Study reveals a double standard in how we view AI-assisted communication

People expected others to use AI-assisted communication tools more often than they would use them themselves, according to a study published in the British Journal of Psychology. The research also discovered that people viewed openly disclosed AI use more favorably than secret use when the two options were directly compared.

AI-mediated communication includes technologies that assist in communication by changing or generating part of a message. Examples include writing assistants, tools that improve the sound of someone’s voice and video software that changes a person’s appearance or facial expressions. These technologies can make communication faster or more polished, but they may also raise questions about whether a message is authentic.

The researchers were interested in the social rules people apply to these tools. Individuals often hold ideas about the acceptability of behavior, and beliefs about what others are likely to do, but those two judgments do not always match.

Led by Zoe A. Purcell of Université Paris Cité in Paris, the team conducted two online studies. Study 1 included 477 UK adults (mean age 48.48 years; 200 males, 266 females, and 11 participants in other categories). Study 2 included 765 U.S. adults (mean age 45.47 years; 379 males, 375 females, and 11 participants who were nonbinary or preferred not to say).

Participants viewed examples involving text, audio and video. In Study 1, they rated both openly disclosed and secret AI use on a five-point scale. They also rated how likely they would be to use the technology and how likely they thought other people would be to use it. Study 2 used a preregistered design in which participants were placed into separate conditions and rated acceptability and likelihood on a scale from zero to 100.

Free daily newsletter

In Study 1, participants rated open AI use as more acceptable than secret use, with average scores of 3.41 and 2.72 out of 5, respectively. They also expected others to use AI communication tools more than themselves, with average likelihood ratings of 3.73 for others and 2.80 for themselves.

Study 2 produced a more nuanced result. When participants were not shown open and secret versions side by side, the difference in acceptability was small and not statistically significant. Open use received an average rating of 56.87 out of 100, compared with 53.62 for secret use. Transparency also did not change expectations about use. However, the self-other difference remained strong: participants rated the likelihood that others would use the tools at 71.41, compared with 40.44 for themselves.

Participants also considered AI use slightly more acceptable when judging other people than when judging themselves. Acceptability was more closely connected to expected behavior when people considered their own use than when they considered others’ use.

Additionally, greater familiarity with AI was associated with higher acceptability and greater expectations of use. By contrast, people who believed AI communication caused information loss or revealed something negative about a user’s character tended to rate it less favorably. These results suggest that concerns regarding AI-assisted communication may ease as these tools become more mainstream and individuals become more accustomed to them.

Purcell and colleagues summarized their findings: “Self-other differences are a central factor for understanding people’s attitudes and expectations for AI-mediated communication.”

Some limitations are to be noted. For instance, the word “others” was deliberately broad and did not specify whether participants were thinking about friends, colleagues, strangers or family members. In addition, data were collected in 2021 and 2022, before many current AI communication tools became widely used.

The study, “People have different expectations for their own versus others’ use of AI-mediated communication tools,” was authored by Zoe A. Purcell, Mengchen Dong, Anne-Marie Nussberger, Nils Köbis, and Maurice Jakesch.