Turns out automating any work where is a specific desired outcome is really difficult because the type of AI we have today doesn’t really have any understanding of what it’s doing in any meaningful sense. This makes it pretty much impossible to guarantee that it will behave in a correct way when trying to accomplish a task. The cost of mistakes when applying AI in the real world is very high. A robot that does that wrong thing can cause a lot of physical damage and can even kill people.

On other hand, a lot of creative jobs are low risk. If an article has a typo or an illustration has some weird artifact that’s typically not a big deal. Worst case you can get a human to do a second pass to clean it up. This makes AI an attractive proposition for companies since it allows them to cut their staff and just keep a skeleton crew that manages the algorithm.