First published on LinkedIn on 26 September 2025, and expanded for this site in October 2026. Read the original LinkedIn post.
In September 2025 I went to a BCS Edinburgh Branch event to hear Daljit Rehal, then President of BCS, The Chartered Institute for IT, speak about digital transformation in the age of AI, and to chat with him afterwards. As well as his BCS role, he is Director General and Chief Digital and Information Officer at HM Revenue & Customs, and Computer Weekly’s UKtech50 2025 named him the second most influential person in UK technology.
One question from that evening has stayed with me. Are we effectively outsourcing our thinking to AI and simply accepting what it tells us? And if we are, how would we know whether its answers are right?
What the research says
A peer-reviewed study published in January 2025 suggests the concern is real. Michael Gerlich surveyed 666 people in the UK, and interviewed 50 of them, about their use of AI tools and their critical thinking (Gerlich, 2025). He tested two hypotheses and found support for both:
- Higher use of AI tools is associated with weaker critical thinking. People who used AI tools more scored consistently lower on critical thinking assessments.
- Cognitive offloading explains much of that link. The more mental effort people handed over to AI, the weaker their critical thinking.
Younger participants, aged 17 to 25, relied on AI tools most and scored lowest.
It’s worth applying critical thinking to the study itself. It measures association, not cause, and it relies on self-reported data, which the author acknowledges; he calls for experiments to establish causation. It doesn’t prove that AI makes people worse thinkers. But it’s consistent with what many of us see: when an answer arrives instantly and sounds confident, the temptation to accept it is strong.
Why this is an organisational risk
For an individual, losing the habit of questioning is a personal cost. For an organisation, it’s a decision risk. AI is now part of how strategies are drafted, options compared and evidence summarised. If the people reviewing that work have stopped challenging it, errors pass straight into decisions, carried by a confident tone that makes them harder to spot.
This is automation bias: the tendency to trust an automated answer over our own judgement, even when there’s evidence it’s wrong. It isn’t new, but generative AI makes it much easier to fall into, because its answers are fluent, fast and always available.
It isn’t only me saying this
At Davos in January 2024, IBM Chairman and CEO Arvind Krishna put it plainly: “If the lower half of cognitive work gets taken over by genAI, it implies that you’ve got to learn critical thinking.” Critical thinking, he said, becomes “the skill that is far, far more needed” (World Economic Forum).
Ron Carucci, writing in Forbes the following month, made the same point from a leadership angle: “No clever prompt we type into an AI tool will ever be able to replace human critical thinking.”
And in January 2025, Dr Philippa Hardman’s summary of recent research on generative AI and learning concluded that general-purpose AI tools, used without careful design, can encourage over-reliance and weaken critical thinking.
What leaders can do about it
The answer isn’t to stop using AI. It’s to make challenge a deliberate part of how AI-supported work is reviewed:
- Form your own view first. Write down your own answer before asking the AI, so its answer is something to compare against, not a starting point.
- Ask the AI to argue against itself. Ask for the strongest case against its recommendation, and for the evidence that would prove it wrong.
- Check the sources, not just the summary. If a claim matters, find where it comes from.
- Run a pre-mortem on important decisions. Imagine the decision has failed and ask why. AI-supplied assumptions are often where the answer lies.
- Decide where AI is not the final word. Agree in advance which decisions need a named person to own the reasoning, not just the outcome.
These are habits, and habits can be taught. That’s the heart of the critical thinking training and decision support I deliver: practical, repeatable ways for teams to question a plan, including one produced with AI, before they commit to it. Careful checking pays off. In my own work, questioning a source most people take on trust led to a correction to how Thinking, Fast and Slow summarised Professor Shane Frederick’s research.
How are you managing this risk in your organisation? If your team needs to test AI-assisted analysis before acting on it, explore AI-assisted Decision Support.
Get in touch now: Contact me