Deskilling Is Already Happening: Evidence Across Domains

The Pattern Beneath the Hype

It is tempting to treat the deskilling AI can cause as a future risk, something to worry about once the technology is more capable. The evidence says otherwise. Across very different fields, driven by the same human factors, the erosion is already measurable. The lesson from each domain is the same one, dressed in different clothes.

What Deskilling Actually Looks Like

Deskilling is rarely dramatic. It shows up in four recognisable ways:

• Reduced situation awareness. Monitoring an automated system without actively engaging with it lowers awareness and slows recovery when the system fails.
• Automation bias. Too much trust in AI weakens independent checking and judgement, producing both errors of omission and errors of commission.
• A collapse when AI is removed. Performance holds steady with assistance, then falls sharply the moment the AI is withdrawn, exposing how much underlying skill has been lost.
• Skill atrophy. The slow deterioration of cognitive and procedural skills that are simply no longer practised.

The first three are warning signs. The fourth is the underlying disease.

The Evidence Across Domains

This is not theoretical. The same effect appears wherever capable automation meets skilled human work.

• Medicine. A 2025 multicentre endoscopy study observed an effect on operator performance and confidence when AI assistance was withdrawn after a period of routine use.
• Diagnostic judgement. In controlled trials, diagnostic accuracy fell by roughly eleven percentage points when clinicians were shown biased model recommendations, a clear signature of automation bias.
• Aviation. Regulators report that heavy reliance on automation erodes manual flying proficiency, to the point that pilots are now required to practise manual skills deliberately and regularly.
• Automated driving. Drivers using vehicle automation show slower reactions and weaker vehicle control when they have to take back the wheel at short notice.

Different professions, different stakes, one pattern. Cognitive and procedural skills follow a simple rule regardless of domain or expertise level: use it or lose it.

A Predictable Curve

Read across the evidence and the trajectory is strikingly consistent. Performance with AI follows a predictable curve: an initial improvement as the tool takes load off the user, followed by growing dependency, followed by erosion of the skill underneath. Early work also suggests this degradation accelerates under time pressure, precisely the conditions where AI is most tempting to lean on and where lost skill is most dangerous.

The highest risk sits where the stakes are high, where errors carry significant consequences and AI assistance is becoming common at the same time. That combination, serious consequences plus thinning human skill, is exactly the one organisations cannot afford to drift into unmonitored.

The Talent Pipeline Problem

There is a further effect that is easy to overlook. Junior positions have always done double duty: they get work done, and they build the foundational skills that produce tomorrow’s experts. When AI quietly absorbs that junior work, the immediate output looks fine, but the development pathway disappears.

Follow that logic forward and you get a familiar, avoidable failure:

The expertise gap timeline. Each step looks reasonable in isolation. The endpoint is a system no one understands.

The way out is to build apprenticeship back in: design AI so that it augments learning rather than replacing it, and protect the practical experience that turns juniors into experts.

Why This Is a Business Continuity Issue

Here is the part that moves this from an interesting research finding to a concern for the board. AI presents a genuine paradox: the same technology that can capture and preserve expertise can also be the cause of its erosion.

The opportunity is real. As experts correct AI, their tacit knowledge can be captured, documentation can emerge from daily workflows, and institutional memory can become searchable and transferable. We have written about that upside in AI for knowledge transfer.

But the risk runs in parallel. Skills atrophy without practice. Experts retire without successors. Teams lose the ability to validate what the AI tells them. And the capacity to innovate, which depends on deep human understanding, quietly diminishes.

The conclusion is blunt: without skill continuity, there is no business continuity. An organisation that has outsourced its critical judgement to a system it can no longer check has not become more efficient. It has become more fragile.

This is the second post in our series on AI and human expertise. The first looked at the cognitive cost of convenience and the research behind it. The third turns to the practical question: how to design AI systems that protect expertise rather than drain it.

If you want a clear view of where your organisation is most exposed to skill erosion, an AI readiness assessment is a good place to start.

Concerned about deskilling in your own teams? Get in touch.