Any prediction is going to be speculative, and has a good chance of being out by a fair margin. But this is my current sentiment:
My expectation for this decade is that we will see relatively slow and steady progress. As new datasets become available, we will develop new narrow tools which will gradually be incorporated into clinical workflows (hopefully following rigorous validation). Those providing direct financial returns (often in the form of savings) will be adopted most readily.
I believe we will continue to see many new start-ups looking to apply machine learning to healthcare, but that there will be a high failure rate among them. Common causes for failure will be insufficient advantage gained from using machine learning, and insufficient high-quality data for further development of the minimum viable product. As ‘promising’ start-ups fail, the hype around AI will dip (although not disappear entirely). I don’t think it will be either an AI summer or AI winter… I’ve heard the term AI autumn thrown around.
I believe this lull will last until a new, non-deep learning area of ML comes to the forefront and a fresh round of potential arises.
Either way, I’m excited to see what there is for me to write about at the start of the 2030s…