How Tech Workers Contend with Hazards of Humanlikeness in Generative AI

Tech workers themselves are uncertain about AI humanlikeness risks: The study found an "unsettled knowledge environment" among 30 professionals across six job functions, indicating that even those building and deploying generative AI lack clear shared understanding of what makes these systems seem human and what dangers that creates.

Humanlike qualities drive both adoption and harm: While AI's human-seeming characteristics are accelerating its use in workplaces, these same qualities create specific hazards—particularly "overtrust in system outputs"—where people may rely too heavily on AI recommendations without appropriate skepticism.

The risks extend beyond individual users to entire fields and society: The research identified potential harms at multiple levels, not just for individual workers but for entire knowledge work sectors and broader societal impacts, suggesting the problem is more systemic than commonly discussed.

"Humanlikeness" itself needs clearer definition: The paper argues that technology workers need better conceptual frameworks to understand what exactly makes AI seem human, as conflated or vague notions of humanlikeness make it harder to identify and prevent specific risks.

Support systems for responsible AI deployment are inadequate: The findings suggest that current guidance and resources are insufficient for helping tech professionals navigate these challenges, pointing to a gap between the pace of AI deployment and the infrastructure needed to manage its humanlike qualities responsibly.