Correct!

However, this is just an analogy. In reality programmers simulate "reward" and "punishment" using different techniques that effectively result in the same thing happening. In neural networks, those are represented in weights in biases, for example.

Yea, this. GOTO is not neccessarily a bad thing. It's a thing. More popular codebases (mostly lower level) use GOTOs than people think. It can be handy in rare cases when you are in control and using GOTO won't actually hurt readability / predictability.

It's just that it's really easy to fuck up when using GOTO so you might as well just not at all (especially when you are not an expert on how things tend to work with it).

Same can be applied to other commonly accepted opinions, like using "eval"-type commands.