We could score comments with sentiment analysis. AI, though not necessarily LLMs, can do it very well. Depending on how child comments reflect on the parent, the parent could be scored for relevance, agreement, helpfulness, or any other metric. This would also fuzz the metric itself (karma), and make it hard to game. The Goodhart's law problem in social media (karma farming, ratio-ing, etc) could be solved.

Of course, that presents many other problems. A corporation using AI to steer what discourse is worthy of promotion and what isn't in social networks has already caused much harm. Many things can be said about it.

But as a thought experiment, I think it's an interesting one. Much good could be done, advertisers and financial interests permitting.