Let them make their AI if we have to. Let them use it to cure cancer and whatever other disease, but I don't think we should be allowing it to be used for commercial purposes.
Public information and the ability for public to analyze, understand and eventually decide what's best for them is by and large the most relevant aspect. Your decisions are drastically different if you learn soemthing can or cannot be avoided.
You can't dissallow commercial purposes. You can't even realistically enforce property rights for illegal training data, but maybe you can argue that the totality of human knowledge should go towards the benefits of the humans, regardless of who organizes it.
However there's a lot that can be done like understanding the implications of the (close to) zero-sum game that's about to happen and whether they are solvable in the current framework and without a first principles approach.
Ultimately, it's a resource ownership and resource utilization efficiency game. Everyone's resource ownership can't be drastically change but their resource efficiency utilization can as long as the implications are made clear.