More Reactions on "Your Brain on ChatGPT"
More people point out flaws with the headlines related to the MIT Paper
A few days back, I wrote that ChatGPT won't make you dumb unless you want to. This was essentially a counterpoint to an MIT paper called “Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task”. Somehow, people missed this, and a number of them sent me a link to the MIT paper asking me for my reaction.
Clearly, this paper has captured the attention of a lot of people, and they didn’t pay attention to my earlier article, so I’m doing another article, this time with reactions of other people, all pointing out that you can mostly ignore that paper.
The Study was Lame
This paper shows the same effect as other studies of "cheating" with AI - if you use AI to do the work (as opposed to using it as a tutor), you don't learn as much.
But note: the results are specific to the essay task - not a generalized statement about LLMs making people dumb.
—
The crisis caused by AI cheating is real, and students using AI to cheat are learning less (just like students copying from the internet learned less).But AI can also act as an effective tutor. It is very much dependent on how we use the technology
—
This study is being massively misinterpreted. College students who wrote an essay with LLM help engaged less with the essay & thus were less engaged when (a total of 9 people) were asked to do similar work weeks later.
LLMs do not rot your brain. Being lazy & not learning does.
—
This line from the abstract is very misleading: “Over four months, LLM users consistently underperformed at neural, linguistic, and behavioral levels.”
It does not test LLM users over 4 months, it tests people who had an LLM help write an essay about that essay 4 months later.
Less Brain Activity Is Not Always a Bad Thing
The MIT study pointed out that students brain activity reduced when using ChatGPT to create the essay. So, what, asks Tyler Cowen. When you use less brain on a low-level activity (which even ChatGPT can do), you free up mental energy that can be applied elsewhere. That’s a good thing.
It took me a lot of “cognitive load”—a key measure used in their paper—to memorize all those state capitals in grade school, but I am not convinced it made me smarter or even significantly better informed. I would rather have spent the time reading an intelligent book or solving a math puzzle. Yet those memorizations, according to the standards of this new MIT paper, would qualify as an effective form of cognitive engagement. After all, they probably would have set those electroencephalograms (EEGs)—a test that measures electrical activity in the brain, and a major standard for effective cognition used in the paper—a-buzzin’.
The important concept here is one of comparative advantage, namely, doing what one does best or enjoys the most. Most forms of information technology, including LLMs, allow us to reallocate our mental energies as we prefer. If you use an LLM to diagnose the health of your dog (as my wife and I have done), that frees up time to ponder work and other family matters more productively. It saved us a trip to the vet. Similarly, I look forward to an LLM that does my taxes for me, as it would allow me to do more podcasting.
If you look only at the mental energy saved through LLM use, in the context of an artificially generated and controlled experiment, it will seem we are thinking less and becoming mentally lazy. And that is what the MIT experiment did, because if you are getting some things done more easily your cognitive load is likely to go down.
But you also have to consider, in a real-world context, what we do with all that liberated time and mental energy. This experiment did not even try to measure the mental energy the subjects could redeploy elsewhere; for instance, the time savings they would reap in real-life situations by using LLMs. No wonder they ended up looking like such slackers.
It Didn’t Work Because Students Aren’t Managers
As we discussed yesterday, we are all now managers of LLMs, and hence we need to learn people/LLM management skills. Venkatesh Rao believes that the reason the students who used ChatGPT in the MIT study flopped is because they did the exercise as “students” and not as “managers”. He says that this is exactly the problem new managers face until they learn the ropes of how to be good managers:
So Why Did the Students Flop?
Because freshman comp doesn’t teach management.
Drop novices into a foreman’s chair and they under-engage, miss hallucinations, and forget what the intern wrote. Industry calls them accidental managers.The cure isn’t ditching the intern; it’s training the manager:
delegation protocols
quality gates
exception handling
deciding when to tolerate vs. combat sameness
His overall reaction to the paper is:
The EEG paper diagnoses “cognitive debt,” but what it really spies is role confusion.
We strapped apprentices into a manager’s cockpit, watched their brains idle between spurts of oversight, and mistook the silence for sloth.
and
Better managers, not louder brains, are the upgrade path.
Maybe AI Will Make Us Intellectually Sedentary?
I worry that AI will make most people intellectually sedentary
When we no longer had to do as much manual labor thanks to new technology, we became more sedentary and baseline fitness declined
At the same time, a minority of people got fitter than ever because they exercised with intention, for status or self-care
AI could create a world where most people become intellectually slovenly, while the privileged or determined few give themselves mental superpowers
How to Fix Students? Learn from Chess?
Technology is banned in chess matches but players can still use it when they train. Paradoxically, the ban makes it easier to use technology to help you learn.
We should copy this in education - banning technology in exams will promote better use of it in the classroom.
Allow students the full use of ChatGPT when learning, but make it clear to them that at least some of the exams will require them to complete tasks without access to ChatGPT, forcing them to practice those skills? (This doesn’t preclude additional exams which require them to use ChatGPT expertly to achieve more complex things.)
Funny, But Maybe True?
TheZvi complains about the entire premise of “write an essay on a topic” as being a useful homework assignment in the age of AI:
That’s because homework is a task chosen knowing it is useless. To go pure, if you copy the answers from your friend using copy-paste, was that productive? Well, maybe. You didn’t learn anything, no one got any use out of the answers, but you did get out of doing the work. In some situations, that’s productivity, baby.
Arnold Kling, in Student Assessment in the Age of AI points out something intriguing:
Frankly, I suspect that what students learn about AI by cheating is probably more valuable than what the professors were trying to teach them, anyway.