There is a concept in philosophy called the Is-Ought gap, which is to say that you can’t tell what ought to be simply by looking at what currently is. This is something that applies to AI that I think is often overlooked.

A computer can tell me what my bank account IS, and a smart enough program could use supporting information to tell me what it WILL be or even MIGHT be… but it can never tell me what it OUGHT to be. Especially in a world without perfect interoperability (or with some things done offline/cash). Similarly a can tell me what IS on a monthly performance report, and in a steady and routine business environment it will tell me what MIGHT be useful on the report, it can never tell me what OUGHT to be on it