At most one will create an easily teachable AI they can utilize. But that’s most definitely not the same as training a NN by specifying input/layers, etc.
That's what I meant. Trying to get the already-trained AI to do exactly what you want by modifying the prompt could be thought of as a type of programming and could become a skill-set in itself.
I believe it will either be "fool-proof" and so intuitive as "Follow the wizard and click on the button you want to be clicked when this and that happens", or it won't happen.
I think the biggest obstacle standing between the general populace and computers is the amount of detail one has to provide for a task to become programmable. But it comes up in real life as well.
That ticks the boxes of deceptively simple interface and programs a few lines long.
It also won't require as much formal training and will be more intuitive and trial error based.
GPT-3 and DALL-E have given us a glimpse of what that will look like.