Cognition in an era of AI – what separates knowledge and expertise?

Knowledge has become a hollow noun. Anyone with access to the internet can search for what they need; nowadays we can all ask our AI agents to scan the internet for us. But asking for the correct answer isn’t the same as understanding it. Expertise isn’t the same as knowledge. With the daily use of AI and continuous cognitive offloading, is our ability to develop expertise and maintain it diminishing?  

We’ve seen this difference up close and personal and make it a core part of our business to find true expertise. That’s not a positive, it’s a necessity. Working with advanced STEM projects around the world, there’s no room for second guessing. As one partner explained ‘...we want a team that doesn’t just deliver, but questions and challenges.’  

So, what’s happening to expertise, why is it getting harder to find and how can businesses separate knowledge and expertise?  

How quickly can you lose expertise while using AI tools?  

Within months. Habitual cognitive offloading can atrophy professional expertise quicker than previously thought, perhaps as quickly as a few months. 

One study found that regular use and reliance on AI tools can erode professional expertise within a few months, not just years. Diagnosing doctors saw their successful diagnoses of prostate cancer fall by 6% within a matter of months after using AI tools. (TheLancet, 2025) 

While this doesn’t necessarily ring alarm bells, it does illustrate just how quickly our hard-fought expertise can erode if not exercised. If we take this idea and apply it to other industries, and especially to new graduates entering the market, there is a significant concern that despite passing exams, new professionals will never properly develop the expertise of their older peers. But what about those existing experts?  

Do AI tools prevent professionals from developing expertise?  

Yes. AI mediation activity impedes human capacity cultivation, disrupting human-to-human interactions which help nurture know-how and understanding.

Essentially replacing nuanced conversations, probing questions and debate for quick answers removes the opportunity for expertise to develop. Avigail Ferdman of the Technion–Israel Institute of Technology argues that the more AI replaces valuable human activity, the more it risks deskilling humans of their human capacities. She introduces the concept of ‘capacity-hostile environments’ - settings where AI mediation actively impedes human capacity cultivation. (HBS, 2023) 

What is capacity cultivation? It refers to the idea that acquiring agential control over skills requires a gradual process of habituation. Habituation, in turn, depends on learning from others: the "know how" of the skill, as well as a shared understanding of the value of the skill. (Erigo, 2025) 

Which types of roles are most affected by diminishing expertise? 

Narrowly focused specialisms. AI can do a lot, closing the knowledge gap between novice and expert, but a gap will always remain. That gap widens depending on the role, with highly specialized positions harder to synthesize with AI knowledge.  

A prevailing assumption of AI is that it will democratize knowledge. And to a certain extent, that is true. However, knowledge isn’t equal to expertise and with that in mind AI still has a gap to bridge. In a study of 78 employees at a UK-based global trading company, AI narrowed the gap between web analysts and marketers – but a wall appeared for developers and data scientists. (Springer, 2025) 

What is the difference? Web analysts and marketers share a foundational sensitivity to audience needs, conversion strategies, and effective marketing copy; that background let them use AI's suggestions wisely. Employees in more specialist roles could not replicate the same results.

The erosion is reversible. But only if it’s noticed and often it is only noticed when AI is removed from the equation.  

AI induced cognitive erosion is not dementia; brain cells aren’t dying; they are just atrophied. The very same neuroplasticity which means skills can decline, makes it possible to strengthen ‘lost’ skills. Think of it as a muscle, when you use your own mind for a task repeatedly your cognitive ability increases. Don’t use it, and you’ll lose it. (GoodwinLaw, 2025) 

But there’s a catch. New research has uncovered a phenomenon known as ‘the false expert transition’ - situations where apparent expertise masks knowledge gaps. Studies focusing on law and medical professionals show measurable cognitive decline within months of using AI. So, while these professionals may seem accurate, they are increasingly dependent on AI tools. If that cycle continues, it might be difficult to catch cognitive decline before serious impacts are noticeable for businesses or individuals. (Anthropic, 2025) 

It’s complicated. Generally, it is believed that the low, methodical way of learning; research, reading, drafting, debating etc., develops key aspects of expertise.

One of the best illustrations of this new dynamic comes from studies exploring medicine and law; two traditional professions with long learning runways and high stakes. Surveying nearly 900 UK lawyers, one report found AI tools have dramatically reduced the time juniors spend on research, drafting and document review – the bread-and-butter tasks that have previously defined the early years of a legal career and serve as an informal training ground for developing legal instincts. 72% of respondents identified deep legal reasoning and argumentation as the biggest skills gap among junior lawyers. Just 2% believe AI actually strengthens learning. (Legartis, 2026) 

It’s not just lawyers and doctors who are affected. Many students delegate the highest-order cognitive tasks to AI more than any other. Across 574,740 student conversations analysed, students delegate creating and analysing to AI more than any other cognitive level. These are precisely the tasks through which novices develop judgment. (LexisNexis, 2026) 

This reality is so concerning that some US law firms are restricting juniors from delegating tasks to AI before they have fully grasped critical concepts, reasoning and developed effective judgement. (PwC, 2026) 

Yes, although it’s not a one-size-fits-all answer.  

The real question is, what kind of expert are you? Some roles are improved more effectively with AI than others. In most cases AI tools can widen the gap between good and great, helping those with developed expertise maximize their time and focus on tasks which require their judgment and instincts.  

Professionalised jobs that allow AI to amplify human expertise are growing twice as fast as democratised ones, and with 42% higher wage growth. The new tasks being added to AI-exposed roles are 2.5 times more likely to rely on skills like empathy, judgment, and creativity. (ThomsonReutersInstitute, 2026) 

Novices are also benefiting from AI; underperforming professionals discovered a 43% positive increase in their output compared with a 17% increase for high achievers. Although that extra boost comes at a cost. Known as the ‘sycophancy trap’, there’s a body of evidence which suggests misconceptions and assumptions are rarely challenged by AI systems and usually agreed with. Without adequate judgement, novice professionals can get stuck in a web of confirmation bias with little to no pushback. (HBS, 2025) 

Where can you find the best experts?  

Coalesce Management Consultancy. Many consultancy firms provide a generalist approach – filling the brief with as little push back as possible. CMC is different.  

We see the big picture, not just the parameters of a brief; we ask probing questions, cross reference our experts to be sure they can deliver, even when the odds are stacked against them. In short, we only operate with tried and tested expertise.  

Today the big picture is complicated – with AI rolling out throughout industries it seems as if the recent wave of research has done little to curtail business enthusiasm for AI tools. While there are legitimate concerns, the prevailing thought is that we’re in a transitionary period. Growing pains of a new era.  

If you’re looking for the right expert to guide your business, solve a technical headache or free a logistical bottleneck, get in touch. With thousands of STEM expertise deployed around the world – in advanced engineering, energy, data centres and more – CMC possess both the knowledge and expertise needed to meet your needs. Schedule a call today.  

FAQs

If AI is so harmful to expertise, should professionals simply stop using it? 

Nonowledge to their AI interactions consistently achieve better outcomes — they can spot errors, push back on weak suggestions, and use AI as a tool rather than a crutch. The danger is in removing the formative stages of professional development that build that judgment in the first place. Two-thirds of UK legal professionals surveyed believe treating AI as a “thinking partner” — rather than a shortcut — will improve legal judgment over time. The goal is structured AI literacy, not abstinence. 

Are there professions where AI poses little or no risk to expertise development? 

The risk appears greatest in roles where deep tacit knowledge — judgment developed through years of practice and pattern recognition — is the primary source of professional value. Medicine, law, engineering, and psychology all fall into this category. PwC’s 2026 Global AI Jobs Barometer draws a useful distinction between “professionalised” roles, where AI amplifies expert judgment, and “democratised” roles, where AI primarily lowers the barrier to entry. The latter face a greater risk of expertise becoming irrelevant; the former face a different risk — that the pipeline of experts needed to fill those roles begins to run dry. 

Is anything being done to protect expertise development in a world of AI? 

Yes, and the responses are becoming more structured. In medicine, some institutions have introduced deliberate AI-free periods in training to prevent skill erosion. In law, leading firms have developed multi-week bootcamps for new associates combining simulations and drafting drills before AI tools are introduced. At the regulatory level, the European AI Act now requires all companies using AI to ensure that employees have an adequate understanding of it — and violations can be treated as a breach of due diligence. The consensus among researchers is that the solution is not to restrict AI, but to deliberately design learning environments that preserve the foundational tasks through which expertise is built.