That is definitely not inevitable. It could very well be that we reach a point of diminishing returns soon. I'm not convinced, that the simplistic construction of current generation machine learning can go much further than it already has without significant changes in strategy.