Human Intelligence in the Age of AI: The Skills Machines Can't Replace

Ask an AI model to plan a workshop and within seconds you will have an agenda, a few icebreakers and a neat closing ritual. It will be good. Annoyingly good.

But is it answering the question you actually meant to ask?

This is why I keep coming back to the idea at the heart of FOG OFF. Clarity begins before the tool, with a question you have actually wrestled with. A simple daily prompt helps: write the question you are about to ask AI, then rewrite it once as if you were asking a wise friend.

That is the part nobody puts in the product demo. The machine is brilliant at answers. It has no idea whether you asked the right thing.

What skills can AI not replace?

Let me start with the data, because the conversation about "future skills" is usually all vibes and no numbers.

The World Economic Forum's Future of Jobs Report 2025 surveyed more than a thousand large employers worldwide. Their estimate is that 39% of workers' core skills will change by 2030. Analytical thinking is still the skill employers rank as most essential. Right behind it sit resilience, flexibility and agility, then leadership and social influence. Among the skills expected to grow fastest, next to AI and big data, you find creative thinking, curiosity and lifelong learning.

Read that list again slowly. Half of it is technology. The other half is deeply, stubbornly human.

Curiosity. Resilience. Creative thinking. Knowing how to influence a room. None of these are new. What is new is that they used to be "soft skills", the nice-to-have section of a CV. Now they are the part of you that a model cannot quietly do on your behalf.

Is AI making us better or worse at thinking?

Both. That's the honest answer, and the research is quite precise about it.

In 2023, a team led by Fabrizio Dell'Acqua at Harvard Business School ran an experiment with 758 consultants at Boston Consulting Group. On tasks that sat inside what the authors called AI's "jagged technological frontier", consultants using GPT-4 completed about 12% more tasks, worked roughly 25% faster and produced work rated around 40% higher in quality.

Then came the twist. On a task deliberately designed to sit outside that frontier, consultants using AI were 19 percentage points less likely to reach the correct answer than those working without it.

Put simply, the frontier is jagged, and you often don't know where it is until you fall off it. That is my reading of the lesson, not the authors' words. The tool was so fluent that smart people stopped checking whether the problem was the kind of problem the tool could solve.

A year later, Michael Gerlich at SBS Swiss Business School published a study in the journal Societies (2025) with 666 participants. He found a significant negative correlation between frequent AI tool use and critical thinking scores, and the link ran through what psychologists call cognitive offloading: handing the thinking to the tool. Younger participants showed the strongest effect. It is a correlational study, so it doesn't prove AI makes us dumber. It does suggest that the habit of outsourcing is where the risk lives.

The danger is not that machines think. The danger is that we stop.

Why asking the right question is the real human advantage

There is a line often attributed to Picasso: "Computers are useless. They can only give you answers." Quote Investigator traces it to a 1964 interview by William Fifield, so the attribution is plausible, if not airtight. Sixty years later, it reads less like a joke and more like a job description.

Here is how I put it to my clients:

The machine gives answers. The human chooses the question.

A model optimises inside the frame you hand it. It does not ask whether the frame is wrong. It will not stop you in a meeting and say, "Wait, why are we solving this at all?" It will happily write the perfect email to the wrong person.

Choosing the question requires things a model doesn't have. A body that gets tired. A history. Stakes. Someone at home you'll have to explain the decision to. Taste, which is really just memory with an opinion.

Rilke wrote to a young poet in 1903 that he should try to love the questions themselves, "like locked rooms." I think about that line more now than I did at twenty. Our culture rewards speed to answer. The people I see thriving right now are the ones who can stay a little longer inside the question before they let anyone, human or machine, close it.

The Unopposed Mind: what happens when no one disagrees with you

There's a pattern I've started calling the Unopposed Mind. It is a brilliant mind with no one left to disagree with it.

It used to happen to senior executives. You rise high enough, people stop pushing back, and your blind spots get a corner office.

Now it can happen to anyone with a chat window.

Large language models are, by design, agreeable. Researchers even have a name for it: sycophancy, the tendency of models to align their answers with what the user seems to believe. Ask a leading question and you often get a confident, well-formatted echo of yourself.

"The first principle is that you must not fool yourself, and you are the easiest person to fool." Richard Feynman, Caltech commencement address, 1974

Feynman said that long before chatbots. It has never been more practical advice.

A future skill nobody lists in the reports: building your own opposition. Deliberately seeking the person, the book, the data point that will tell you you're wrong. And asking the model to argue against you rather than for you.

Human skills in the age of AI: the short list I actually believe in

I'm wary of lists. Most "top ten future skills" articles are the same ten nouns in a different font. So here is mine, and why each one resists automation.

Judgement under ambiguity. A model gives you probabilities. Someone still has to decide when "probably fine" is not good enough. This is the skill the BCG consultants lost on the outside-frontier task.

Question design. Framing the problem. Reframing it when the first frame fails. This is where my engineering training at ITU and my Brain and Cognitive Sciences studies at MIT keep meeting each other: the best solution is useless if it answers the wrong question.

Connective thinking. Seeing that a problem in your sales team looks like a problem in a jazz quartet, or in a beehive. This is the heart of what I call The Polymath Way™. Models can retrieve analogies when prompted. Noticing that you need one, in the middle of a Tuesday, is still ours.

Emotional and relational intelligence. Trust is built in moments of risk, and a machine has nothing to risk. People follow people who have skin in the game.

Resilience and self-regulation. The WEF list puts it near the top for a reason. When everything changes every eighteen months, the ability to stay steady, curious and kind to yourself in the middle of change is the operating system for every other skill.

Notice that none of these are anti-technology. The people who will do best are not the ones who avoid AI. They're the ones who use it without handing over the steering wheel.

How to train the skills AI can't replace

Not a ten-step plan. A few small experiments you can run this week.

Write the question before you open the tool. On paper, in one sentence. Then ask yourself, "If I got a perfect answer to this, would my problem actually be solved?" If the answer is no, the question needs work, not the prompt.

Ask the model to disagree with you. "Give me the three strongest arguments that my plan is wrong." Read them with genuine openness. Notice which one irritates you most; that's usually the one worth sitting with.

Do one hard thing slowly. Once a week, solve something without assistance. A draft, a calculation, a decision memo. Not out of nostalgia. Out of the same logic that makes you take the stairs.

Collect a friendly opposition. Name two people who reliably see the world differently from you. When did you last ask them what they think?

Borrow from far away. Before your next big decision, ask yourself how a chef, a nurse or a chess player would approach it. Write down one idea that didn't come from your own field.

Borrowing from far away is basically the whole philosophy of Creative Solutions for Ordinary Problems. Creativity is a muscle, and cross-field thinking is one of its best workouts. One of the experiments I love is picking a random profession each week and solving one of your work problems the way they would.

So, what are the future skills, really?

The Future of Jobs Report will be revised again. The tools will change before this post is a year old. Some of the specific skills on today's lists will look quaint.

But the shape of the answer probably won't change much. As machines get better at producing answers, the scarce resource shifts upstream. Toward attention. Toward judgement. Toward the quiet courage of asking, "Is this the right question?" in a room where everyone already has an answer ready.

That is not a lesser kind of intelligence. It may be the oldest kind we have.

So I'll leave you with one question, and I'd love you to resist the urge to ask a machine:

What is the question you've been avoiding, because you already suspect the answer?

With love,

Gizem

If this raised a question you'd like to think through, we can look at it together. Book a free 30-minute Discovery Call.