The new research highlights the potential drawbacks of using artificial intelligence in the workplace. People who rely on AI tools to get the job done are often judged more harshly by others. Despite the possibility that AI can increase productivity, workers may face negative social assessments, particularly in terms of ability, effort, and motivation, just to use it. Research published in Proceedings of the National Academy of Sciencesprovides evidence that this perception can even affect employment decisions.
Generic artificial intelligence refers to tools such as ChatGpt or other systems that can generate human-like text, images, or code. These tools are becoming increasingly common in workplaces across the industry, providing support for writing, analytics, creative tasks and more. It can improve performance, but some people worry that relying on AI makes them seem lazy or replaceable. That concern may be unfounded.
The research team – social scientists studying how people evaluate each other in the expert context – wanted to understand whether using AI actually harms people's reputation in the workplace. Using attribution theory, which explores how people explain the behavior of others, researchers reasoned that observers might view AI use as a sign of lower ability or effort. As previous research has shown that people sometimes assume that seeking help reflects personal weaknesses rather than situational needs, the team predicted that AI users could face similar biases.
“After conversations with employees from several different organizations who were hesitant to hesitate to use generated AI in the workplace, not because of the issue of output, but because of social dynamics associated with AI use,” he said. “These discussions, coupled with industry research suggesting that people are hiding AI use in the workplace, made me want to look into whether employees with social assessment penalties are expected to be real.”
The researchers conducted four experiments with over 4,400 participants. The research is premised, meaning that the team has documented methods and hypotheses in advance to reduce bias. In all experiments, researchers looked at both people's expectations about how they are judged by using AI, and also how they can actually evaluate others who use AI tools.
In the first study, we imagined 497 participants completing tasks at work using either a generator AI tool or a traditional dashboard tool. They then rated that their managers and colleagues thought they would perceive them. Those who imagined using AI are expected to be judged as lazy, more interchangeable, capable and unindustrious than those using non-AI tools. They also said they are unlikely to disclose AI use to others. These responses suggest that workers are aware of possible stigma with regard to AI use.
But are these fears justified? In the second study, the researchers asked 1,215 participants to read a brief description of the virtual employee and assess the various characteristics. Employees were said to have received help from either AI, human colleagues or no support. Participants consistently rated employees who used AI to be more lazy, less competent, hardworking, and non-independent than employees in the other two conditions. The effect was not tied to the employee's age, gender, or occupational type. These results show that people really have negative views of AI users, even if the help received is essentially the same.
“We were surprised that the social assessment penalty in our study occurred regardless of the age, occupation or gender of the employees we described,” Reif told Psypost. “In total, we compared 384 unique stimuli that systematically altered these attributes, but in addition to whether the target was explained to be getting AI help, getting help from another source or not at all, I thought the ratings of AI users, for example, would differ depending on whether he had discovered his career.
The third study tested whether these perceptions influenced actual decisions. In this case, one group of participants (801 people) completed the visual task and reported how often they used AI. The second group (1,718 people) played a role in hiring managers and selected among candidates who used AI every day or never used it at all. They said their pay depends on the candidate's performance, which gave them an incentive to make careful choices.
Overall, managers who did not use AI themselves tended to support candidates who also did not use AI. In contrast, managers who used AI were more likely to prefer candidates who used AI every day. This suggests that people's own experiences with AI influence how they view others who use it. People new to this technology may be more skeptical or suspected of their users.
The final study aimed to understand why these biases occur and whether they could be reduced. The researchers asked 1,006 participants to evaluate virtual candidates who would apply for either manual (e.g., handwritten notes) or digital (e.g., sending personalized emails). Some candidates were described as using AI regularly, while others used traditional tools like Microsoft Office. Participants who did not use the AI themselves were more likely to consider AI users as lazy. This recognition led to a low rating of job fit, especially for manual tasks.
However, if the task was digital and clearly suited to AI support, the penalty disappeared. In fact, AI users were even seen as slightly more suited to digital tasks than non-AI users. Researchers also found that frequent AI users are less likely to consider AI users as lazy, suggesting that familiarity can help reduce stigma.
“The main point from this work is that the use of AI is social costs,” explained Reif. “In our study, employees who are described as using generative AI were rated as lazy, competent and less industrious than employees using other tools and workplace help sources to perform the same task. The irony of our findings is likely to use AI because they are motivated to be more productive at work, but other employees may be less motivated.”
One strength of the study was its experimental design, allowing the team to separate the effects of AI use from other variables. However, the author warns that there are some limitations to his work. All studies relied on online samples rather than actual organizations, which could affect how well the results were converted into a real workplace. Additionally, the description of AI tools may be intentionally broad and may not reflect the various tools and use cases that exist today.
“A significant limitation of our study was that raters were not personally aware of the targets they were evaluating, and simply reported first impressions when they read about them,” Reif noted. “For example, we cannot say how the perceptions of employees with a long-standing reputation as hardworking will change when they start using AI. If the evaluator has more knowledge about the target or existing work relationships, the effectiveness of observing in the research may be weaker.”
“One of the future directions I'm excited about is to unpack the reasons why social assessment penalties we document are occurring. For example, evaluators may be assuming how people using AI are saving.
The study, “Evidence of Social Assessment Penalties for Using AI,” was written by Jessica A. Rafe, Richard P. Larick, and Jack B. Sol.
