Industrial requirements for ML application technology

Given the fact that machine learning research exists almost half a century by now one may ask the question what the industrial value is of the things we have done. Well there is good news and bad news. The good news is that the research is flourishing. We have found a number of robust simple techniques that really seem to solve a class of interesting learning problems. In the next decade applications of ML will abound: intelligent cars, houses, wordprocessors, selfhealing networks, adaptive systems it all seems within reach. There is also bad news. Especially from an industrial point of view. ML is facing problems that look very much like the one we had in AI about a decade ago. The market is shattered. We offer point solutions in a variety of industrial settings, but no strong ML industry seems to be emerging. Our technology is shallow, i.e. easy to copy by our competitors. Application areas are, like in AI, divers. No strong vendors of AI tools currently exist in the market. In my opinion no strong vendors will emerge in the ML market. Still, ML is a very powerfull technology and we must start to develop methodologies for the deployment of ML in an industrial context. In the following I will draw on my experience as an R&D manager over the past few years.