If you’re going to avoid psychology, do it because of the replication crisis. What is being called “AI” should play no role on that. Here’s why.
Let us suppose for a moment that some AI 7y from now is able to accurately diagnose and treat psychological issues that someone might have. Even then the AI in question is not a moral agent that can be held responsible for its actions, and that is essential when you’re dealing with human lives. In other words you’ll still need psychologists picking the output of said AI and making informed decisions on what the patient should [not] do.
Furthermore, I do not think that those “AI systems” will be remotely as proficient at human tasks in, say, a decade, as some people are claiming that they will be. AI is a misnomer, those systems are not intelligent. Model-based text generators are a great example of that (and relevant in your case): play a bit with ChatGPT or Bard, and look at their output in a somewhat consistent way (without cherry picking the hits and ignoring the misses). Then you’ll notice that they don’t really understand anything - they’re reproducing grammatical patterns regardless of their content. (Just like they were programmed to.)