Christian Heilmann

AI is Dunning-Kruger as a service

Thursday, October 30th, 2025 at 8:52 pm

On January 6th, 1995 two bank robbers in Pittsburgh confused law enforcement by not making any attempts to conceal their faces but instead brazenly looking at security cameras as if they were invisible. The reason is that they actually thought they were.

Clifton Earl Johnson had convinced his fellow in crime, McArthur Wheeler that covering their faces in lime juice would make them invisible to cameras. Much like lime juice can be “invisible ink” until you heat the paper. As a test, Johnson had taken a polaroid of Wheeler that showed his face smudged. That a camera fault might be the cause, or doing a second test didn’t get to their mind.

This baffling over-confidence in their flawed approach inspired two psychologist, Justin Kruger and David Dunning to see if there is a common bias in people when it comes to assessing their skills and their actual performance in doing them. They found out that there is such a thing and it is now known as the Dunning-Kruger Effect.

A cognitive bias, where people with little expertise or ability assume they have superior expertise or ability. This overestimation occurs as a result of the fact that they don’t have enough knowledge to know they don’t have enough knowledge.

One could say that the Dunning-Kruger effect is the opposite of Impostor Syndrome. Instead of people not being able to interiorise their obvious successes, people declare themselves as great and experts at things they have no or just a rudimentary clue about.

Over the last few years we’ve been on a constant path to make this the standard mindset in the technology world. It started with a demand for everything to be released incredibly fast and to be a huge success in numbers from day one. Anything not growing exponentially is not a success.

Fakers instead of makers

“Fake it till you make it” is given as advice devoid of any irony. Instead, deception and inflation of numbers is seen as a smart move until you have the resources and knowledge to properly do the task. KPIs and OKRs are meant not to reflect delivery goals but aspirations. When you’re not gunning for a promotion every half year you’re not seen as a go-getter or having a growth mindset. In other words, we encourage bragadocious behaviour and language. Some of the things you hear from heads of states and other politicians in interviews sound like Muhammad Ali at press conferences before a fight in the 60s or old school rappers in the 70s and 80s.

AI bots excel at faking knowledge

But even worse, any interaction I have with AI chatbots gives me the same vibes. They give utter nonsense answers with high confidence and wrap errors in sycophantic language making me feel good for pointing out that they wasted my time. A correct answer is a lot less important than a good sounding one, a positive one or one that makes me interact more with the system. Time in product is the goal, not helping me find the right answer.

GenAI makes you a genius without any effort

The siren song of generative AI to turn anyone into an artist, wordsmith, composer or videographer by using “intelligent” tools is a push into Dunning Kruger territory. Vibe coding or vibe anything really focuses not on the craft, but the result. We’re not meant to create by learning the ropes and understanding the art. We’re much too clever and busy for that. Give it a prompt and create a product, an app or an agent that does your bidding. We’re continuously reminded that we all are capable of genius – if only we let the machines do the boring work for us. Our egos are fed, we are barraged by digital cheerleaders and confidence tricksters.

Stop wasting time learning the craft

Adding human effort into things, really creating and writing yourself is taunted as wasting your time and not embracing change and progress. But the cost is that we forget about the craft and we lose the joy of creating. Creativity of any kind is messy and fraught with error and drawbacks. But all of these make us human and who we are. As Leonard Cohen put it: “There is a crack in everything, that’s how the light gets in”. Sure, you might not be good at painting, composing, writing or shooting movies. But a terrible, human effort still is worth so much more than asking the machine to build you a boring solution focused on crowd pleasing more than being a thing in itself.

I am not happy about this, and I don’t see it as progress. If anything, it is deception and watering down craft and art. Politics have become an attack on intelligence, decency and research in favour of fairy tales of going back to “great values” of “the past when things were better”. Social media has become devoid of the social part and is a numbers game and addiction machine. But you know what? I don’t care. I keep doing what I do. I write down things I consider important at that time. I paint things although I suck at it. I publish on the web and my own blog because nobody stops me. Sure, I feel like a fraud when people applaud what I do more often that not. And yet – the joy of creation is something we should never give up on. Do you feel like what you do isn’t good enough or worth while? It is, and even if what you did isn’t amazing quality, you’ve created it and it is yours. And maybe, just maybe you are not the best judge to assess the quality of what you did anyways. One person’s disappointment may well be a joy to others. Keep creating and keep striving to improve and if others impressed you, tell them about it.

Share on Mastodon (needs instance)

Share on BlueSky

Newsletter

Check out the Dev Digest Newsletter I write every week for WeAreDevelopers. Latest issues:

160: Graphs and RAGs explained and VS Code extension hacks Graphs and RAG explained, how AI is reshaping UI and work, how to efficiently use Cursor, VS Code extensions security issues.
159: AI pipelines, 10x faster TypeScript, How to interview How to use LLMs to help you write code and how much electricity does that use? Is your API secure? 10x faster TypeScript thanks to Go!
158: 🕹️ Super Mario AI 🔑 API keys in LLMs 🤙🏾 Vibe Coding Why is AI playing Super Mario? How is hallucinating the least of our worries and what are rules for developing Safety Critical Code?
157: CUDA in Python, Gemini Code Assist and back-dooring LLMs We met with a CUDA expert from NVIDIA about the future of hardware, we look at how AI fails and how to play pong on 140 browser tabs.
156: Enterprise dead, all about Bluesky and React moves on! Learn about Bluesky as a platform, how to build a React App and how to speed up SQL. And play an impossible game in the browser.

My other work: