Relying on AI at Work May Quietly Erode Your Expertise

October 28 — AI already acts as an assistant in many professional fields; doctors, researchers and software developers all use it at work. The trouble is that this produces what researchers call skill degradation: people hand work to AI and their own professional competence declines with it.

Kaisa Kärki, a researcher in theoretical philosophy, examines the problem in a paper titled “Who should be responsible for dealing with skill degradation?”, published in the journal Human Affairs.

“One difficulty with skill degradation is that practitioners find it hard to notice the change in their daily work, partly because of cognitive bias about relying on AI. When we talk to a large language model and get a plausible-looking reply, it is easy to become over-reliant. But these models make mistakes too.”

Kärki uses medicine as an example: if a doctor relies on an LLM, fails to spot wrong information the model provides, and a medical incident follows, who should be responsible?

“Generally, philosophical theories of responsibility hold that an agent is only responsible for actions taken knowingly. The unnoticed cognitive biases that come with AI use challenge that view: after all, it is hard for us to notice our own biases or our tendency to over-rely on AI. So whether we should bear responsibility for this silent loss of competence, and for the mistakes that follow, is not settled.”

Kärki points out that skill degradation is hard to notice, so responsibility for the problem cannot be pushed entirely onto individual practitioners: it is a social and structural problem.

Each professional community needs to define which capabilities are core to its field and must remain firmly in human hands.

Pilots repeatedly practise what to do when autopilot fails. Other industries should ask themselves the same question: when necessary, can practitioners do the work independently without AI?

“In medical education there is already an argument that basic skills training should include phases where AI is banned. Diagnostic reasoning, for instance, is a core medical capability, and so is a holistic understanding of the field and its work.”

If workplaces keep expanding AI use while ignoring the risks of skill degradation, we may in future lose the ability to train new people without AI. In other words, people will become deeply dependent on the technology and lose the ability to tell when it is wrong.

The resilience of human communities rests on a diverse reserve of human skills.

“Skill degradation carries enormous risk. Societies and professions should consider this problem in advance, before competence declines across the board at an industry level.”

Practitioners must be made aware that using AI can quietly cause skill degradation;

even experienced veterans should be periodically assessed on core competences;

some work should be reserved for humans and not handed to AI;

employers need AI-use policies that account for the risk of skill erosion;

professional education should include learning stages where AI is banned;

incentives should be designed to encourage people to maintain their professional skills;

in fields where skill degradation would threaten critical social functions, regulation should govern AI use;

industry communities should define the core skills their field must keep in human hands;

and research into AI-driven skill degradation should be extended beyond medicine to more fields.