Dissociated Responses to AI: Legal AI Advisors Are Persuasive But Not Trustworthy?

Aydin, Zeynep

Full Metadata

Overview

Year:: 2024
Contributor:: Aydin, Zeynep (creator); Malle, Bertram (Advisor); Netter, Julia (Reader); Sloman, Steven (Reader); Brown University. Department of Cognitive, Linguistic, and Psychological Sciences (sponsor)
Genre:: theses
Subject:: human-computer interaction; Artificial intelligence; Reasoning; Social cognition
Extent:: iv, 45 p.

Files

Description

Abstract:: Abstract of DISSOCIATED RESPONSES TO AI: LEGAL AI ADVISORS ARE PERSUASIVE BUT NOT TRUSTWORTHY?, by ZEYNEP AYDIN, Degree ScM., Brown University, MAY 2024. The rapid development of Artificial Intelligence (AI) advisors in the domains of health, finance, and military poses pressing research challenges. For the successful implementation of such artificial advisors, individuals must be able to critically evaluate AI advice. Empirical work on people’s perceptions of AI advisors has found evidence for both over- and under-reliance on their advice, known respectively as "algorithm appreciation" and "algorithm aversion." My thesis explores how people react to artificial advisors regarding their willingness to change their opinions, approve of and trust the advisor. I study these questions within the legal domain, a promising yet underexplored application for AI advisors. I apply a dual-process framework (distinguishing between central and peripheral processing routes) from the persuasion literature to interpret my findings of dissociated responses to AI advisors. In three survey studies involving legal dilemmas, I analyze how people respond to advice from human versus AI advisors. In Study 1, participants made judgments about legal dilemmas and updated their judgments based on advice they received either from a human or an AI advisor. People also expressed their approval of the (human or AI) advisor. Study 2, using a similar design, pitted two types of advisors (human and AI), offering opposing arguments against each other. Here, participants also updated their judgments in response to the advice, expressed their approval, and provided trust ratings. Both studies showed that people were equally persuaded by both human and AI advisors. However, their trust and approval ratings were consistently lower for the artificial advisor, revealing a potential dissociation between the deliberate consideration of the contents (central route) and affective responses to the source (peripheral route). This suggests that while there is no clear pattern of algorithm aversion or appreciation in scenarios involving central-route information processing, people have affective reservations towards AI advisors. In both studies, arguments were of the same type and quality. To better understand how people perceive different types and strengths of arguments from different sources, Study 3 (currently ongoing) will examine the effects of argument quality and type on people’s possibly dissociated persuasion and trust responses to AI advisors.
Notes:: Thesis (Sc. M.)--Brown University, 2024

Content

Citation

Aydin, Zeynep, "Dissociated Responses to AI: Legal AI Advisors Are Persuasive But Not Trustworthy?" (2024). Cognitive, Linguistic, and Psychological Sciences Theses and Dissertations. Brown Digital Repository. Brown University Library. https://repository.library.brown.edu/studio/item/bdr:a8ev4cst/

Relations

Collection:

Cognitive, Linguistic, and Psychological Sciences Theses and Dissertations

Theses and Dissertations for the Cognitive, Linguistic, and Psychological Sciences department.
...