When Bots Become Bosses: Study Finds Humans Still Shape AI Moderation

Organizations increasingly delegate oversight of employee work to automated software agents, deploying bots to apply consistent rules and evaluate output. But these systems are not infallible, and new research suggests that the humans they govern are finding ways to push back.

Jason Thatcher, the Tandean Rustandy Esteemed Endowed Chair in information systems and analytics at the Leeds School of Business, University of Colorado Boulder, studied how artificial intelligence functions as a moderation agent. His research examined actual ratings generated by a bot on Wikipedia that assesses the accuracy of contributions to the platform's pages.

Thatcher describes such a bot as an automated software agent granted the authority to fact-check or verify the correctness of human contributions. These agents scan records for inconsistencies and flag unusual activity. In one example, the bot detects vandalism on personal profiles — such as mischaracterizing the record of football coach Deion Sanders — and reverts the page to its prior state.

The appeal of inserting bots into platform regulation, Thatcher notes, is that it creates a perception of consistency and fairness. Yet because bots make mistakes, human governance remains necessary.

What the study found was not passive acceptance. People subject to bot moderation did push back. While they could not directly alter the bot's behavior, they could escalate their concerns to the committee that decides what the bot can and cannot do. That committee would then approach the developers, who would adjust the bot to address its deficiencies.

The result, according to Thatcher, is a more effectively functioning ecosystem in which bots regulate people appropriately rather than making repeated errors in correcting contributions. He characterizes the process not as simple bot regulation but as a bot-human-bot negotiation.

Thatcher's broader expertise covers how people think about and use technology in organizations and daily life, including cybersecurity, strategic decision-making, and innovation under workplace stress. He points out that automated agents are already pervasive — appearing when a worker makes a typo in a form, proactively ensuring correct formatting, flagging inappropriate tone or sentiment, and returning the work for correction.

While the prospect of bots regulating human activity may appear alarming, Thatcher highlights a more encouraging narrative: even when governed by automated systems, people can take action to change how those systems operate when they fail. The human voice, he argues, still matters — content creators and contributors retain a say in how their work is interpreted and published.