Learning the minutiae of a particular piece of hardware isn't important. Learning to understand and debug a program's is very important. Unity or Godot still work on logic - if A causes B and C causes D but A and C doesn't cause B and D, there'll be a reason for that (even if that reason is just "a bug in the Unity implementation") and a programmer has a hope of learning what it is. Probabilistic tools don't have that, and I worry about the implications for learning. (For the same reason I've always been very skeptical about including tools like ElasticSearch or Splunk in a dataflow - even if they seem to work reliably, how can you be confident they will continue to do so in the future?)