It created a gigantic procedurally generated training dataset using the formulaic rules of the game.
That doesn’t work irl.
Discussion
Why should it not work irl? It could try billions of possibilities to build better, faster airplanes for example. Could also easily find best marketing approaches by millions of parallel a/b tests etc.
Neither of those things are innovative.
There probably isn’t enough aircraft design data to train a reliable model. You might be able to take a 747 design and make incremental improvements to some components.
But it isn’t going to design a previously unseen hypersonic passenger jet with new engine technology.
The hardware necessary to support that kind of program doesn’t exist.
I've asked now GPT4. Let's see ehat the future brings 🙃
With the rapid advancements in artificial intelligence and machine learning, it is possible that AI could come up with new inventions in the future, including super-fast airplanes and other advanced technologies. By analyzing vast amounts of data from previous designs and simulations, AI could potentially develop innovative solutions that humans may not have thought of before.