A new wave of tools from Microsoft, OpenAI, Anthropic, and Google is framing AI agents as digital colleagues with human-like capabilities. This shift is raising concerns about how humans interact with and manage these tools. According to Emma Wiles, a Boston University business professor, managers who treat AI agents as coworkers are more likely to make errors. In her study, participants caught 18% fewer errors when the work was said to come from an agentic 'AI employee' rather than a chatbot.
This suggests that how we label AI tools significantly affects our performance with them. The research also found that participants saw themselves as less responsible for the output of AI tools labeled as employees. They were 44% more likely to escalate questionable work to a manager rather than correct it themselves, undermining the efficiency of using AI agents. This mislabeling could have broader implications beyond the workplace, especially as AI agents are integrated into fields like healthcare, warfare, and government.
The study highlights the risks of assigning human-like roles to AI tools, which could lead to misplaced blame for failures. The researchers emphasize that AI agents should be designed to enhance human capabilities rather than replace them. They suggest that the current approach of treating AI as coworkers may be counterproductive, as it sets unrealistic expectations and undermines human responsibility. The findings raise important questions about how organizations should integrate AI tools into their workflows.
Source: mittr
