Are LLMs Rotting Your Brain? Study Claims AI Acts Like A 'Cognitive Virus'

A multinational team of researchers comprising mathematicians and data scientists from Spain, Italy, and the US have just published a paper called "Large-Language Models as a Cognitive Virus" that is making some waves in the AI community. The premise of the paper is fairly complex, but the short version is that the researchers believe that using chatbots to do your thinking for you doesn't just hurt your own thinking ability, but it also may be self-reinforcing if others around you are doing the same.

The paper is dense and full of advanced math, because it's not an experimental study. Instead, the scientists collated existing research on the mental effects of AI dependence and, by applying epidemiological math (like that used to track biological viruses), came up with the conclusion that a society might transit between three states over time: uncoupled, coupled, and dependent. Uncoupled users don't rely on LLMs for thinking, Coupled users use LLMs as part of a cognitive process, and Dependent users have offloaded most of their thinking to AI.


So the argument is less that "AI rots your brain," because that's already known from prior research. Unsurprisingly, using a machine to do your thinking for you produces negative results for your ability to think. The argument in this paper is that widespread LLM use can have systemic effects on a population. According to the paper, a persistent pattern of offloading cognition to AI can function similarly to a biological virus; as people benefit from it, they spread its use to others, and then they slowly become dependent on it if they don't carefully guard against that.

Interestingly, the authors of the paper aren't anti-Ai. In fact, the paper sort of goes out of its way to say that they don't think AI use is intrinsically harmful, and name a number of ways in which Ai use, including LLMs, can fundamentally aid human cognition rather than replacing it. That's basically the trick of how you avoid becoming "infected" with this mind-virus, by the way: don't let the AI do the thinking for you.

That doesn't mean "don't use AI," but rather, it means use the model for deep research, information gathering, and even instruction. As the authors say, "the distinction is not between using and not using AI, but between forms of coupling that extend human competence and those that replace the cognitive operations through which competence is maintained." The authors point to quite a bit of existing research showing that using AI critically, without replacing human cognition, is the key to avoiding this cognitive decline, and metaphorically, to inoculating yourself against the "virus".


Of course, it's important to keep in mind that, as I noted early on, it's not an experimental study. Rather, the authors are applying a lot of very advanced mathematic theory to existing research to derive their conclusion. Most of the paper is extremely impenetrable if you're not a mathematician or data scientist as a result, but the Discussion section toward the end is quite digestible and a pretty interesting read, so head over there if you're curious.

By the way, while we sometimes use AI for illustrative images, all of the text content on HotHardware is 100% human generated. If you'd like to support human-generated, fully independent tech journalism, check out our Deals posts! Besides finding you the best deals around, your purchases help to support the site and help us remain independent since 1999.