AI/ML - specifically Deep Bayesian approaches. The Deep neural network might have run its course, but Bayesian approaches (Probabilistic Graph Models) hold a lot of promise.
I think "run it's course" is a little needlessly pessimistic when there are still major strides being made each year in that area.
This comes off as pretty biased. "Run its course?" - really? Remind me, did DeepMind use Probabilistic Graphical Models to beat Lee Sedol at Go? Does any of OpenAI's research use them?