency in the<br>age of AI.</p>
format Recurso digital
id zenodo_https___doi_org_10_5281_zenodo_17909961
institution Zenodo
language
publishDate 2025
publisher Zenodo
record_format zenodo
spellingShingle From Oracle to Mapmaker: Protecting Human Agency via Structured Anti-Bias Prompting
Ishibashi, Ryuhei
<p>As Large Language Models (LLMs) become ubiq-<br>uitous in decision-making, a critical sociotechnical<br>challenge has emerged: Automation Bias, where<br>users surrender agency to authoritative-sounding “Or-<br>acles.” We attribute this interaction failure to the<br>Certainty Bias Trap, a structural mechanism where<br>autoregressive models prematurely converge on a sin-<br>gle narrative.<br>To reshape this interaction, we introduce Struc-<br>tured Anti-Bias Prompting (SAP). Originating<br>from engineering root cause analysis in Sep 2025, SAP<br>transforms the LLM from an “Answer Engine” into a<br>“Mapmaker.” By mathematically enforcing diver-<br>sity via Multi-Perspective Validation, SAP acts<br>as a Cognitive Forcing Function. It prevents the<br>model from making the decision for the user, instead<br>providing the cognitive scaffolding necessary for the<br>user to act as the “Navigator.” This paper proposes<br>SAP not merely as a technical fix, but as a philosoph-<br>ical framework to restore Epistemic Agency in the<br>age of AI.</p>
title From Oracle to Mapmaker: Protecting Human Agency via Structured Anti-Bias Prompting
url https://doi.org/10.5281/zenodo.17909961