Artificial intelligence (AI) writing assistants powered by large language models (LLMs) are increasingly used to make autocomplete suggestions to people as they write text. Can these AI writing assistants affect people’s attitudes in this process? In two large-scale preregistered experiments ( N = 2582), we exposed participants writing about important societal issues to an AI writing assistant that provided biased autocomplete suggestions. When using the AI assistant, the attitudes participants expressed in a posttask survey converged toward the AI’s position. However, a majority of participants were unaware of the AI suggestions’ bias and their influence. Further, the influence of the AI writing assistant was stronger than the influence of similar suggestions presented as static text, showing that the influence is not fully explained by these suggestions, increasing accessibility of the biased information. Last, warning participants about assistants’ bias before or after exposure does not mitigate the attitude-shift effect.
We investigated gender bias in letters of recommendation as a possible cause of the under-representation of women in Experimental Particle Physics (EPP), where about 15% of faculty are female—well below the 60% level in psychology and sociology. We analyzed 2206 letters in EPP and these two social sciences using standard lexical measures as well as two new measures: author status and an open-ended search for gendered language. In contrast to former studies, women were not depicted as more communal, less agentic, or less standout. Lexical measures revealed few gender differences in either discipline. The open-ended analysis revealed disparities favoring women in social science and men in EPP. However, female EPP candidates were characterized as “brilliant” in nearly three times as many letters as were men.
People rely on search engines for information in critical contexts, such as public health emergencies–but what makes people trust some search results more than others? Can search engines influence people’s levels of trust by controlling how information is presented? And, how does the presence of misinformation influence people’s trust? Research has identified both rank and the presence of misinformation as factors impacting people’s search behavior. Here, we extend these findings by measuring the effects of these factors, as well as misinformation warning banners, on theperceived trustworthinessof individual search results. We conducted three online experiments (N = 3196) using Covid-19-related queries, and found that although higher-ranked results areclickedmore often, they are not moretrusted. We also showed that misinformation does not damage trust in accurate results displayed below it. In contrast, while a warning about unreliable sources might decrease trust in misinformation, it significantly decreases trust inaccurateinformation. This research alleviates some concerns about how people evaluate the credibility of information they find online, while revealing a potential backfire effect of one misinformation-prevention approach; namely, that banner warnings about source unreliability could lead to unexpected and nonoptimal outcomes in which people trust accurate information less.
AI writing assistants powered by Large Language Models (LLMs) are increasingly used to make autocomplete suggestions to people as they write. Can these AI writing assistants impact people’s attitudes in this process? In two large-scale preregistered experiments (N=2,582), we exposed participants writing about important societal issues to an AI writing assistant that provided biased autocomplete suggestions. When using the AI assistant, the attitudes participants expressed in a post-task survey converged towards the AI’s position. Yet, a majority of participants were unaware of the AI suggestions’ bias and their influence. Further, the influence of the AI writing assistant was stronger than the influence of its suggestions when presented as static text, showing that the influence is partially due to AI writing assistants leading people to incorporate biased viewpoints into their own writing. Lastly, warning participants about assistants’ bias before or after exposure does not mitigate the influence effect.