The more your team trusts the AI, the less they question it. Why that is the real risk.

The AI's answer went up on the screen, and the room nodded. It was fast, well-phrased, and it sounded right. Nobody pushed back. And a beat too late you realised that you had not pushed back either, and you were not sure anyone had actually checked it. The room had quietly outsourced its scepticism, and no one had decided to. That small moment is the whole issue in miniature.

Here is the reframe. The danger with AI at work is not mainly that people forget how to do things. It is that they stop questioning what the tool hands them. The more they trust it, the less they look. And the moment that matters most, the moment the tool is confidently wrong, is exactly the moment nobody is looking.

Does relying on AI actually reduce critical thinking?

The evidence says it can, and it names the mechanism. Researchers at Microsoft Research and Carnegie Mellon University surveyed 319 knowledge workers about 936 real tasks where they used AI. The pattern was clear. The more confidence a person had in the AI, the less critical thinking they applied to its output. The more confidence they had in their own skill, the more they checked and refined it. Trust in the tool lowered scrutiny; trust in themselves raised it.

The study found more. Across most kinds of thinking, from analysis to evaluation, people reported that the AI reduced the effort they put in. And groups using AI produced a less varied set of answers to the same task. So the shift is subtle. Not a loss of ability, but a drop in the questioning that ability depends on. It is a close cousin of the trust problem at the centre of every AI decision: the smoother the answer, the less we interrogate it.

Why is over-trust the real risk, not lost skills?

Because a wrong answer nobody checks is far more expensive than a skill nobody practises. When the AI is right, uncritical acceptance costs you nothing, which is exactly why the habit forms. It is rewarded, again and again, until it is automatic. Then comes the case the tool gets confidently wrong, delivered in the same fluent, assured voice as every correct answer before it. Fluency is not accuracy, but it reads like it, and an unquestioned mistake travels straight into a decision.

The lower-altitude question here is which AI tool is most accurate, or which vendor hallucinates least. Useful, but not the point. Every tool will be confidently wrong sometimes. The higher question is whether your people are still thinking when it happens. That is a question about the culture around the tool, not the tool itself, and it is the same reason the human capability around AI is what you actually have to build.

So how do you keep judgement switched on?

You use AI in a way that invites scrutiny rather than sedates it. The goal is a team that leans on the tool and argues with it, both at once. Build it in this order:

1. Make checking the norm, not the exception. Treat an AI output as a draft to be tested, never a verdict to be accepted. Say so, and model it yourself.
2. Ask for the reasoning, not just the answer. A team that has to explain why an AI answer holds stays engaged with it. An answer taken on trust switches thinking off.
3. Protect dissent. Make it safe and expected for someone to say the AI is wrong. The quiet room is the dangerous one.
4. Match scrutiny to stakes. Let low-risk work flow. On the decisions that carry weight, require a human to interrogate the output before it counts.
5. Measure thinking, not just speed. Watch whether AI is sharpening your team's judgement or dulling it, and treat that as a real metric.

What does this make possible?

A team that is faster and sharper, not faster and softer. AI carries the load on the routine work, and your people spend the freed attention interrogating the answers that matter, testing them, improving them, catching the confident mistakes before they cost anything. The tool does not replace their judgement. It gives them more room to use it. That is a stronger operation than one that adopts AI and quietly stops thinking.

Picture the same meeting a year on. The AI's answer goes up on the screen, and this time someone leans in and asks how it got there. Another spots the flaw. The tool made them faster; the culture kept them sharp. That is the team you want when the stakes are real: quick to use the machine, quicker still to question it. Speed and judgement are not a trade-off here. Built with care, they compound.

About the author

Thomas Green

British technology futurist, AI keynote speaker and advisor. Thirty years across enterprise technology and AI strategy, helping leaders navigate the future of work. The futurist who died.