nobody's in control if the approach to ML/AI is not fully known. we're still reinventing blackbox neural network architectures. training data is only curated to a finite degree. 5 years from now, all bets are off. 10 years from now, all bets made in 5 years are off.

I would be worried most about corporations like Tesla and Amazon that can afford the industrial scale to deploy robots for ML feedback and model refinement. here's a summary of what got covered at tesla AI day: https://www.youtube.com/watch?v=ABbDB6xri8o