There's definite possibility here - even potentially in marking and monitoring lessons. How much to "Automate" learning in general is. To put it crudely, many LMIC primary school education systems are based on rote learning. So one big question (and debate) in education circles is, should we then make that rote learning as effective as possible? Which is what orgs like Gates foundation funded bridge acadamies have tried to do.

https://www.bridgeinternationalacademies.com/

Or should we try and transform learning environments and teaching styles, so that classrooms are transfrormed into the kind of interactive and exploratory spaces we have in higher income countries?

I don't have a strong opinion on this, but lean towards the "improve the rote learning" in places like Uganda where i live, especially if the government isn't putting in a huge effort to transform education sustem

LLMs even right now could easily play a big role in improving rote learning, but I'm not sure they are at the stage yet to play much of a role in transforming classroom spaces - but that could come in the near future. 

Yes, this also came top of the ideas when we did a discussion on potential use cases for AI in education for LMIC (https://ai-for-education.org/working-group-discussion-ai-use-cases/)

There’s a few people trying this - my concern though, and something we just got a grant to think about, is how we make sure the content is good qualiTy. So we will start the year thinking about benchmarks for AI in education. 
 

lots of charity ideas here, and something we’re fortunate to get funding from BMGF and others to explore. 

Another reason I've been thinking about recently is the fact that most people forget most of the knowledge they learned in school, very soon after finishing school. I don't think there's a plausible mechanism by which this forgotten knowledge generates benefits for the individual or wider society.

It's quite likely imo that the primary intellectual benefit of school is not knowledge (easily forgotten) but the learned cognitive endurance that makes it easier to do cognitively demanding jobs later in life. Those jobs are also better paid, and they have larger benefits to society in terms of helping a country grow. If this is the main benefit, then negative spillovers likely won't be large, because they are taking jobs that less educated people couldn't do. Plus, noncognitive benefits of school, in terms of better socialization, are real.

I'm not aware of direct evidence on negative spillovers, but it's also important to point out that positive spillovers are also very plausible. More educated people can help their peers learn, not just in the classroom but also on the job. Plus if more educated people are better able to create successful businesses, then they create jobs for others in a positive-sum way.