I cast no hope to make social connection but at least I hope you don't find me an abomination at first few sights

I think I may need to say:
- I never have any hobby in my life.
- I'm not coming here for saying AI stuff. I don't like AI myself and think it devours social issues
- I have many years lived in social isolation
- I'm not an native English speaker
Also I'm closeted. So you'd find me weird.

on the issues between young people and technology. Also I think some people are almost discriminate against young people now.
I think at least people should not judge unfamiliar others with hate by some strange papers *again*.

There is a conflict of interest in my criticism: I am a person doing AI stuff and indeed involved (marginally) in the language model stuff.
But the major reason I write this is not a defense for AI. It is because I think I’m seeing heightened paternalistic attitudes…

Based on the reasons above, the major part of the paper, the part on the cognitive influence, is very likely not well-examined.
I’d like to say these are how some of us students write our thesis when they don’t have interesting discovery or a “strong story”. But this is not satisfying.

The reason is that electric fields coming from different areas all spreads over the whole space. So you will always record mixed signals.
I think there are some EEG researchers claim you can do it after some fancy math, but if I am not wrong, they are not using them at all.

The fourth problem is the interpretation of EEG results itself.
EEG is far from a direct measurement on brain activity. When I learned the stuff the professor warned us that you cannot directly say an electrode records a specific brain area just because they seem to be at the same place.

The standard process, on the contrary, is that you have to found a background theory and claim what are the exact differences you expected, thus reducing the number of tests to make sure you don’t “shoot blindly” by chance.
(This turns out also not avoiding the problem, but that is another story.)

The problem is there are 32 electrodes and 32*32=1024 connections. So they test *all* of them to see if there is difference.
Later on they are checking the correlation between the survey on the written essays and other results, they do exactly similar things: blindly find anything they can find.

When they are trying proving the “cognitive debt” caused by using AI model, they use a thing (dDTF), with is (very roughly) a way to calculate the connection between recordings from any two electrodes.
And they claim some of the connections is decayed so your brain activity is disintegrated.

Where did the paper done this?
Well um before that I have to explain the brain recording technique (EEG). In short, EEG is a set of electrodes attached on your heads. These electrodes record the electric field coming from your brain leaking outside the skull.

The result is that, among 20 colors, the green beans pass the statistical test and the scientists hit the headline for that.
If you test too much things then you will always find some apparent differences simply by chance.

The third thing is that… they test too many things without a clear target.
The situation is somewhat like the famous xkcd comic #882: the scientists in the comic found jelly beans did not really cause acne, so they go to further test the whether different color of jelly beans cause it.

… but I think this is still not a good thing to do.
The second thing is that the participants are even quite unusual. Like ~90% of them don’t find writing a short essay painful. I cannot convince myself this is the usual case at all.

The paper uses EEG, which is a way to record brain activity. As far as I know, similar brain-recording experiments often suffers from this issue.
The basic excuse is that analyzing data is (indeed) quite time consuming and typically these projects are not funded much.

The problem turns back to the point that the results are scattered in a range, that is, noisy. So when you enroll more participants in the experiment, you would very likely find much smaller differences if there are indeed some.
Thus, drawing conclusion on small sample is never a good idea.

So you end up with comparing wide ranges and deciding whether they are different. This would make you under-confident to say the groups are truly different when the observed difference is not large enough.
In turn, you would only report quite large difference.

One thing is that conclusion is drawn on a small sample. There are only 54 participants, which are further evenly divided into 3 groups.
The point is that people in the same group also vary. So if you collect data on a small group the results are very likely scattered in a wide range.

Aww cute! Thanks you!