
The ‘bureau’ where I did lots of thinking as a postdoc (circa 2014). Based on a photo taken on an old smartphone, sharpened up by GPT-6.
Image by Dominic Orchard, licensed under CC BY 4.0.
A colleague recently pointed me at this lovely article in the Economist, titled “AI often makes writing worse. (Even if it makes it better)” [1]. The article includes the following argument:
Real writing is not writing at all. It is thinking; specifically it is thinking: “What do I think?”
It goes on to quote Prof. Emma Smith (University of Oxford): “Often, ‘you only really know what you think by writing it’”. This reminded me of two other quotes which I have pinned in the introduction page of a “team” repository that I share with all my students and postdocs:
If you’re thinking without writing, you only think you’re thinking. […] Everyone thinks they think. If you don’t write down your thoughts, you are fooling yourself.
Writing is Nature’s way of showing you how sloppy your thinking is.
In all of these quotes, people are tapping into the idea that many people have discovered and discussed [3][4][5][6], that the act of writing gives you a chance to really think about an idea, problem, or argument. Writing helps you generate new ideas, deepens your exploration of a problem, and helps you understand how to evaluate it.
This aligns with the advice Simon Peyton Jones gives, that as soon as you have an idea, you should start writing the paper about it [2]. Why? Because it forces you to confront what you don’t yet understand and what evidence is needed. Writing opens up a way to have a dialogue with others who can critique and collaborate on the idea.
Going back to the Economist article, their point is that if we give writing over to AI we lose that opportunity to think for ourselves. I know some find it daunting, but I relish starting with an empty page and a fresh idea because I know that is where the action happens! Getting it down on paper helps my brain to think it through rather than “fooling myself” (Lamport) that I understand the idea well enough. And if I get stuck, I can always switch things up and try some different writing strategies. (I have written a whole list of them, to help me and my students; useful, I hope, when hitting writer’s block!)
If writing is thinking, then I believe the same also applies to programming. I do not think this will be a controversial point (at least not amongst my computer science colleagues), but in the current ‘AI summer’ this idea seems to be getting lost, and may be missed by those without a background in programming.
Programming is thinking
We are in a world where AI agents are writing large amounts of code, and people are wondering whether software engineering is a dying art with code increasingly becoming just an opaque box we do not need to, or indeed cannot reasonably, inspect. But just as writing is thinking, programming is thinking too! At least, it can be. Many times I’ve grappled with a problem or idea, from a variety of domains, by writing code about it. I often tell undergraduate students to ‘implement their mathematics’ as a way of helping them to get a better intuitive grasp of the ideas. By programming some concept or theory, the student gains a deeper appreciation for something abstract. A germane quote:
People have said you don’t understand something until you’ve taught it in a class. The truth is you don’t really understand something until you’ve taught it to a computer, until you’ve been able to program it.
(I couldn’t find the origin of the quote, but it is mentioned in a Communications of the ACM interview with Donald Knuth which touches on similar ideas [7].)
The point is that computers (in their traditional way of working) are very stupid. To ‘explain an idea to them’ you have to spell things out, in code, in considerable detail. But doing so makes sure that you yourself understand the idea. And then you have an object ‘in your hand’ which you can inspect and probe, by running it with particular inputs or applying other kinds of analysis. If we just automatically turn to an LLM to do our programming work, we miss out on the opportunity to actually explore a problem or an idea. Not all programming is like this, but certainly some is. This fits into the broader notion of ‘computational thinking’ [8] which was big about 10–15 years ago, where we leverage ideas from computing as analytical techniques to help solve problems and design systems.
So writing is thinking, and programming is thinking, and I think that a similar argument could be made for other kinds of ‘content production’. Perhaps a bit more of a parochial example, but for similar reasoning, I propose that:
Writing slides is thinking
I personally find writing slides for talks and lectures to be a hard slog, and it is a significant component of my job. But the process of writing those slides (which can take me as long as a day for a one-hour talk or lecture) gives me a chance to think through what key messages I want to get across, how I want to say them, in what order, what story I want to tell, and what examples to include. The deep work of staring at blank slides, moving bits of text and pictures around, is a time when my brain is preparing all the groundwork for the actual spoken, embodied delivery. I’ve sat through a lot of obviously (or explicitly stated) LLM-generated slide decks recently and I just cannot imagine being able to do a good talk myself in this way;1 maybe others can do it, but I need that time, that sometimes painful time, to sit soaking in the material trying to drag out a coherent story through it. For research talks, this almost always helps me better understand, analyse, and evaluate the key ideas in the research itself.
Passing large parts of these tasks (writing, programming, slide making) to AI then robs us of the chance to think. I don’t think this necessarily means there is no way to use LLMs effectively in the process of programming, but I think we need to make careful choices about what we are willing to not think about.
So, writing is thinking, and programming is thinking, and preparing slide decks is thinking.
Let’s keep thinking!
(No AI was used in the writing of this blog post).
References
- The Economist. AI often makes writing worse. (Even if it makes it better). The Economist, September 2026. ↩
- Simon Peyton Jones. How to write a great research paper. Slides. ↩
- Keith Oatley and Maja Djikic. Writing as Thinking. Review of General Psychology, 12(1), 9–27, 2008. ↩
- Richard Menary. Writing as thinking. Language Sciences, 29(5), 621–632, 2007. ↩
- Hans Sluga. Thinking as Writing. Grazer Philosophische Studien, 33, 115, 1989. ↩
- Mary S. Lawrence. Writing as a Thinking Process. University of Michigan Press, 1996. (Review by Robert Retherford, TESL-EJ, 2(4), 1997.) ↩
- Donald Knuth. Interview: Donald Knuth: A Life's Work Interrupted. Communications of the ACM, 51(8), 31–35, 2008. ↩
- Jeannette M. Wing. Computational thinking. Communications of the ACM, 49(3), 33–35, 2006. ↩
-
This is similar to how I find it very hard to reuse someone else’s slides for teaching; I need the time to think about how I’m going to explain it, otherwise I will not be able to communicate the ideas as effectively. Relatedly, I saw a talk by a senior academic who admitted the slides were AI-assisted; partway into one slide they said “I don’t like this slide, AI has made it too complicated” and then just skipped it! ↩