A wild guess, but a lot of the advances in neural networks can be directly applied to control problems normally be handled by PID controllers. It will be interesting to see where this leads
A lot of big industrial processes have a ton of different control loops that try to optimize parameters. The systems are usually complicated enough that the ideal state for these parameters is unknown, so controllers try to approximate.
An easy example is the engine controller in cars. It tries to optimize for power, efficiency, emissions, engine longevity, etc...
Machine learning could crunch tons of historical data from these systems to build a CNN that performs better than the usual controller.
In some systems, like stoplight timings or freight packing.... An AI driven controller could possibly be revolutionary, performing orders of magnitude better than current algorithms.