From Future of Work to Future of Workers: Addressing Asymptomatic AI Harms for Dignified Human-AI Interaction

Review

This paper examines a subtle but consequential risk emerging from AI deployment in professional environments: systems can improve short-term productivity while quietly degrading the human capabilities required to supervise them effectively. Drawing on a year-long qualitative study of AI-assisted radiation oncology planning involving clinicians, dosimetrists, and physicists, the authors identify what they call an AI-as-Amplifier Paradox. AI can increase throughput and consistency while simultaneously weakening professional intuition, vigilance, and long-term expertise.

The authors describe a progression of “asymptomatic harms.” Early effects include reduced attentional engagement and declining reliance on domain intuition as practitioners increasingly trust automated outputs. Over time these effects may develop into deeper structural harms such as deskilling, dependency on automated systems, and professional identity erosion. The paper frames this final stage as identity commoditization, where experts begin to perceive their role as merely validating or operationalizing AI outputs rather than exercising independent judgment.

Methodologically, the study combines think-aloud sessions, interviews, and participatory workshops over a year-long period. This longitudinal design is particularly valuable because it allows the researchers to observe slow changes in professional behavior rather than only immediate reactions to new tools. The findings are synthesized into a conceptual framework for Dignified Human-AI Interaction that spans worker, technology, and organizational levels.

The paper’s central contribution for AI governance lies in reframing the “human-in-the-loop” assumption. Simply keeping humans present in a workflow does not guarantee meaningful oversight if their expertise is gradually weakened. Governance systems must therefore evaluate whether AI deployments preserve the human capacity to detect errors, challenge automated decisions, and maintain institutional accountability over time.

One of the paper’s most practical proposals is the concept of Social Transparency. Rather than focusing only on algorithmic explainability, the authors emphasize visibility into the broader socio-technical context of AI decisions: who interacted with the system, what assumptions shaped the output, and how responsibilities were distributed across humans and machines. This approach aligns more closely with how real organizations assign responsibility and conduct oversight.

For public sector AI systems and digital public infrastructure, the implications are significant. Governments often measure AI success through efficiency gains, throughput improvements, or cost reductions. The paper suggests that such metrics can conceal deeper governance risks if they fail to track whether human expertise and institutional judgment remain intact.

A practical governance response would treat human capability retention as a safety control objective. This could include mechanisms such as periodic AI-off drills, independent verification tasks, skill maintenance protocols, and oversight metrics that measure not only output accuracy but also the robustness of human review capacity.

In short, the paper argues that the long-term safety of AI systems depends not only on the quality of their outputs but on the durability of the human institutions that supervise them. Systems that appear successful today may quietly undermine the human expertise required to govern them tomorrow.

Key Insight

AI systems can improve visible performance while gradually eroding the human expertise, intuition, and professional agency needed to detect and correct system failures. Governance frameworks must therefore treat human capability retention as a safety objective.