My idea is that you should combine ML with something that is highly reliant on human labor--especially where that reliance is starting to be the key bottleneck

My idea is that you should combine ML with something that is highly reliant on human labor--especially where that reliance is starting to be the key bottleneck preventing new capacity from starting up. A good example of this would be slaughterhouses in the US; the pork processing business for example. Get some experienced workers from an existing facility to wear google glass type cameras, put sensors on the base and tip of their electric knives, and project a grid of red laser dots on the carcasses, and record the process of carving them up from start to finish many times with many different carcasses. Get enough labeled data that way and you could replace human workers in a dirty/dangerous/low-paid industry.