What is the point of creating an AI that will learn on its own, if it is faster and cheaper to teach it right away?

Question

Artificial intelligence

I don’t understand why do so much for the AI ​​to learn, for example, to dig the ground with a shovel, if it can be taught right away and this is 900 times faster and 900 times cheaper? What's the point of this?
Well, if you dig deep (not the earth, but a question of intellect), then you mean very deep in-deep programming of a universal organism, but this is still the same set of conditions, actions within the conditions, and then an analysis of new changed conditions as a result of actions. These are three whales. But there is a fourth, semantic - strategic task. In the end, everything comes down (I repeat): to a set of conditions, actions within the conditions and analysis of new conditions as a result of previous actions, and then again a new cycle based on a strategic task.
What is the training itself? In search of 100500 ways? So inside each of the 100500 iterations there is already an algorithm for this very process. It turns out that AI is endless networks of networks of algorithms that are constantly replenished with new ones and, as a result, for the task assigned to it (AI), it (AI) will simply give out the most economical option at the moment (faster, smaller, easier, simpler)? Is this big brain just obtained with a bunch of unnecessary (in 99.999-9% of cases) logical pattern networks like that?

Answer 1

Is it cheaper and easier to teach how to solve problems that no one has solved before?
How is it cheaper and easier?
It's cheaper and easier to show how tasks are solved - and then he himself.
Give an analogy with a person - you want only those who need to ideally chew on all tasks and who, with a slight deviation from the task, fall into a stupor, or do you still want those who work for you to be at least a little independent in making decisions, so that you don't spend 200% of your time on them being their babysitter all the time?