Kristjan Vingel

How to Prevent AI Brain rot (according to scientists)

By the end of this article, you will know exactly how to protect your brain from erosion caused by AI usage. So let the AI bro’s do what the sheep do. We’ll be much smarter about this.

I read this paper called “Large-Language Models as a Cognitive Virus.” Its main premise is this: LLM usage spreads like a virus. The consequence is that once a tipping point is reached, a whole bunch of people can become rapidly dependent on AI, with “abrupt losses in cognitive competence.”

But fret not, the authors also give actionable advice on how to avoid this brain rot. This is what interests me the most, and so this is also the main topic of my article. I’ll leave the math aside today.

Before we go, a remark: This is not a summary of the article – If you want “everything”, I recommend reading it yourself. And, before you turn to Chad Gippity with “Summarize this for me” – be aware it’s actually not a good idea… But I am getting ahead of myself (more on this later).

Nuff said. “DJ, spin that shit!”

If you’re massively overweight, getting back into shape takes more effort than the effort of staying healthy from the start. Undoing is harder than prevention. That’s one thing. The other thing is that it’s harder to stay healthy when everyone around you becomes fat. Once more and more people become fat, there’s a tipping point with a sudden drop – people become fat at an unprecedented rate. Because with each slip, the support weakens – which means even more people slip. And to go back to a point where most are healthy again takes more effort than it took to go from fit -> fat.

Ever been to a concert or a theatre play with a standing ovation at the end of the performance? It starts with one person standing up, then a few more until a tipping point is reached and the remaining people will stand out almost at once – not because they want to but because of social pressure.

By the way, one or two times I deliberately did not stand up just to get a sense of what it feels like to be the only one sitting. It’s an interesting feeling and not that easy to go against what the group is doing. But it’s good practice to feel the weight of that pressure and to be able to withstand it.

The moral is: it’s a wise choice to protect your thinking muscle from becoming obese in the first place. Keep it fit.

Here’s how we can do it.

1. Solve Your Problems

Solving problems is hard work. We have to use our brains to do this “thinking.” “But we can just ask Claude to do it for us!?” Well, that’s like not walking at all and just taking Bolt or the electric scooter everywhere.

Don’t avoid cognitive friction. Do your thinking. Read textbooks, check Wikipedia, take notes, brainstorm, sit down with a pencil and piece of paper and “do the work.”

For example, there was a period in my life where I got up every morning before 5 AM and studied mathematics. I did proofs and trig and all that. I want to get back to it – I miss that time! But AI bros are like, “bruh, AI can do math so well…”

2. Verify

It’s easier to check AI’s output when you have a clue about the topic. The bias is for us to accept whatever AI spews at us as holy gospel when we’re a novice at that particular thing. It’s the Gell-Mann amnesia effect or the automation bias and the related Dunning–Kruger effect.

Recently I read a really sad story about a university student who trusted the AI’s recommended medicine dosage and died because of that.

Let’s try to keep that bias in mind.

3. Deliberately Disengage

The authors recommend having deliberate periods of disengagement from AI.

A day a week? A week every few months?

Sounds good to me.

Bill Gates used to do “Think Weeks” where he’d go into some cabin in the woods with a bunch of books and “think.” No phones, no computers. Apparently several of the famous Microsoft products came out of those sessions.

I predict those sorts of retreats and detox periods become even more prevalent and necessary as time passes.

4. Maintain Your Non-AI Skills

Write, draw, make something with your hands. Not everything needs to be done with the help of AI.

Btw, (good) writing may be one of the most resistant areas to AI. AI can, of course, generate flat, soulless text, but it’s not capable of producing prose like an actual writer with a voice.

5. Think First

Think first, ChatGPT later. Remember the beginning of the article where I said that it’s not a good idea to right away summarize the paper?

In this study, the group that did their own thinking first and then used Gippity did better than the “all-in AI” group and the “no-AI at all” group. However, this is just one study that showed a short-term effect, not the long-term consequences (which we are much more interested in). Plus, you can probably find many more shortcomings and holes if you want.

Regardless, this is basically how I use AI on my software projects (I described it in more detail in my previous article): I solve the problem myself first, then Claude will act as a personal Berkeley-level professor and check my work, and I’ll see if I missed something or if I can improve upon it. Sometimes I agree with it. Sometimes I disagree and say: “Nah, this abstraction is not worth it,” and leave it at that.

I don’t use this method for writing my articles, though. In fact, I don’t use any AI to “help” with my postings.

What to do If You’re an Addict Already?

If you’re too deep in the weeds, these actions may not be enough. You might need psychological intervention such as therapy.

Seriously.