"An AI does not know human psychology. " I don't really agree. Even boring Facebook algorithms have proven to know people better than their own wifes (or husbands). Of course they also excel at maximizing peoples attention and compusion patterns. And that's just basic algorithms. With the right training GPT can do a lot more.
Also article's title says writers should not worry about AI but then mentions competing with AI won't be easy "If anything, it is extremely risky". Is it not better to worry about extremely risky things?
I do agree however with predictions, who might get hit the hardest etc. All of these seem reasonable.
An AI is neither sentient nor sapient. It cannot know anything about someone or something outside of its programming. What it can do is to gather input from a platform (i.e. someone's browsing behaviour on Facebook) and match it against models and data sets. If someone looks at X, he is probably interested in X. If someone looks at a bunch of related content, he probably possesses personality trait Y. If he does Z, he will probably do more of Z. That's how these models usually operate.
It takes a human to develop these models, and refine them.
In the same vein, an AI cannot natively maximise a person's attention and create compulsion patterns. What it can do is simply what it was programmed to do. It takes a human to understand the human mind, to determine how human attention and compulsion works, and then to develop a set of instructions for the AI to execute.
'It is extremely risky' refers to competing against AI-generated generic content on the basis of quality. Since the writing market does not care about quality, this is a risky strategy. To mitigate risk, you must do something else, or offer more than just quality.