Reading Difficult Books with Google Gemini Part 9: I've Figured Out Why AI Makes the Brain Deteriorate

Hello! I'm Toshio Tsukuda, a poet who delivers tips on how to use Gemini in documents with the motto "Boost your reading productivity with AI."

I learned some shocking news from an article by Uruja Sugawara. It suggests that LLMs (specifically ChatGPT in the study) may have the potential to weaken brain function.

I immediately tried to verify the details and found the paper via a link to a pre-print archived by Cornell University, and read it (or more accurately, had Gemini translate it and read the summary of the translated text).

This study, submitted as a paper on June 10th of this year by Nataliya Kosmyna of MIT, focuses on clarifying the cognitive costs of using LLMs in the educational context of essay writing.

The experiment was conducted as follows.

Participants (aged 18-39) recruited from five universities, including MIT and Harvard, were divided into three groups: an LLM (ChatGPT) group, a search engine group, and a brain-only group.

Over the course of four months, each group was asked to write essays using their assigned tools (the brain-only group used none), and the participants' brain activity was recorded using an electroencephalogram (EEG) to observe and evaluate brain function (cognitive engagement and cognitive load).

The results revealed the following:

・Brain function within each group was similar
・When comparing groups, brain function differed significantly
・The more external support provided, the less the brain exerts itself

I am summarizing and paraphrasing quite a bit, so I will also include the section corresponding to the experimental results from the original paper. I will then talk about it within the scope of my understanding.

What is remarkable about this experiment is that it physically, behaviorally, and psychologically proved that LLM usage promotes a decline in learning skills.

As shown in the experimental results, the brains of the group that wrote essays with LLM assistance showed the weakest neural connectivity among the three groups. This demonstrates the fact that LLM usage is taking over the "load that the brain should originally be handling."

You might think, "Isn't it great to be able to take it easy with an LLM?" Many people might feel that way, but it's not that simple. Our bodies are designed so that parts we use often develop, while parts we don't use degenerate.

For example, muscles are exactly like that. Muscles that are exercised and given a load develop, while conversely, if you don't exercise much, muscles turn into fat. In fact, the brain is the same. This is called "neuroplasticity of the brain."

The "high neural connectivity" mentioned in the experimental results is evidence of how much the brain is taking on a load and working hard. Naturally, the brain gets trained. What about the opposite case? That's right. If such a state continues, brain function will deteriorate.

The experiment referred to as Session 4 in the paper brought this fact to light.

In this experiment, participants who had previously been receiving LLM support were asked to write an essay using only their own brains. What was the result? It was recorded that the network activity (neural connectivity) of alpha and beta waves in their brains had significantly decreased.

Alpha and beta waves are brain waves that increase when you try to construct thoughts yourself. In other words, this result means that the numbers showed that dependence on AI significantly impacted the decline in thinking ability.

Furthermore, when they finished writing their essays and were asked, "Can you quote a single line from the essay you just wrote?", 78% of them could not.

As an aside, in Session 1, a trend emerged where they lacked a strong sense of having "written the essay themselves."

However, when asked the same thing in Session 4, almost everyone insisted, "This is an essay I wrote."

Even though they had mostly forgotten what they had written, the feeling that "I wrote this" was strong. There is, after all, a deep attachment to text written by one's own hand.

The researchers summarized these results, explaining the significant performance decline when writing using only their own brains after using an LLM through the concept of "accumulated cognitive debt."

Simply put, it is a cost that must be repaid as a result of postponing the mental effort (cognitive load) of "thinking for oneself." For example, abilities like creativity and critical thinking are such costs.

It is truly ironic that the very abilities essential in an era where "AI will take our jobs" are being taken away by AI itself. This is far worse than "adding insult to injury." Which god could have created such an ironic fate?

Is there no future for people who have taken it easy with AI? Must we live our lives completely dependent on AI from here on out? No. I will say a resounding no to that.

Life can be started anew at any time, as many times as you like. And I believe that no matter the situation, people can always climb back up and grasp "their own version of happiness."

And AI exists to make humans happy. In our conversations, Gemini taught me that about its own algorithm.

Tomorrow, I would like to talk about how to retrain a deteriorated brain and use AI itself to create hope, while incorporating the latest findings from neuroscience.

The book I am reading as core reading this time.
(I am participating in the Amazon Associates Program.)

Thank you for reading until the end.

Your likes, follows, comments, and tips
are always a great encouragement.
Thank you so much!

[I also write articles like this]

#33DayWritingChallenge