Home » Insights » Are we thinking less? AI, effort and the skill we can’t afford to lose

Are we thinking less? AI, effort and the skill we can’t afford to lose

TL;DR — About 60% of employees in a new IBM study worry AI is eroding their skills, with critical thinking top of the list. Early research suggests the risk is real but uneven: it is highest when people lack domain knowledge, doubt their own reasoning and trust the tool too much. The fix isn’t avoiding AI. It’s protecting the effortful parts of thinking through better task design, subject-based teaching and workplaces where questioning AI is expected.

AI can now do much of our thinking for us. The open question is what that costs the thinker.

Earlier this year, Tsinghua University education researcher Lixiang Yan was setting up an online course with an agentic AI tool. He soon noticed the AI could complete every assignment, answer every question and join every discussion on its own, with no student mind engaged at all (Nature). His conclusion was not that teaching is obsolete. It was that teaching people to reason for themselves now matters more than ever.

That tension is no longer confined to lecture halls. A new IBM study suggests employees are feeling it at work too, and many of them are worried about what is happening to their own minds.

The worry, in numbers

The concern is widespread, and it is coming from students, educators and employees alike.

Who was askedWhat they saidSource
Employees (IBM, September 2026)About 60% worry AI is eroding their skills; critical thinking is the decline they mention mostForbes on IBM
Chief HR officers (same IBM study)57% rank critical thinking among the most important capabilities staff needForbes on IBM
Higher-education students who use AI (global, 8,000+ respondents, 2026)Around 20% already find it harder to work through problems without itNature
US faculty (1,000+, 2025)90% expect generative AI to reduce students’ critical thinkingNature

The IBM result is the striking one. The skill employers say they value most is the skill workers feel slipping. These are self-reports of worry, not measured declines. But a workforce that suspects its own judgement is weakening is a psychological finding in its own right.

Why effort is the point

Learning depends on mental effort, and effort is exactly what AI is built to remove.

Cognitive psychologist Daniel Willingham of the University of Virginia puts it plainly: “Cognitive processing is typically effortful, and that’s what AI effectively removes” (Nature). Psychologists call handing mental work to an external aid cognitive offloading. We have always done it, with notebooks, calculators and calendars. What is new is that AI can take over the reasoning itself, not just the remembering.

Critical thinking also isn’t a single, portable skill. Researchers describe it as an umbrella for memory, reasoning, decision-making and problem-solving, and it depends heavily on knowledge of a specific field. Being sharp in one domain does not make you sharp in another. That matters, because the people with the least domain knowledge are the ones most tempted to let the tool decide.

Uppsala University’s Thomas Nygren names the risk for science students: they may outsource the very activities that build scientific thinking, such as forming hypotheses, weighing evidence and sitting with uncertainty (Nature).

What the research actually shows

The early evidence leans towards concern, but it is thinner and messier than the headlines suggest.

  • Learning with a crutch, then without it. In a 2024 study, secondary-school maths students did better with a ChatGPT-style tutor. Once it was taken away, they performed worse than peers who never had it (Nature; Bastani et al., PNAS, 2025).
  • Offloading as the mechanism. Michael Gerlich of SBS Swiss Business School studied 666 people in the UK. More frequent AI use went with lower critical-thinking scores, largely explained by cognitive offloading. The 17 to 25 age group used AI most and scored lowest (Societies, 2025).
  • Trust in the tool versus trust in yourself. A Microsoft Research and Carnegie Mellon survey of 319 knowledge workers found that confidence in AI went with less critical-thinking effort, while confidence in one’s own ability went with more (Campus Technology). People often skipped scrutiny when they lacked the skills to check the output.

None of this settles the question. Gerlich’s data are correlational. The Microsoft study measured how much thinking people felt they did, not how well they caught errors. Yan, who has reviewed the literature, notes that the studies conflict and that AI can push learning in either direction (Nature). Rigorous, long-term trials are only now being funded; the UK’s Education Endowment Foundation committed up to £2.5 million to them in June.

What the studies agree on is the pattern of risk: low confidence, low domain knowledge and high trust in the tool.

The South African picture

Locally, the debate has moved from whether students use AI to how universities teach and assess with it in the room.

At the Universities South Africa conference last week, Higher Education Minister Buti Manamela argued that AI makes the university’s job of building critical thinking and testing evidence more important, not less. He urged institutions to make practical choices about teaching and assessment (SAnews, 8 October 2026).

South African research points both ways:

  • A UNISA study of distance-learning English students found AI could strengthen critical thinking and metacognition, but only when used reflectively and with lecturer guidance (Frontiers in Education, 2026).
  • A University of Johannesburg conference paper on Gauteng students warns that AI can mask skill gaps while deepening existing divides (UJ, 2026).
  • At the University of the Free State, lecturers report polished but vague, contextless essays, and some are returning to handwritten work and close-reading exercises (com, via The Conversation).

The local stakes are particular. Many first-year students arrive from under-resourced schools and study in a second or third language. For them, AI can feel like a lifeline. That is exactly why how it is used matters so much.

What helps

The research does not say “stop using AI”. It says protect the effortful parts of thinking on purpose.

In classrooms and lecture halls

  • Design the task, not just the rules. Yan’s work suggests AI supports learning when it gives feedback or prompts reasoning rather than ready-made answers. The catch: some of his students abandoned a thinking-prompting AI for ChatGPT (Nature). Good tools need assessments that reward the reasoning.
  • Teach thinking inside subjects. Educational psychologist Paul Kirschner argues critical thinking must be built within specific disciplines, across a coherent curriculum and over years. Willingham adds lots of practice on varied problems so students recognise a reasoning trap when it appears in real life.

At work

  • Train people to question AI, not only to prompt it. Diane Hamilton, writing on the IBM findings, argues that organisations should spend as much effort teaching staff to interrogate AI output as teaching them to use the tools (Forbes).
  • Make questioning safe. Hamilton points to fear, assumptions, technology and workplace culture as the things that shut curiosity down. A manager who asks “What surprised you in the AI’s answer? What would you challenge?” invites thinking. One who only rewards speed invites silence.
  • Protect the early-career learning curve. Junior staff build judgement by struggling, making mistakes and figuring things out. If AI absorbs all of that, they may become fluent with the tool without learning when it is wrong.

For yourself

  • Try the problem first, then compare your answer with the AI’s.
  • Ask what the output assumes, what it leaves out, and what would change your mind.
  • Notice when you accept an answer simply because it sounds confident.

The bottom line

AI is not quietly making anyone less intelligent. But it does make it very easy to skip the mental effort that builds and maintains good judgement.

The early research and the IBM numbers point to the same place. Critical thinking holds up best in people who know their field, trust their own reasoning, and work somewhere it is safe to ask “are we sure?”. Those are conditions schools, universities and employers can build deliberately. The tool can answer in seconds; deciding whether the answer is any good is still our job.

Sources

Shopping Basket
Scroll to Top