Yeah, this also reminded me of the way that Neural networks that play video games might make a new connection that is not super efficient, but by a combination of reasons ended up being more effective than it might usually be in that scenario.

The AI's often end up giving that pathway a try a few more times before giving up on it.