My idea is simple.
I guess that one can take advantage of machine learning in early prototype construction. The idea is to make the prototype smart from the very beginning. So for instance. If a mechanism is controlling pressure then replace that with a machine learned algorithm.
Then you have to figure out how the software can help you in improving your prototype. I wrote before that maybe machine learning can be used to recognize cause and effect. I think something like that could be used.
So for a 4d smart prototype it would then indicate how to self-evolve. Then you alternate between improving the prototype physically and running the tests.
// Per Lindholm