There have been a number of systems both experimental and mainstream over the years that have tried to answer this question. One of the most sophisticated attempts in this respect is Slashdot with its system of "karma", thresholds, moderation and meta-moderation as possibly the best attempt so far at collaborative filtering.
At one point in the mid to late 1990's, Slashdot was the biggest and most successful forum of its kind, arguably at least partly due to the way it worked. But the complexity of the system was such that it may also have worked against it in some respects.
Perhaps a more successful attempt at an online space for civil discourse is Metafilter, which has concentrated on (and perhaps simply been lucky to have) fair and balanced moderation. As an attempt to create a non-polarising and empathetic online space, you could do worse than to consider their methods.
Other attempts have focused on issues such as ego. For example, 2Channel, at one point the biggest online forum in the world, enforced complete anonymity on the principle that if nobody knew who you were and there was no ability to build a reputation, you would be more likely to behave better.
In terms of design principles in the abstract, we might assume the main engine of polarisation and lack of empathy in online forums is the near total reliance participants have on written language. This, compounded by people's differing skills in written expression, leads to subtle misunderstandings which makes people tend to react in ways that in person they would probably not consider.
One way of countering this would be to have a forum that relied much more on voice communication (and perhaps also video). This came to pass with Discord and Twitter Spaces, for example. But this has issues with things like gender bias or indexing and discovery.
In the offline world, one of the most interesting examples of a technique that promoted empathy between people who might not otherwise have seen eye to eye was in post apartheid South Africa. Here, the use of "open space" was seen as successful in resolving issues between communities. This was before the online era, but it might provide some clues to better online systems, as might some ideas about human thinking and discourse in general, as reviewed here.
In short, while there are theoretical ways to enforce empathy, nobody has yet been generally successful in implementing them online. The various successes and failures in this area have also been studied by the likes of Clay Shirky and his writings on networks and culture, which are worth considering too. He feels we are (or at least have been) in a golden age of experimentation in the area of human discourse.