The AI Skill Nobody Is Teaching
AI has created a new problem for smart people.
You can get an intelligent answer to almost anything in seconds. Which means you can also get an intelligent justification for almost anything you already want to believe.
Keep prompting. Change the wording. Ask another model. Challenge the answer and watch it produce a better one. Eventually you arrive somewhere that feels convincing.
And because every step contained legitimate intelligence, it doesn’t feel like confirmation bias. It feels like research.
Which is why I have come to think one of the most important AI skills of the next decade has almost nothing to do with AI. It is the ability to notice what your own mind is doing while you are using it.
AI Moves the Thinking
Microsoft researchers surveyed 319 knowledge workers who use generative AI regularly and collected 936 examples of how they actually used it at work.
The more confidence people had in the AI’s ability to perform a task, the less critical-thinking effort they reported applying. Confidence in their own ability ran the other way.
Thought was relocating rather than disappearing. People spent less effort gathering information and more verifying it. Less solving the original problem, more integrating the machine’s response. Less performing the task, more overseeing what it produced.
That may be the cognitive shift underneath everything happening with AI right now. As the machine gets better at producing answers, the human becomes responsible for deciding what deserves to be believed, rejected, questioned or acted on.
Which sounds straightforward until you use these systems all day.
Tomorrow’s forecast, in your inbox.
One email. Five minutes. Written for owners, not engineers.
Ten Intelligent Answers
Say you are considering a new business.
You ask AI to evaluate it. The analysis looks promising. You ask what could go wrong and get six risks. You explain why three of them probably don’t apply, and it agrees and updates the analysis. You ask another model, which likes the opportunity too but sees it differently. So you bring that answer back to the first one and ask what it thinks.
Forty minutes later you know vastly more about the opportunity than when you started. You may also be less certain what you actually think.
Or something subtler happens. You already wanted to pursue the idea, and one of those responses finally articulates the case in exactly the way you needed to hear it.
You feel clarity.
But was it? Did new evidence change your mind, or did the machine find something you genuinely hadn’t seen? Or did you keep an infinitely patient intelligence talking until it produced the argument that made the discomfort go away?
When intelligent answers are unlimited, knowing what is happening inside your own mind becomes part of evaluating the answer.
Trust Creates Its Own Problem
A 2026 study in Scientific Reports put 295 people through a decision-making experiment involving real and AI-generated faces. Participants received advice they believed came from either a person or an AI, and the advice was deliberately correct only half the time.
Among those receiving AI guidance, more favourable attitudes toward AI went with worse performance at telling the real faces from the synthetic ones. The researchers found that reliance on AI advice could introduce bias into human judgment, particularly among people predisposed to trust it.
This gets harder as the systems improve. Bad AI trains you to be sceptical, because you see the mistakes and learn to check. A system that is right ninety-five times out of a hundred creates a very different psychological environment, where the rational response is to trust it more — which makes noticing the other five increasingly difficult.
Your Brain on AI
An MIT Media Lab experiment came at this from another direction. Fifty-four participants wrote essays across three sessions using an LLM, a search engine or nothing at all, with brain activity recorded by EEG. A smaller group returned for a fourth session and swapped conditions.
The LLM group showed weaker neural connectivity during the task than the other two. They had more difficulty recalling what they had written and reported less ownership over it.
Small preprint, one narrow task, so it establishes nothing general about AI and cognition.
The crossover session is the part worth the attention. Participants who had written unaided and were then given the AI showed stronger neural engagement than the LLM group had shown in any of its own sessions. Only eighteen people completed that stage, which makes it the thinnest result in an already thin study.
They had engaged the problem themselves first. Then they brought in the machine.
That sequence may matter, because there is a meaningful difference between bringing AI into your thinking and asking AI to begin the thinking for you.
The Skill Nobody Is Teaching
There is an old word for observing your own thinking. Metacognition.
Mindfulness trains something closely related. A thought appears and you notice it before automatically following it. An emotion appears and you notice the emotion. An impulse appears and there is enough distance to decide whether to act on it.
AI may give that ability an entirely new economic purpose. Consider noticing any of these while working:
I really want this answer to be true.
I’m asking the same question again.
That sounded convincing, but I haven’t seen any evidence.
I’m no longer exploring this. I’m trying to make myself feel certain.
I rejected three answers and accepted the first one that agreed with me.
I’ve spent another hour analysing something that can only be answered by trying it.
Those moments are easy to miss, because AI rarely feels like something influencing your cognition. It feels like something helping you think. Often it is. The difficulty is recognising when that changes.
The Machine Can Push Back
Research presented at CHI 2026 tested conversational AI designed to prompt people to examine their own reasoning during decision-making rather than simply supplying conclusions. The approach reduced over-reliance on AI compared with conventional assistance, though it demanded more mental effort to use.
Which points somewhere the industry is not currently heading. The best systems may not always make thinking easier. Sometimes they might deliberately make it harder.
Instead of answering immediately, an AI might recognise that you have asked variations of the same question twelve times. It might tell you that no new evidence has entered the conversation. It might notice that your latest prompt contains the conclusion you want it to reach. It might refuse to generate another strategy and tell you the next useful information has to come from the world rather than from it.
That would make it something more interesting than an answer machine. It would make it a mirror for how you think.
The Forecast
Metacognition becomes a core AI skill.
Today we teach prompting, agents, automation and workflows. Those matter, and the machines will increasingly handle them without us.
The harder skill belongs to the person on the other side of the interface. Can you tell when AI expanded your thinking, and when it replaced thinking you needed to do yourself? Can you recognise the difference between searching for evidence and searching for reassurance? Can you stay uncertain after an extraordinarily intelligent machine has offered to make the uncertainty disappear?
Those questions get harder as the systems become more persuasive, more personalised and more familiar with us.
Within five years, your ability to use AI may matter less than your ability to notice how using it is changing you.
Which is why I think mindfulness is about to acquire a role almost nobody associated with meditation ten years ago. It may become part of how people preserve judgment while surrounded by intelligence far greater than anything we have had access to before.
We spent the first years of this learning how to prompt the machine. The next discipline may be learning to observe the person doing the prompting.
Sources: Lee HP, Sarkar A, Tankelevitch L, et al. The Impact of Generative AI on Critical Thinking. CHI 2025 · Kosmyna N, et al. Your Brain on ChatGPT. MIT Media Lab, 2025 (preprint) · Pearson J, et al. Examining human reliance on artificial intelligence in decision making. Scientific Reports, 2026 · Understanding the Effects of AI-Assisted Critical Thinking on Human-AI Decision Making. CHI 2026