In today’s day and age, the use of AI has almost become ubiquitous across many different fields and platforms. It has evolved so quickly that many institutions, whether for work or education, are still learning how to deal with it. In universities and education-based settings, for example, many teachers have adopted their own approach, with some people permitting AI with citations, and others outright banning it. With all this discourse, studies on the social impacts of AI adoption are necessary and contribute to the broader conversation around AI usage in educational settings.
Alex Kale, Assistant Professor at the University of Chicago Department of Computer Science, recently published a paper titled, “Underreporting of AI Use: The Role of Social Desirability Bias” alongside coauthors Yier Ling (PhD student in the Department of Economics) and Alex Imas (Professor at Booth School of Business). Making its debut at the leading human computer interaction conference, CHI 2026, the paper sought to understand the social impacts of AI adoption, and furthermore, whether social desirability bias plays a role in discussions and environments fostering AI use. Social desirability bias is a phenomenon where people self-report their behavior in a way that would be viewed more favorably by an outside party, because even when anonymized, people generally don’t want to admit to something that would be viewed as unfavorable.
“I was fascinated by this topic because the problem is really important right now,” Kale stated. “A lot of educational institutions are thinking about the use of AI in schools and classrooms, and how it changes learning and the environment. This was a project that was really trying to look directly at the underappreciated social impacts of AI adoption, and a broader conversation around the way that people view their own AI usage and the AI usage of their peers.”
To mitigate potential self-reporting bias in these surveys, the authors approached this survey using both direct and indirect questioning: the same set of questions are asked to students but framed slightly differently. One question may be asked directly, “how frequently do you use AI?” while the other asks indirectly, “how frequently do your peers use AI?” Interestingly, when asked these sets of questions, respondents describe a large discrepancy between their own AI use versus that of their peers: approximately 60% of students reported that they used AI, compared to the reported 90% of their peers. Given every respondent is part of the same population as their peers, these numbers should be estimates of the same statistic at the population level and should not be so far off on average. The authors sought to understand where the large discrepancy between the reported percentages is coming from. One hypothesis they had was that the reporting gap was driven by social desirability bias.
To check whether this explanation is consistent with students’ perspectives, the authors then created a second survey to ask undergraduate students what they make of the first survey’s results. Students largely pointed to feelings of shame and concerns about being perceived as competent enough to do the work. This provided evidence that AI adoption was associated with negative emotions, which may lead students to underreport their usage and confirm that social desirability bias explains the discrepancy.
“These findings were really striking to me, not only because of its magnitude, but because of what it reflects about the campus environment,” Kale reflected. “The thing that struck me most was just how emotionally charged all of this is. It was such a stark reminder of the degree to which students are in a difficult bind with AI. On one hand, they feel that they need to use it to keep up, whether that’s in terms of class performance or the time it takes to juggle their course load, the perception that virtually all of their peers are using it, or even labor market pressures outside the university. On the other hand, there’s a lot of things in the campus environment, from peers and professors, that suggest that AI use is bad. Students are in a truly impossible situation.”
Kale had many great conversations with other faculty and educators at CHI 2026 that were confronting the same issues and concerns with AI use in the classroom. He views this paper as part of an ongoing conversation about how educators and institutions should appropriately react to AI use. In his own teaching, he has become more permissive about AI use and includes explicit instructions on using and experimenting with AI, to foster an environment of epistemic humility where students are encouraged to explore ways of responsibly enriching their work with these tools.
In the future, he is looking ahead to more collaborations on AI related projects. Kale hopes that the publication of this paper sparks an important conversation and that more studies, including those on quantification of social desirability bias and impacts of different AI policies, can lead to different social norms around AI use that are better for teaching and learning.
To learn more about his work, visit Kale’s website here.
People
