AI confidence and the Dunning-Kruger effect

I love models. Models about how the world works, or ways of thinking about the world. Economic studies are essentially that, a framework for thinking being my favourite 3 words to describe the discipline. Here’s a model of how things work, and here’s what happens when x changes ceteris paribus - all else held the same. They are not how the world actually works, because the world is complex, but by compartmentalising concepts, powerful simple forces can be predicted.

Today I write about one particular model, the Dunning-Kruger effect. I see an overzealousness about AI from people who… don’t actually use AI, but spend a lot of time reading about AI, and facing strong external incentives to push for AI. What C-level leader wants to present themselves as old-school, of course they’re full steam ahead on AI, this is the future after all!

How the Dunning-Kruger effect is commonly represented

For my own story of this effect I will use the youthful confidence that comes from being a teenager and learning maths. I did calculus and statistics at high school, passing fine enough, all the concepts made sense. At university I also did well in my first year, maths was a part of almost every paper and any maths thrown at me was easy peasy, lemon squeezy.

Then came 2nd year, and advanced mathematics in a specific paper.

Matrices were being taught to me for the first time. An entire field of mathematics that I have no idea about, and couldn’t make intuitive sense of was very exposing. Me, someone that knows maths? No way, not any more. Matrices are only one tiny branch of many in maths too! The peak of confidence was smashed and never to be recovered. I fell, hard and fast right down the curve somewhere towards the bottom. I passed the course, but I know enough now, to know that I really don’t know maths at all!

Now for some fun irony. The Dunning-Kruger effect is not even the above chart at all. That’s a made up representation, great for story telling but not the real one.

I guarantee you there will be many, who think that’s the “Dunning-Kruger chart”, with full confidence. The real graph is more boring and is below. Quite simply in studies, it shows the relationship between perceived performance and actual performance. Lower competence people rate themselves on average to perform better than they actually do, with only high competence people perceiving themselves to have done worse than they actually did on average.

In laymans terms, incompetent people are often confident, and competent people are often too humble about their abilities. — In layman's terms, incompetent people are often confident, and competent people are often too humble about their abilities.

The real data does tell a related story to the common representation of the Dunning-Kruger effect, which takes that knowledge and applies a suggested twist of time. “As you learn more you’ll feel less confident” for example is not something you’d directly garner from the actual data plot.

And now for my social commentary part

I have played with AI tools a lot, I have built software, automations and use it daily in my work. I am somewhere down the valley with regards to my competence, and confidence.

I worry deeply in the current climate about leaders who don’t do, but just read about what AI can do. AI can do wonderful things, but as of today (May 2026), it’s still not quite there, and we don’t even know to what extent it’ll get there.

The enthusiasm to implement AI everywhere, and belief that it’ll be able to gain all the wonderful efficiencies without too many difficulties I believe is misguided from people who don’t sit down and wrangle with the systems day-in-day-out. I don’t know many builders who would be happy letting AI run wild in production. Humans for now must be in the loop somewhere.

I believe C-level leaders are at peak confidence right now. I am extremely curious to know where they are in 6-18 months, when they know more.