Not an expert, but it's something to do with how they're made. Basically, most AI completely lack the ability to take in surrounding context, so an image generator doesn't know it's making a keyboard, it's just guessing what pixels come next. So for example, Chat GPT knows what letter is coming next, but it can't tell you what letter came before. I'd imagine this is a pretty easy fix tho. Why they haven't just given AI the ability to store context or use a calculator is beyond me, but they probably have that ability behind closed door.