How would you design a better social network?
Platforms are not neutral: they sort attention, and attention goes spontaneously to whatever generates strong emotions. Designing better means changing what is rewarded.
Four principles emerge from what we know and what has been tested.
Reward what connects rather than what divides. Pol.is provides the clearest demonstration: instead of surfacing whatever provokes the most reactions, the algorithm surfaces the statements that bring together otherwise opposed groups. Taiwan used it at national scale on subjects reputed to be deadlocked.
Correct the distorting mirror. A very active minority produces most of the visible content and appears to represent its camp. Making the real distribution of opinions visible reduces hostility measurably, because our meta-perceptions are wrong in the direction of the worst.
Restore the context. Most online conflict arises from statements taken out of context, where assuming bad intent becomes automatic.
Slow down the tipping moments. Minimal friction before sharing or replying in the heat of the moment has measured effects, without censoring anything.
One warning that applies to all of the above: Goodhart’s law. As soon as a measure of quality becomes a target, it stops measuring quality.