AI, writing, and me

Substack announced yesterday that they’re introducing a feature which will allow users to check whether a post was written by AI or not. It doesn’t seem to be working for me yet - perhaps it’s being rolled out in different regions at different times. But as far as I understand, you will click on the button with the three dots at the top of the post and then be able to select an option which will perform the check.

So long as it actually works and doesn’t give false positives, I think this is a good idea. In fact, it addresses a problem that I’ve been thinking about a lot recently.

I have never used AI to write, edit, draft, or redraft (or any other verb) any post for this newsletter, and I can’t imagine that I ever will. To be honest, I’ve never even really gotten as far as thinking about the moral problems with doing it, because I just don’t see any reason to even think about it as a practical proposition. I am very comfortable - and fast - at writing in my own voice, and AI cannot write in my own voice better than I can. So I just have no reason to even think about doing it at all.

LLMs have gotten good at many things, but I still don’t think they’re very good at writing. Or rather, they’re “good” in a narrow technical sense, but they produce writing which lacks some intrinsic human quality. One reason for that is that their output tends to be very samey. The voice always sounds to me like a parody of a voice, or like someone going through the motions without there being any real human feeling behind it - a kind of philosophical writer zombie. And then there are the annoying tics, which range from repetitive phrases like “it’s not X, it’s Y” through to the combination of rhetorical sophistication with extreme shallowness.

One of the things that I think is interesting about all of these problems, though, is that they’re not unique to LLMs. There have always been human writers who reused the same tics, whose points just didn’t seem to connect to each other, or who substituted rhetorical fanciness for having a deep point. Everyone’s favorite writers vary, but there are some pretty famous (or at least internet-famous) writers whose work I’ve somehow just never connected to, for these reasons or others. Bad writing is subjective, but it’s always been with us.

However, the problem that LLMs generate is that they dramatically reduce the cost of producing writing which has these flaws. Back in the day, even to be a bad writer on geopolitics or ancient history or whatever, you still needed to put in the hours. You had to have a conceptual and theoretical foundation, a knowledge of historical cases, and then you had to follow the news to keep up with current events. Now people don’t need to do any of these things, or they only need to do them a little bit, perhaps to the extent that they can edit the LLM output and ask it relevant questions in the first place.

So although bad writing is not a new thing, I am genuinely worried that as the ease of producing it comes down, there is a real risk of platforms like Substack becoming submerged in it. This is particularly the case because “bad writing” and “unpopular writing” are not the same thing. What usually does best on online platforms is writing with a strong moral message, particularly one that is identity-affirming for the reader. Another very popular type of content is stuff that over-simplifies a complex topic or puts forward a stark binary - they bad, us good - view of the world. These are all things AI can produce a lot of, quickly.

So I think any tool that allows us to see when AI is being used to produce writing is a good thing. If people want to read AI-produced writing then they absolutely should. But I think it’s vital to maintain a niche of stuff written purely by humans.

My goal at America Explained has never been to be the biggest newsletter on Substack. In fact, I deliberately make choices that have the effect of keeping America Explained smaller. I try not to sensationalize or provide simplified us vs. them views of the world. I don’t get mean or too polarizing. I specialize in explanations that some people probably find kind of boring, particularly if what they come on the internet to do is have their views and prejudices affirmed.

All of this probably limits my growth, but it helps me sleep at night and maintain my professionalism as a historian and commentator. I’m not really worried about AI taking over my niche - it’s not me, and it can’t write like me! The number of people who want to pay to read my writing was always going to be bounded anyway, and that’s fine. But I would like my niche to be still discoverable, rather than drowned in AI slop

Finally, I hope that younger authors don’t lean too hard into AI. Writing is a craft that you have to learn over decades, and if you’re leaning on AI to do it from the start then you’re not learning. You might be highly productive in terms of cranking things out, but you’ll hit a plateau of quality and never find your own voice. And having a voice is a precious, precious thing.

If you’d like to support this artisanal, human-made newsletter, please upgrade to a paid subscription - currently 30% off! And if you can’t afford it or don’t want to, thanks for reading anyway - I truly appreciate it.