Why AI doesn't replace human intuition in decisions
By Aleks Mosaz ·
What makes AI different from our brains
I'll say something that is probably controversial and obvious at the same time: the main problem with AI is that it "doesn't know how to think."
Yes, it's excellent at analysing, forecasting, assembling a solution from an enormous set of inputs, writing code and text, designing interfaces, finding patterns and doing a thousand other things we long considered exclusively human. Just a couple of years ago, we were laughing at "pig wings" and other AI glitches. We still are.
It's not even about the mistakes
AI makes mistakes and hallucinates all the time. It can calculate the wrong thing, make a bad decision, lose a piece of context and then, with a completely straight face, explain why you're the one who misunderstood everything. GPT's favourite kind of gaslighting.
But there's something else.
AI has no reason of its own to consciously choose a solution that it considers worse. Read that sentence again and imagine two business scenarios.
The first is straightforward.
It follows best practices. There are numbers, data and case studies. Similar companies have done it and made money. There's a model in which stepping to either side is practically a firing offence. Everything is logical.
The second is some completely unhinged nonsense dreamed up with a fever of 37.5°C.
Nobody does it that way. There are no data. Colleagues think you've lost your mind. The presentation makes it look as though the whole team has spent three weeks taking something illegal. Someone gets fired right after presenting it. Someone else has their competence questioned but is given a chance to try.
And sometimes a person says: "Actually, that's an idea! Let's do the second one! We need to test it."
Not because that option is objectively better. Often it's worse than the transparent, predictable one. But because: "I feel like there's something there."
Behind that feeling and that gut instinct is a wild mixture of everything that makes a particular person who they are:
Experience, associations, intuitions, tastes, ambitions, fears, childhood traumas, upbringing, people they've met, films they remember from when they were fourteen, their first job, their first failure, the desire to prove something to someone. Add as many ingredients to that cocktail as you like.
And a simple internal feeling: "Let's just bloody try it."
Perhaps that's why people are so inconveniently different. Put ten smart people in one room, give them the same data, and you can get ten different decisions and arguments until morning, ending in a few people telling each other where to go.
Nine of them may be absolutely certain that the tenth is proposing utter rubbish. Sometimes the nine will be right. But sometimes the tenth will build a multimillion-dollar company on that decision. There are plenty of examples.
A person can stake money, a career, months of their team's work and their own reputation on a hypothesis.
A model has no internal "I have a feeling; I want to test this."
It can be excellent at helping you test it, suggesting ten ways to implement it, finding weaknesses or playing devil's advocate. It can do in five minutes what used to take a week.
But the original "What if we deliberately do this the wrong way?" still more often comes from somewhere else.
AI's cultural averaging
I think this is one reason there's so much AI slop around. You see it every day: people seem different, but everything they produce in social media and products looks the same, as if copied from one template.
Look at modern SaaS products, landing pages, presentations, content carousels, ad creatives and dashboards. Most look like copies of one another.
It's not because AI is stupid. Quite the opposite: it's trying to do the best possible job.
Its training already reflects an enormous number of existing human decisions. Common, understandable patterns that have proved effective have a natural advantage.
So when you ask it to "make something cool," it quite logically gives you something statistically similar to what humanity has already agreed is cool.
Now millions of people send that request every day.
The result is a curious circular procession. People ask AI to make things better at scale. AI reproduces the characteristics of what is already considered good at scale. Those decisions fill the internet. More and more new decisions are then made in a world already filled with AI's previous decisions.
Welcome to cultural averaging.
Where autopilot ends
I think the value of a person working alongside AI will increasingly be neither pressing buttons faster, a contest we've already lost, nor remembering more information, another pointless competition we'll lose.
It will be the ability to pull AI towards the place where its autopilot ends.
To say: "Yes, a round wheel is objectively better. Today we're making a square one."
Not because square wheels are better. In 99 cases out of 100, it will be a complete failure; let's agree on that.
But in a sense, all human progress consists of people who occasionally agreed to look like idiots for the sake of that hundredth case.
AI is excellent at making round wheels. Perhaps soon it will make the best round wheels we can imagine.
Our job is to look at them and say: "Very good. Amazing. Now make them square."
Not because it's reasonable, but because we have one crazy hypothesis that AI can't "reasonably" understand.