Biased AI improves human decision-making but reduces trust

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Lai, Shiyang, Kim, Junsol, Kunievsky, Nadav, Potter, Yujin, Evans, James
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866908495177056256
author Lai, Shiyang
Kim, Junsol
Kunievsky, Nadav
Potter, Yujin
Evans, James
author_facet Lai, Shiyang
Kim, Junsol
Kunievsky, Nadav
Potter, Yujin
Evans, James
contents Current AI systems minimize risk by enforcing ideological neutrality, yet this may introduce automation bias by suppressing cognitive engagement in human decision-making. We conducted randomized trials with 2,500 participants to test whether culturally biased AI enhances human decision-making. Participants interacted with politically diverse GPT-4o variants on information evaluation tasks. Partisan AI assistants enhanced human performance, increased engagement, and reduced evaluative bias compared to non-biased counterparts, with amplified benefits when participants encountered opposing views. These gains carried a trust penalty: participants underappreciated biased AI and overcredited neutral systems. Exposing participants to two AIs whose biases flanked human perspectives closed the perception-performance gap. These findings complicate conventional wisdom about AI neutrality, suggesting that strategic integration of diverse cultural biases may foster improved and resilient human decision-making.
format Preprint
id arxiv_https___arxiv_org_abs_2508_09297
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Biased AI improves human decision-making but reduces trust
Lai, Shiyang
Kim, Junsol
Kunievsky, Nadav
Potter, Yujin
Evans, James
Human-Computer Interaction
Artificial Intelligence
Computers and Society
Current AI systems minimize risk by enforcing ideological neutrality, yet this may introduce automation bias by suppressing cognitive engagement in human decision-making. We conducted randomized trials with 2,500 participants to test whether culturally biased AI enhances human decision-making. Participants interacted with politically diverse GPT-4o variants on information evaluation tasks. Partisan AI assistants enhanced human performance, increased engagement, and reduced evaluative bias compared to non-biased counterparts, with amplified benefits when participants encountered opposing views. These gains carried a trust penalty: participants underappreciated biased AI and overcredited neutral systems. Exposing participants to two AIs whose biases flanked human perspectives closed the perception-performance gap. These findings complicate conventional wisdom about AI neutrality, suggesting that strategic integration of diverse cultural biases may foster improved and resilient human decision-making.
title Biased AI improves human decision-making but reduces trust
topic Human-Computer Interaction
Artificial Intelligence
Computers and Society
url https://arxiv.org/abs/2508.09297