Monitoring Anthropomorphism of AI on Misinformation and Mainstream News Websites

Presenter

Algis O. Petlin

Group Members

Kaitlyn Alexa Malsky, Kate Sorz

Campus

UMass Amherst

Sponsor

Ankita Gupta, Department of Computer Science, UMass Amherst

Schedule

Session 2, 11:30 AM - 12:15 PM [Schedule by Time][Poster Grid for Time/Location]

Location

Poster Board A31, Campus Center Auditorium, Row 2 (A21-A40) [Poster Location Map]

Abstract

Generative AI's rising popularity has raised concerns like human-level intelligence and potential to displace many from jobs, topics often weaponized to spread misinformation. In this work, we investigate different news websites for their tendency to anthropomorphize AI and whether unreliable websites anthropomorphize more than reliable ones, potentially aiding in misinformation analysis. To conduct this study, we collect 2700 articles discussing AI (X reliable, Y unreliable), between January 2022-February 2024. We identify a website’s reliability using Media Bias Fact Check (MBFC)’s ratings. To measure degree of anthropomorphism, we use a linguistic metric [Cheng et al., 2024] which computes the log-ratio of the probability that the entity could be replaced by human pronouns versus non-human pronouns. A positive A-score implies tendency to anthropomorphize (i.e., the entity is more likely to be framed as human). We find that unreliable websites (0.27 ± 0.2) anthropomorphize more than reliable websites (-0.52 ± 0.4), scores aggregated over a website’s articles. Among reliable websites, futurity.org and amgreatness.com have the highest A-scores, which are mostly factual though have few failed fact-checks as per MBFC. Among unreliable websites, SHTFPlan.com, breakingwide.com, and off-guardian.org have the highest A-scores, also reported for using emotional/loaded language by MBFC, establishing convergent validity. Analyzing A-scores over time, we observe a Spearman’s coefficient of 0.02 (reliable) and 0.03 (unreliable), suggesting a near-zero correlation. Overall, our findings suggest that anthropomorphism has the potential to serve as an indicator for sensationalism, aiding future misinformation research effort.

Keywords

Anthropomorphism, Natural Language Processing (NLP), Misinformation

Research Area

Computer Science

SIMILAR ABSTRACTS (BY KEYWORD)