How You Treat an AI Agent Determines the Results You'll Get, Says Professor Taha Yasseri

Why It Matters

Misaligned human‑AI interaction can erode productivity and amplify bias, threatening ROI on enterprise AI deployments.

Key Takeaways

• Human mindset drives AI performance more than technology
• Gendered AI voices trigger sexist bias in decisions
• Delegated AI input receives lower peer evaluation
• Training needed for effective human‑AI collaboration
• Early regulation parallels electricity commercialization risks

Pulse Analysis

Recent insights from Professor Taha Yasseri, head of the Joint Centre for Sociology of Humans and Machines (SOHAM), underscore that the quality of results from AI agents hinges less on raw algorithmic power and more on how users engage with them. By applying a ‘theory of mind’ to machines—understanding their capabilities, limits, and decision processes—workers can unlock a synergistic partnership that outperforms even the most advanced models. Yasseri’s studies reveal that when employees treat AI as a transparent tool rather than an omniscient oracle, task efficiency and accuracy improve markedly.

The research also surfaces troubling social dynamics that travel with AI adoption. Voice‑based agents with feminine tones provoke stronger sexist reactions, leading to harsher feedback on decisions made by a female‑presented AI manager. Likewise, contributions routed through an AI intermediary are consistently rated lower than those authored directly, despite identical quality. These biases mirror long‑standing human‑to‑human prejudices and can erode trust in automated systems, skew performance metrics, and ultimately diminish the return on investment for enterprise AI initiatives.

Enterprises must therefore embed structured training and change‑management programs that teach employees to calibrate expectations, recognize AI limits, and mitigate unconscious bias. Drawing a parallel to the early regulation of electricity, Yasseri argues that proactive policy, design standards, and continuous knowledge generation are essential to prevent costly missteps as AI agents proliferate across workflows. By institutionalizing these safeguards now, organizations can accelerate adoption while protecting against reputational damage, legal exposure, and productivity loss, positioning AI as a reliable partner rather than a source of hidden risk.