Comment by glimshe
1 day ago
I'll answer one of your points partially: if AI builds a better sorting algorithm and proves its performance characteristics, it's useful. I'd be able to use it to make my programs faster even if I don't/couldn't understand it.
It would be a bit disappointing but still useful and make humanity slightly better.
It's interesting to measure how much we believe in something, and how much trust we've lended in order to have a working model of our reality: enough understanding for us to get around, move about, and be content.
I'm sure you would only trust the improved 'blackbox' AI sorting algorithm after it has been proved out through benchmarks. Once you've seen better, repeatable numbers: your trust would rise and eventually you'd feel confident enough to use the blackbox in other areas of your application. You'd build on top of the trust you lended to the blackbox. And you would continue measuring yourself as you build out, making sure you trust the foundation as you go.
A proper engineering mindset if you ask me, but it's only useful in the physical world when solving physical problems.
The Mathematicians build 'castles in the sky' with vast equations that link up together in shapes that make sense. There's trust being lent to the linking as you go. How do you validate these 'castles in the sky'?
Through understanding. But then, how much understanding is needed? This is where Theory meets Application: and the Article is purely in the Theory territory. Your measure is purely in the Application territory.