Maybe scaling personalized education is not that complicated.Basically you just create wide arrays of explanations to something , and automatically try to offer each student the explanations he will benefit most from. Yes it demands work on content creation , and it demand a way(automated, crowdsourced, by teacher) to offer each student recommendations on content. But the gains could be immense: offering great education globally.
After you solved this problem, offering social support becomes somewhat easier.Instead of the teacher focusing on lecturing to full classes, creating lesson plans, he can focus on working with each student individually and offering social support. That's the whole idea of reverse classroom:children listen to lectures at home, and do homework at school with help of the teacher.
But that's just one way of a scalable solution, but there are many startups working on finding the solution and building interesting things. I wouldn't bet against all of them.
>> Most students do experience long-term learning in these environments.
That's interesting, because that doesn't match the many positive reviews from students. That also doesn't match the fact that khan offers exercises to some students and see results improve.