I agree that it's a shame that many research results do not get to be "industrialized" and actually used, but also I feel like many research results are created in such a sterile way that they wouldn't be applicable to real world scenarios.

I think what we got really good at is "perceptive" ML, like speech and image recognition, and those things do see industry applications, like self-driving cars or voice assistants.

I'd be interested to know where you see unrealized potential.