AI Surveillance in the Workplace Linked to Employee Resistance, Turnover

Workers tend to react negatively to on-the-job surveillance of any kind, but recent research from Cornell University found that being monitored by AI in particular results in even more dissatisfaction and greater resistance.

Some organizations are implementing AI and algorithmic tools to monitor employee behavior and performance, which may be decreasing productivity and leading to more people quitting, the research concluded. AI-based surveillance most often takes the form of tools that monitor keystroke movement and track time online to gauge worker activity.

“We found consistent evidence that algorithmic surveillance led participants to perceive they had less autonomy and to engage in more resistance behaviors, such as complaining more, performing worse, and intending to quit,” said study co-author Emily Zitek, associate professor of organizational behavior at Cornell’s School of Industrial and Labor Relations in Ithaca, N.Y.

“No one likes the idea of Big Brother,” agreed Katheryn Brekken, senior research analyst at the Institute for Corporate Productivity (i4cp) in Seattle. “What’s at stake is customer and employee trust. i4cp research shows that productivity flourishes in environments of high trust, and out of all the dimensions of trust we studied last year, employees’ trust in senior leadership was the most impactful.”

When employees know their activity is monitored, “it leads to performance [not genuine] work—counterintuitive efforts like mouse jiggling—that are anything but productive,” she said. “These AI surveillance tools are supposed to be a mechanism to understand employee performance, but clearly, if not used or implemented appropriately, they can sink productivity.”

Brekken said the good news is that so far, very few organizations are conducting AI surveillance. According to a 2023 i4cp study, only 6% of large companies reported they use employee surveillance tools.

AI Can Give Helpful Feedback

Interestingly, the Cornell study also found that when participants were told that an AI tool would be used to provide developmental feedback, they resisted less.

“The lack of perceived autonomy and increased resistance can be alleviated if algorithmic surveillance is viewed as developmental,” Zitek said. “Thus, algorithmic surveillance will not necessarily lead to adverse outcomes for individuals—it just depends on how it is implemented. The problem occurs when people feel like an evaluation is happening automatically, straight from the data, and they’re not able to contextualize it in any way.”

Brekken said that a consistent theme emerges when thinking about how AI should be implemented at work. “How these tools are communicated and used is crucial,” she said. “People analytics leaders often stress how important it is for organizations to inform employees what data is being collected and for what purposes. The reporting needs to be in aggregate numbers so that individuals cannot be identified. To ensure this happens, particularly as the amount of data [that] companies have access to continues to grow, organizations need strong data governance and ethics policies.” 

The Cornell research raises interesting questions about who should receive the data and why that matters in terms of trust, Brekken added.

She explained that if AI is framed as a way to enhance personal and team development, it has powerful potential to improve coaching and development. “The AI in this case could give you direct feedback after a meeting to tell you how to perform better,” she said. “This is like having a personal fitness tracker. Once you see data on your performance, it’s hard to ignore and kind of addicting. But the point is you’re in charge of the data and you get to decide what to do with it. You have agency.”

If, on the other hand, the data is sent directly to a manager and the employee’s performance is perceived incorrectly, the technology has high potential to cause harm, Brekken noted.