When ML surpasses human-level performance

Course Q&A Deep Learning Specialization Structuring Machine Learning Projects
1

Does that result in “updating” Bayes Optimal Error?

2

Yes, you should because it is the best possible error, and now you have proof that you can do better than human-level performance. Bayes optimal error can never be surpassed. Usually human-level performance a proxy for the bayes optimal error.