Highlights
- I expect that efforts to integrate AI into every part of our lives will be followed by attempts to return to AI-free approaches. Post This
- A recent MIT study, though undergoing peer review, indicates that using ChatGPT negatively impacts individuals' cognition when writing essays. Post This
- Foregoing generative AI keeps my work intrinsically enjoyable and allows me to further develop my own programming and writing abilities. Post This
Generative AI tools like ChatGPT, GitHub Copilot, and Gemini have become very popular for both programming and writing. Because much of my professional and personal life is spent writing, coding, and analyzing data, I would be a prime candidate for many of these applications. However, I have concluded that not using AI1 is the best decision for me, both professionally and personally. I don't want to lose the intrinsic enjoyment I derive from my work, and I'm also concerned about the cognitive effects of relying on AI tools.
I love what I do for a living. Data analysis and coding offer an exciting combination of problem solving, creativity, and design in which I can get immersed for hours. It's very rewarding to convert a raw dataset into a series of analyses and charts that I can then share with others.
1. I'd Rather Not Delegate Writing and Coding to a Chatbot
There are times when programming feels more like a game than labor. I encounter a variety of challenges, then try out different tools to resolve them. It's gratifying to build up each script, line by line and function by function, until it develops into a finished product. (The experience is a bit like building a metropolis in Sim City, but with fewer earthquakes.) I certainly get frustrated at times when programming, just as I would with a challenging game, but the more vexing a particular obstacle or error message can be, the more exciting it feels to overcome it.
My love of coding is also a prime example of the "IKEA effect," a term coined by Norton, Mochon, and Ariely to describe how "labor alone can be sufficient to induce greater liking for the fruits of one's labor." Although I make extensive use of open-source code that others have written, my own code is usually more meaningful to me, even though it may not be as advanced or flexible as third-party tools.
If I assigned these coding and writing tasks to a chatbot, I expect that I would not enjoy my work as much as I do now. I would feel less of a connection with the final product, thus missing out on the IKEA effect, and would also forfeit the many intrinsic rewards of the writing and programming process. The prospect of spending much of my working life correcting a chatbot's errors, only for new errors to pop up along the way, seems less appealing (and not necessarily faster) than simply learning how to do a task myself.
My concerns about delegating work to AI are expressed well by Hippolyte, a character in Dostoevsky's The Idiot, who proclaimed that what matters is "the process of discovering, the everlasting and perpetual process, not the discovery itself." I'm sure that AI can sometimes reduce the amount of time needed to generate a finished product. However, the process of getting to that discovery—debugging my code, posting on Stack Overflow and GitHub, and testing out different methods from documentation files, all while a working product slowly takes shape—is often more fulfilling than the completed work.
Is this desire to enjoy programming counterproductive and out of touch? It's not, argues Colton Voege in an excellent piece titled "No, AI is not Making Engineers 10x as Productive." He writes that enjoying the coding process is "essentialin our field. If you force yourself to work in a way you hate, you're just going to burn out."
2. AI Could Have Negative Cognitive Impacts
The more I code, the more I agree with Kristen Nygaard that "programming is understanding."2 Jeff Duntemann concurs with this statement, writing: "Learning to be a programmer, furthermore, is almost entirely a process of learning how things work."3 Therefore, in order to perform my work well, I need to continually deepen my understanding of statistics, coding, and data visualization. If I were to rely on generative AI tools instead, this knowledge might never develop—just as a student who jumps immediately to the answer key of a textbook would fail to understand how to arrive at that answer.
New evidence supports my concern that AI has negative cognitive implications. A recent MIT study, though still undergoing peer review, indicates that using ChatGPT negatively impacts individuals' cognition, at least when writing essays. The study's authors found that, "over the course of 4 months, the LLM group's participants performed worse than their counterparts in the Brain-only group at all levels: neural, linguistic, [and] scoring." The authors note that using ChatGPT appears to have resulted in "cognitive debt," a phenomenon in which individuals outsource their own thinking to a chatbot. They explain that cognitive debt "defers mental effort in the short term but results in long-term costs, such as diminished critical inquiry, increased vulnerability to manipulation, [and] decreased creativity." I expect that these long-term drawbacks of AI will become increasingly more apparent in the coming years.
Programming lets me 'level up' my abilities by learning new skills, strategies, and languages. As with many role-playing games, this leveling up comes through experience and repetition.
Alex Green, a public policy lecturer at Harvard Kennedy School, provides further evidence in a recent WSJ op-ed of the cognitive risks of AI use. Many of his students, he writes, report "that after relying on AI to draft their papers and emails, their ability to write, speak and conduct basic inquiry is slipping away." While an incoming OpenAI executive hypothesized that "AI can give everyone more power than ever," Green is more critical of its actual effects, stating: "no new technology has produced such a terrifying admission of stark and fundamental disempowerment by my students as AI has."
My own experience supports the idea that over-relying on technology can hold one back mentally. When I lived in New York, I used Google Maps to navigate to our local grocery store, even though it was only an 11-minute drive away. One day, I decided to try driving to the store without using this tool. To my surprise, I went straight past the left turn to the shopping center where it was located.
This was 'cognitive debt' in action. Typically, I would learn to get somewhere by taking the same route over and over. However, my reliance on an app, while convenient, short-circuited this process. (I still use Google Maps, admittedly, but I'm trying to rely on it less.)
The authors of the MIT study suggest that, once users have achieved "self-driven cognitive effort," they could then consider incorporating AI into their work. This approach is more ideal than relying on AI from the beginning. However, I suspect that it would still cause my own command of the tasks I delegate to AI to atrophy, thus making it harder to learn new tasks that rely on them.
To return to the game analogy, I appreciate how programming lets me 'level up' my abilities by learning new skills, strategies, and languages. As with many role-playing games, this leveling up comes through experience and repetition. Having your older brother beat a tricky video game boss can save you time, but you'll then remain unprepared for the even-harder final boss. Similarly, using AI as a shortcut to produce work could unintentionally keep my own skills stuck at a lower level, thus undermining my ability to tackle harder tasks later on.
3. Foregoing AI Helps Me Enjoy My Work and Develop My Skills
Tech CEOs and AI startups may give you the impression that, if you do not immediately adopt generative AI tools, you are doomed to be 'left behind'. But this hasn't been my experience. Foregoing generative AI keeps my work intrinsically enjoyable and allows me to further develop my own programming and writing abilities.
Granted, not everyone enjoys programming or writing to the same extent. Does this mean that generative AI is the right choice for them? Not necessarily. AI has serious accuracy issues that could be worsening; it may actually slow coders down; it might infringe on others' copyrights; and it can have negative effects on human relationships. Therefore, there are solid reasons to continue to write and code without AI 'assistance.'
There's a popular mindset that, because 'AI is here to stay,' we must simply adapt to the new world it's creating. However, growing support for school cell phone bans, along with successful efforts to increase the minimum age for accessing social media, both demonstrate that new technological developments are not unavoidable: societies, communities, and individuals can indeed choose a 'tech exit.' I expect that efforts to integrate AI into every part of our lives will be followed by attempts to return to AI-free approaches; if this prediction comes true, I'll have saved lots of time and effort by not basing my workflows on AI to begin with.
Ultimately, everyone needs to decide for himself or herself whether, and how, to apply generative AI tools. My choice to forgo them4 may not be right for everyone, but I'm confident that this approach will ultimately make for a happier and more satisfying life.
Ken Burchfiel is a Research Fellow at the Institute for Family Studies.
Photo credit: Shutterstock
Endnotes
1. Whenever I use the term 'AI' in this essay, I am referring to generative AI tools that can produce prose and code, rather than to artificial intelligence in general. Whenever I use a search engine or run a regression analysis, I am also making use of AI, but not generative AI. (And yes, I did disable generative AI summaries within my web search results.)
2. It's also fair to say that "writing is understanding." In the process of writing this article, I needed to think carefully about my beliefs; review what others have to say on the use of AI; and provide a structure for those beliefs. Writing is, among other things, a concrete expression of learning.
3. Duntemann, Jeff (2023). x64 Assembly Language Step-by-Step: Programming with Linux. Wiley.
4. I have played around with AI tools for generating text and images, and I've also read AI-generated summaries of product reviews, so I can't say that I have lived a completely generative AI-free life. However, I haven't applied them for any serious, public-facing work.