The only thing the movie Wall-E got wrong was how quickly the laziness would come. In our AI-driven, automated future, being fully human will be an act of resistance ✊
I work in e-commerce and I hear “just ChatGPT it” way too often at work. I also feel like when we do have successful breakthroughs, no one can explain why because ChatGPT was doing to hard work. I am also the oldest person in the office so if I say something they just shake their heads and say I’m too old school. I’ll just keep flexing my cognitive muscles as long as I can get away with it!
Brad, thanks for so many points in this article. First, for background I should state that I'm a semi-retired attorney. I went to a Continuing Legal Education seminar here in town this spring, and even though it was in-person, definitely not the norm these days, it was about AI in the law and ethical issues surrounding that. Surprisingly, most of the lawyers in the room used AI in their practices. The argument was that in private practice, it might be unethical to bill clients for old-fashioned research that takes hours when AI can do it so quickly and therefore cheaper. The presenter did point out the need to check AI's work because it makes up cases out of thin air. Disturbing. There's more I could say about AI in the legal profession, so maybe I'll write a post about it. Finally, thanks for the heads-up on Derek Thompson. Just subscribed to his Substack and been reading his column on the workforce in the Atlantic for quite a while.
Karl - thanks sharing this. My wife is an attorney, so I am no stranger to discussions about AI in the legal field. As I tried to convey in the piece, I think this is going to be an ongoing case of "it depends" and "a tool's impact results from how you use it." Thanks for reading and stopping by to discuss!
Thanks for this, Brad. This research and your musings on it reflect what I've seen in the world of software: when engineers have used LLMs to write all their code, they don't know how it works, are not able to follow detailed code reviews and are unable to show down enough to think through the code they *do* write. Whereas, many senior engineers have played with LLMs to aid their coding and have largely relegated them to simple tasks (like writing so-called boilerplate code), reserving harder code and higher level architecture decisions to humans.
It is also possible that writers with less cognitive processing ability to begin with were more likely to use ChatGPT for their writing. However, I would not be the least bit surprised that not using our brains Creatively is bad for cognitive functioning.
Based on a study of 54 high end college students in Boston writing SAT style essays and assessed with brain scans and questionable guidelines, you choose to write a fear mongering article like this? This research is not just thin; it’s nonexistent.
While I agree with you, I do think it's *also* worth pointing out that a UK study was done with a larger sample size (600 subjects). https://www.mdpi.com/2075-4698/15/1/6
The actual results are interesting, but, as with all data, interpreting the results are key. I don't know that we can really draw too many conclusions on the ramifications just yet.
I have been outsourcing most of my work to AI, giving it papers to summarise for some months now, it has been efficient but I realised I lost alot. Few hours prior to reading this post I was reading a short research paper, I found it extremely hard to concentrate and make meaning out of the paper, not because I didn't understand the topic but I felt bored easily, spaced out several times. I realised I had lost my ability to concentrate, pay attention. Realising this I made a note to myself to stop outsourcing everything to AI because I was becoming too reliant, loosing my cognitive skills in the process. I am really glad I found this piece. I didn't know there was even a study to back exactly what I have been experiencing lately.
They should have compared brain connectivity metrics on people who didn’t write essays at all during that time (ai assisted or otherwise). Would it be better or worse than the gpt users?
But did you know professors don't read term papers? At least not in the college I recently graduated from.
They only skim through it. They are more interested in grammar than ideas. Conservative universities favour the lack of individual arguments, which I think chatgpt can favour since why would it have drastic opinions on what shakespeare character is the most homoerotic?
I've come to find that AI can be useful for writing in two ways:
1. Sending it a draft and getting its thoughts. 4o is far too sycophantic, but sometimes that can serve as a little motivation boost if you're not feeling inspired, just gotta make sure you remember that the model is designed to tell you that whatever you're doing is amazing. Plus, even that model can make you aware of points that may be worth building on or other things to read. And you can always use other models or give it a prompt like "pretend you're a philosophy professor and give me a scathing criticism" to have even 4o completely rip you apart.
2. Talking with it conversationally to brainstorm and fill in knowledge gaps. If I'm writing something about a complex topic, just shooting some ideas back and forth with it can help me crystallize some of my thinking and get ideas about parts of the topic that I didn't know about. I'll often ask it for a reading list that is tailored specifically to my idea — which saves a lot of time when the idea draws on several disparate fields.
What I DO NOT do is ever put anything that ChatGPT wrote into my pieces. I just feel that's crossing a line. Only exception is if I'm writing an email or text, and I feel like it might come off the wrong way. In that case, I'll send it to ChatGPT, ask its thoughts, and it will tell me how it comes across and then make some suggestions, some of which I will take verbatim. But that's really not much different than just running an important email/text by a person you trust.
Having written this comment, makes me think maybe I should write something up on this...
There is thinking that concerns like this echo those raised when the calculator was developed. I’m not sure. It seems as though perhaps the calculator is like equipment at the gym that aids your workout. Not something that replaces it.
We'll see. In these circumstances, I tend to follow the precautionary principle: if the result is striking but also makes sense, and the action (in this case, being very intentional about how you use ChatGPT and other LLMs) is simple and practical, I take the action.
The only thing the movie Wall-E got wrong was how quickly the laziness would come. In our AI-driven, automated future, being fully human will be an act of resistance ✊
I work in e-commerce and I hear “just ChatGPT it” way too often at work. I also feel like when we do have successful breakthroughs, no one can explain why because ChatGPT was doing to hard work. I am also the oldest person in the office so if I say something they just shake their heads and say I’m too old school. I’ll just keep flexing my cognitive muscles as long as I can get away with it!
Worst of all, people are now getting used to mechanical work, hungering for control and perfection over humanity and hardwork
Agreed. True. Prime directive: stay busy at all costs and keep the numbers up.
Brad, thanks for so many points in this article. First, for background I should state that I'm a semi-retired attorney. I went to a Continuing Legal Education seminar here in town this spring, and even though it was in-person, definitely not the norm these days, it was about AI in the law and ethical issues surrounding that. Surprisingly, most of the lawyers in the room used AI in their practices. The argument was that in private practice, it might be unethical to bill clients for old-fashioned research that takes hours when AI can do it so quickly and therefore cheaper. The presenter did point out the need to check AI's work because it makes up cases out of thin air. Disturbing. There's more I could say about AI in the legal profession, so maybe I'll write a post about it. Finally, thanks for the heads-up on Derek Thompson. Just subscribed to his Substack and been reading his column on the workforce in the Atlantic for quite a while.
Karl - thanks sharing this. My wife is an attorney, so I am no stranger to discussions about AI in the legal field. As I tried to convey in the piece, I think this is going to be an ongoing case of "it depends" and "a tool's impact results from how you use it." Thanks for reading and stopping by to discuss!
Thanks for this, Brad. This research and your musings on it reflect what I've seen in the world of software: when engineers have used LLMs to write all their code, they don't know how it works, are not able to follow detailed code reviews and are unable to show down enough to think through the code they *do* write. Whereas, many senior engineers have played with LLMs to aid their coding and have largely relegated them to simple tasks (like writing so-called boilerplate code), reserving harder code and higher level architecture decisions to humans.
I’m so glad that the last line in this piece was not “this piece was written by AI”. Because that would have been waaaay to 2025.
It is also possible that writers with less cognitive processing ability to begin with were more likely to use ChatGPT for their writing. However, I would not be the least bit surprised that not using our brains Creatively is bad for cognitive functioning.
Based on a study of 54 high end college students in Boston writing SAT style essays and assessed with brain scans and questionable guidelines, you choose to write a fear mongering article like this? This research is not just thin; it’s nonexistent.
While I agree with you, I do think it's *also* worth pointing out that a UK study was done with a larger sample size (600 subjects). https://www.mdpi.com/2075-4698/15/1/6
The actual results are interesting, but, as with all data, interpreting the results are key. I don't know that we can really draw too many conclusions on the ramifications just yet.
Also, as pointed out in the Washington Post https://web.archive.org/web/20250630000444/https://www.washingtonpost.com/health/2025/06/29/chatgpt-ai-brain-impact/ we have feared new technology since recorded history. So, you know, I take these fear-mongering articles with a grain of salt.
I have been outsourcing most of my work to AI, giving it papers to summarise for some months now, it has been efficient but I realised I lost alot. Few hours prior to reading this post I was reading a short research paper, I found it extremely hard to concentrate and make meaning out of the paper, not because I didn't understand the topic but I felt bored easily, spaced out several times. I realised I had lost my ability to concentrate, pay attention. Realising this I made a note to myself to stop outsourcing everything to AI because I was becoming too reliant, loosing my cognitive skills in the process. I am really glad I found this piece. I didn't know there was even a study to back exactly what I have been experiencing lately.
They should have compared brain connectivity metrics on people who didn’t write essays at all during that time (ai assisted or otherwise). Would it be better or worse than the gpt users?
Good analysis. But this headline is a bit alarmist?
Cute essay
But did you know professors don't read term papers? At least not in the college I recently graduated from.
They only skim through it. They are more interested in grammar than ideas. Conservative universities favour the lack of individual arguments, which I think chatgpt can favour since why would it have drastic opinions on what shakespeare character is the most homoerotic?
I've come to find that AI can be useful for writing in two ways:
1. Sending it a draft and getting its thoughts. 4o is far too sycophantic, but sometimes that can serve as a little motivation boost if you're not feeling inspired, just gotta make sure you remember that the model is designed to tell you that whatever you're doing is amazing. Plus, even that model can make you aware of points that may be worth building on or other things to read. And you can always use other models or give it a prompt like "pretend you're a philosophy professor and give me a scathing criticism" to have even 4o completely rip you apart.
2. Talking with it conversationally to brainstorm and fill in knowledge gaps. If I'm writing something about a complex topic, just shooting some ideas back and forth with it can help me crystallize some of my thinking and get ideas about parts of the topic that I didn't know about. I'll often ask it for a reading list that is tailored specifically to my idea — which saves a lot of time when the idea draws on several disparate fields.
What I DO NOT do is ever put anything that ChatGPT wrote into my pieces. I just feel that's crossing a line. Only exception is if I'm writing an email or text, and I feel like it might come off the wrong way. In that case, I'll send it to ChatGPT, ask its thoughts, and it will tell me how it comes across and then make some suggestions, some of which I will take verbatim. But that's really not much different than just running an important email/text by a person you trust.
Having written this comment, makes me think maybe I should write something up on this...
Because they are using LLMs all wrong! You use it as an editing tool to check your own writing, not to cut and paste it into your essay.
Can you please share the refrence of this research paper?
There is thinking that concerns like this echo those raised when the calculator was developed. I’m not sure. It seems as though perhaps the calculator is like equipment at the gym that aids your workout. Not something that replaces it.
I doubt this replicates
We'll see. In these circumstances, I tend to follow the precautionary principle: if the result is striking but also makes sense, and the action (in this case, being very intentional about how you use ChatGPT and other LLMs) is simple and practical, I take the action.