An MIT study found that using AI leaves us more ignorant, yet more certain of our own arguments, reducing our capacity to think even as it leaves us more convinced we are well informed. Here Minh-Hoang Nguyen argues this is a new kind of brain rot: a literal degradation of our ability to think, and a new form of intellectual cowardice. Only by allowing ourselves room to be stupid can we reclaim the courage to think for ourselves.
A new MIT study found that AI is rotting our brains. After participants were asked to write an essay assisted by ChatGPT, not only did the quality drop compared with those using search engines to help inform their thinking, but afterwards there was a noticeable disconnection from their ability to think without the technology. Now, with trillions of dollars invested, the participation of humanity’s greatest minds, and the construction of more than 12,000 data centers, AI is rapidly becoming smarter. Its applications are increasingly versatile, its responses more accurate and aligned with human requests. Especially when we face the challenges of coming to terms with AI being able to make, or at least assist with, landmark mathematics, its performance is gradually surpassing human intelligence on specific cognitive benchmarks.
We are increasingly captivated by the uses of AI. Even the skeptics find it difficult to resist because they face growing pressure to adopt AI simply to keep up with the times. The threat to students, professionals and even academics is that they may be caught out as their peers produce “better” or at least faster work utilizing this technology.
While the AI era may be blossoming, it also marks the acceleration of the brain-rot era: an age in which people can appear intelligent with just a few prompts, a subscription account, and enough electricity to power a browser. We can see, as Deleuze warned, that our society is devolving into a society of control in which people’s capacity to think is modulated and delivered as a service.
And yet the very systems which we are using to generate this artificial intelligence come with a warning. When working with AI, have you ever noticed the statement “AI can make mistakes. Please double-check the information”? Some version of this warning appears beneath almost every AI model, whether it comes from OpenAI’s ChatGPT, Anthropic’s Claude, Google’s Gemini, or Meta’s Llama.
___
Brain rot is distinct from simply not knowing; it’s tied up with our inability to think, the actual rotting away of our mental capacities.
___
The necessity of such a warning is a sign that we are being caught in a trap. Humans increasingly believe in the power of AI’s extraordinary capacity to provide us with a perfect answer. We have stopped checking whether those answers are right or wrong. Although we continue telling one another that humans are intelligent, deep down, we should be terrified by AI’s capacity to learn and regurgitate for us. This attitude of acceptance has begun to paralyze our thinking processes. The ones that should have started as soon as we received the response!
Thus arises the huge absurdity: AI needs to teach humans that AI can be wrong!
Want to continue reading?
Get unlimited access to insights from the world's leading thinkers.
Browse our subscription plans and subscribe to read more.
Already a subscriber? Log in
Love the iai?
Sign up to get exclusive access.
Related Posts:
Related Videos:
7th October 2026
Join the conversation
Brian Balke 7 October 2026
I think that I have finally figured out how to understand AI.
An LLM is massive database defined by tokens (the data) and the weighting of their tendency to appear together.
This is important, because the paradigm in database queries is not procedural. A procedural language (like C++ or Java) defines the steps to be taken in transforming input into output. It does not presuppose anything about the input or output. It can generate answers that can never be realized.
Database queries are instead declarative. They define a criterion for selection from the dataset. "Give me the names and addresses of all the mortgages in arrears."
Now, in the past, you would have had to master a highly technical language to get that data (most commonly, a language called 'SQL'). What happens in an LLM, however, is that the query engine simply maps the words in your query onto the affinities stored in the database. In effect, it allows you to create a custom query language.
The output is generated by extending the words in your query with those most likely to follow. The system actually has no intelligence. It is simply spooling out words without any understanding of their meaning. This is just with a traditional database. The query engine has no idea that the query is going to be used to initiate foreclosure proceedings.
What does this mean for human intelligence? Well, intelligence involves discovery of correlations in our experience and the discovery of behaviors that allow us to benefit from those coincidences. Because we need to start afresh when we interrogate an LLM to solve a new problem, there is no regularity, no correlation, no sense of cause-and-effect, and therefore no foundation for learning. To depend upon an AI engine is to disorganize the mind's capacity to learn.