I won't be using AI

tldr; I have used AI in the past, but I won’t be using AI going forward.

Artificial intelligence can mean different things to different people. I want to make a distinction between machine learning, the maths behind it, and probablity theory on one hand, and techno-capitalist text-extruder tools on the other. So here, I use the term “AI” to mean Generative AI tools such as ChatGPT, Claude and Gemini.

My previous AI use

I was convinced to start using AI through the productivity argument - Since code is now ‘cheap’, it is not worth my time to spend hours thinking and programming something that can be generated within a matter of seconds. Over the next several months, I outsourced the majority of my programming to AI, at the cost of atrophy of my own skills. I also began using AI for my writing so extensively that I couldn’t believe that I had actually written my own thesis without AI - I had lost confidence in my writing skills completely.

I was also increasingly using AI as a sounding board for my ideas. Instead of thinking, being frustrated, thinking more, repeating this a few times and finally distilling my thoughts, I was increasingly typing my stream of consiousness into AI. I think I was mildly addicted to this, because even after I decided to stop using AI, I found my hand itching to open a chat window.

What changed

In 2025, James Zou, a professor at Stanford University, gave the Overton Prize keynote at ISMB. The talk was about AI scientist agents and virtual labs. He said

Now, one of the things that I’m very envious of of these AI scientist agents is that their meetings are much more efficient than my meetings, right? So maybe one of their meetings would take a few seconds maybe a minute. So in the time that I had breakfast, they already had hundreds of these scientific research discussions and meetings. And moreover, because these AI scientist agents, they don’t get tired. They actually run multiple meetings in parallel.

The underlying assumption is that obviously researchers should be using AI, because the humans can never compete with an AI’s efficiency.

I have realised that I cannot get joy in resarch if I am defined by productivity and efficiency goals. I do not want to publish over 200 papers and 14 books in less than a year. Thinking about hard problems requires creativity and time (at least for me).

I found this article by Matilde Marcolli, professor of maths at CalTech, particularly cathartic to read. As she says -

What is for us humans the act of doing mathematics? I’d say mathematics is two very important things at once: a language to make sense of reality and a form of self-expression, much like the arts. When seen from this perspective, the existence of artificial intelligences that can generate solutions to mathematical problems becomes essentially irrelevant …

If you are a poet, the fact that a large language model can compose a dochmiac metered poem in Ancient Greek upon demand (I checked) is entirely irrelevant, because that would simply not be your poem. So don’t ask yourself which are the Big Problems™ and Famous Conjectures™ you should be working on and what is your reason of existence now that an artificial intelligence can get them before you’ve even finished your morning coffee. Ask yourself rather what is the mathematics that lives only inside you and nowhere else.

In another talk, when asked what skills scientists should learn in the future, Prof. Zou said he envisions a world where every scientist has their own personal AI lab behind them, and that we should learn skills such as auditing and management of AI agents. As other people have said, asking an AI to do research requires a different skill set than actually doing research. I would go a step further and argue that aquiring this new set of skills comes at a significant risk of losing existing skills.

Managing a team of AI agents by wading through their slop and auditing their “work” sounds like a nightmare to me. If this is what the future of research entails, life is too short and I would rather not do it.

Other Reasons

My field has been inundated with researchers strongly advocating for AI usage in every step. If AI can cure cancer, we should use it, right? A friend of mine called it “cancer-washing” of AI - similar to green washing by fast fashion brands, where the harms of AI are somehow justified with the tantalizing promise of a cure for cancer.

I reject this premise. I do not think the ends justify the means.

As Winner wrote in his seminal paper, technology encodes power, and artifacts have politics. Current AI systems are trained on data centered on a Western view of the world, and anything that falls outside this purview is immediately excluded. AI is centered on a colonial, heteronormative and sexist ideology. I do not want to become co-dependent on technology that perpetuates this world-view.

I am also convinced that using AI leads to the destruction of labour power, as explained very well here. If a worker is under the threat of being replaced by AI, this disincentivizes the worker from asking for pay-raises, benefits or time off. They have to work harder to “prove their worth”. On the other hand, if a worker is under the threat of being replaced by another worker who uses AI, this creates competetion between workers where they have to fight each other for jobs.

Potential Flipside

My social and professional circles overwhelmingly advocate for AI use, and perpetuate the idea that AI is inevitable - If I don’t use it, I will be left behind. I do not reject that view - more efficiency in research output is definitely important for funding bodies. A higher number of papers possibly indicates to a grant reviewer that they will get the most value for their money.

But I am not a machine and I do not want to be burned out by the pursuit of efficiency. Perhaps this stance will eventually push me out of science. I am ok with that.