A correlation in the "cancelling" (aka termination for violation of company policy) of outliers does not mean we won't "reach new plateaus". You can fire 100 people, and there may be correlations ("they only fired people who like ice cream") but no causation ("only people who like ice cream reach new plateaus"). There is no evidence to suggest a causative link between douchebaggery and innovation in AI research.

On the other hand, if our ability to "reach new plateaus" relied solely on promoting people with personality issues, we have much bigger problems.