Bayesian analysis is “boring” because all analyses have the same workflow: given prior and likelihood, calculate posterior. Then the loss function determines how to summarize the posterior. Maybe perform some posterior predictive checks.
Machine learning is “boring” because, given the data, you just partition into training and test and find the method that minimizes the test error. Or maximizes the accuracy1.
Frequentist statistics might be the only non-boring way to do analysis, you have to be really creative to find the best method for your problem.
Despite how bad of an idea this is, because accuracy is an improper scoring rule.↩