My bachelor’s thesis was about determining the author based on a short paragraph. The idea was to compare word usage statistics against other authors’ texts. Surprisingly, even this simple idea worked quite well. Readers use more than word count: patterns, sentence structure, ways of telling a story. It makes it even easier to feel the style.
AI averages style, so it’s hard to see the author behind it (even with various prompt tricks). Empathy becomes harder without the style. When you know the writer personally, it becomes harder to see the text from the author’s perspective if you know it’s not her style. But even if you have never met a person, non-AI-written text tells you something about this person and their context. Ironically, with AI, context is being lost.
Beyond style, AI-generated text loses precision. Even given a detailed prompt, AI needs to make assumptions while writing to stitch a text together. It’s the same when you need to write a text for another person. This often means choosing more ambiguous, less precise language to avoid logical mistakes. For the writer, it can be hard to spot, as they have a clear idea in their mind, which gives an ambiguous word an exact meaning.
Another source of losing precision is spending less time and effort challenging our own ideas. Putting things on paper usually helps us see inconsistencies and gaps in our judgement, which are easy to miss when we are just thinking about them. AI writing takes away this opportunity, as you jump directly to a result.
The loss of precision and style affects reading speed. You need to switch ambiguous words for clearer ones while reading, and you have less information about the context to do so, as it is lost due to the lack of style. Personally, I feel that I am at least a third slower reading AI-generated text, even if the length is the same. Beyond this, most models are extremely verbose, which makes total reading time much higher. You do not need many readers for the writer’s time savings to evaporate on the readers’ side. This is even before you take second-order effects into account: more questions, more iterations (if it’s a collaborative text), and less trust between readers and the writer.
Was the same argument made against dictionaries and the standardisation of spelling?
Standardisation reduced variation, but made writing more interoperable. And LLMs can be remarkably useful for close reading: interrogating claims, checking evidence and exposing gaps in an argument. The claim that AI removes the intellectual work involved in writing seems too broad.