Helsinki researcher urges AI-free practice to protect professional skills

AI can serve as a professional assistant while weakening the skills needed to catch its mistakes, University of Helsinki philosopher Kaisa Kärki argues. In an October 27 research release, the university highlighted her case for protecting human expertise—and questioned how much responsibility workers should bear when that erosion happens without their awareness.

The paper, “Who is Responsible for Tackling Professional Deskilling?”, appeared in Human Affairs on September 23, 2026. It examines deskilling: the deterioration of professional abilities as people hand tasks to AI. Kärki’s contribution is a philosophical argument about responsibility and prevention, rather than a new experiment measuring how much skill professionals have lost.

When reliance becomes hard to see

Her concern starts with the convincing responses produced by large language models. Professionals may come to trust those responses too much, even though the systems make errors. The danger, she argues, is that people may not recognize either their growing reliance or the weakening of their own abilities. That makes individual awareness an uncertain safeguard against the problem.

Kärki uses a hypothetical medical error to make the responsibility question concrete: a doctor relies on a language model and misses its inaccuracies. Philosophical accounts commonly tie responsibility to actions taken with awareness, she explains. But if the doctor does not notice a bias toward trusting AI, assigning responsibility for the resulting skill loss and mistakes becomes less straightforward.

That does not place every consequence beyond individual responsibility. Kärki’s position is narrower: professionals cannot bear responsibility for all deskilling-related problems when the process can escape their notice. She treats it as a societal and structural issue, placing part of the burden on professional communities rather than leaving each worker to detect the change alone.

Deskilling involves major risks that humanity and professions should pre-emptively consider before skills decay at the community level,

Kaisa Kärki, in the University of Helsinki research release

Practice without the assistant

The proposed response begins with a practical question: which abilities must people retain when AI is unavailable? Kärki points to pilots rehearsing autopilot failures as a model for other professions. Each professional community, she argues, should identify the core skills that need to remain in human hands, rather than treating every task as equally suitable for delegation.

In medical education, she cites arguments for teaching basic skills through stages without AI. Diagnostic reasoning is one example, alongside a general understanding of medicine and its tasks. Her recommendations extend beyond students: experienced professionals should also have their core skills tested regularly. Seniority alone is not presented as a reason to skip those checks.

Other safeguards Kärki recommends

• Reserve some tasks for humans, without the option of handing them to AI.
• Write workplace AI-use guidelines that explicitly address skill decay.
• Develop financial incentives to preserve professional skills.
• Develop legislation for fields where deskilling threatens processes critical to society.

A profession’s ability to teach itself

The longer-term concern reaches beyond one worker’s performance. Kärki warns that unchecked dependence could leave communities unable to train new professionals without AI. In that scenario, the loss is not simply an individual’s ability to finish a task unaided. It is the profession’s capacity to pass on expertise while remaining alert to the technology’s errors.

Kärki also calls for more research on AI-related skill decay outside medicine. The scope of that recommendation matters: her proposed safeguards address a risk that professions should investigate and anticipate, not a measured rate of decline established across doctors, researchers and software developers.