I will say, we are so so so much worse at this right now than people think
For all the talk of social media manipulation and nudges, behavioral economics and gamification theories are so crude, there are major replication crises of studies, most of the touted "outsized" effects (where minor interventions yield large behavioral effects) are basically bullshit.
ML systems that learn through experimentation will probably become pretty good though, and likely are already outperforming our academic theories.
What I think we'll end up with is a place where we know, given a certain context, based on ML training, what works. But we won't know why. We won't have grand theories, just a set of super contextually specific interventions that are effective.