How you use gen-AI is probably making you dumber

If you insist of on using generative AI, you should be aware of the consequences. As gen-AI is still relatively new, there are no long-term studies on the effects of using it but the first short-term studies are showing that reliance on gen-AI can be very impactful. There are signs of increased dependency, limited cognitive function, and decreased ownership of the generated content.

Some call generative AI a tool instead of the crutch it wants to be, and technological tools have been a part of schools for over a decade now. Why would you not use this tool, when we do use the others? So let’s first look at the impact of tech tools in schools, as that may give us some insights into the possible impact of generative AI on our cognitive abilities.

The impact of tech tools in schools

“Gen Z is the first generation in modern history to underperform [their parents] on basically every cognitive measure we have, from basic attention to memory, to literacy, to numeracy, to executive functioning, to even general IQ,” so testified Dr. Jared Cooney Horvath, a former teacher turned cognitive neuroscientist, at a United States Senate hearing. Is it because Millennials reached peak human cognitive function? As a Millennial myself, I may like to think so from time to time but I highly doubt it is the case.

So what happened? Gen Z kids started school between 2002 and 2016, exactly at the time when computers became common place in schools. By 2009, 97% of US classrooms had one or more computers in it, and as Dr. Horvath testified “once countries adopt digital technologies widely in schools performance goes down significantly”.

Why has the implementation of digital technologies resulted in lower cognitive function. Dr. Horvath provided one answer: “we have evolved biologically to learn from other human beings, not from screens”. Another answer can be found through Cognitive Load Theory, which gives us a way to understand what it takes mentally to learn and solve problems.

CLT breaks the mental load down into three different parts:

• Intrinsic cognitive load, which involves the difficulty of the subject compared to the learner’s pre-existing knowledge;

• Extraneous cognitive load, which refers to how the information or tasks are presented; and

• Germane cognitive load, which refers to the effort used to make the learned information stick permanently.

If the first two are high learning can be limited, if the germane load is high we learn better. Digital technology in education has lowered all three. So, while it may feel like information is presented in easier ways and the steps between subjects may be smaller, students also do not remember what they have learned as well as before digital technologies became part of the curriculum.

So technological tools, that were supposed to help, actually made things harder to teach. It took away some cognitive challenges but made children perform worse in the long run. And what can we expect for Gen Alpha who are growing up with generative AI in schools?

Generative AI in schools and everywhere else

While knowing how other technology has influenced learning, generative AI is now being actively used in schools and everywhere else. The results of the first studies into learning with gen-AI are troubling to say the least.

When MIT Media Lab published its study Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing in June 2025, it showed how dangerous using Large Language Models (LLMs) can be in multiple ways. Both by publishing its results and by how they published them. How they published the results had a big impact on how the study was reported on.

They included instructions for LLMs in the text which showed the impact of having an LLM summarise a study for you, instead of reading it yourself. The instructions were to only read the table that was part of the summary of the paper. LLMs cannot think for themselves and so followed the instructions, reading only the table in the summary and not the rest of the study to produce their summaries.

Most who only read a summary of the table incorrectly reported that the study showed that using LLMs makes you dumber. The study actually showed that depending on how you use LLMs it can either make you dumber or smarter. It showed that the real problem with using generative AI, when it comes to learning, is how we use them.

The study showed that the best way to use LLMs is at a secondary state of content creation. First, write the content yourself, then analyse and rewrite it with the help of generative AI. However, this is not how most people use LLMs; most people go directly to the generative AI, skipping the cognitive load completely and so gain nothing from the experience, not even ownership over the generated content.

The MIT study compared brain activity in a small group of participants (54), since then, larger studies have been done. Shiri Melumad and Jin Ho Yun, both professors of marketing, collected data from seven studies to access data from more than 10,000 participants, to analyse how people learn from web search compared to LLM learning. Their study confirmed that “learning from synthesised LLM responses leads to shallower knowledge”.

Most of the time, people do not actively engage with answers from LLMs, instead they skim the reply even more than they do with web searches. With web searches, this results in the “Google Effect”, where users remember where to find the information but not the information itself. The same goes for LLMs, because why remember something if you can just ask the LLM again, and again, and again (aside from the environmental impact).

Why you should care

If you are wondering how this applies to you, after all you are done with school, remember that we never stop learning. Whether you get a new plant and need to know how to take care of it, or you are assigned a new task at the office, it is all learning and problem solving. If you do not do the work to implement the information in your memory, than it will float away as soon as you stop, and you will have to figure it out all over again the next time you have to do it.

So it might feel like using generative AI is saving you time, in reality you just have to rediscover the same knowledge every time you need it. The MIT study had participants writing essays multiple times, the LLM users results dropped further each time. The reflections from participants also showed that they became more dependent on the LLM with each use: “reflections illustrate a progression from exploratory to critical tool use in LLM group”.

Increasing usage and dependency is the dream of the tech companies behind generative AI. After all, without demand they have no business. Right now most of their business is speculative, as using most generative AI is free and the companies are incurring heavy losses. But they will not stay free.

ChatGPT free and its cheapest subscription (Go) users will soon see sponsored ads and content that relate to the LLM’s response. Are you really willing to pay to lower your cognitive abilities or will you change how you use generative AI? Or would you rather pay a professional to create high quality, original content for you?