Feed for good? On the effects of personalization algorithms in social platforms

This paper builds a theoretical model of communication and learning on a social media platform, and describes the algorithm an engagement-maximizing platform implements in equilibrium. Such algorithm excessively exploits similarity, locking users in echo chambers. Moreover, learning vanishes as platform size grows large. As this is far from ideal, we explore alternatives. The reverse-chronological algorithm the DSA mandated to reincorporate turns out to be not good enough, so we build the "breaking echo chambers" algorithm, a modification of the platform-optimal algorithm that improves learning by promoting opposite thoughts. Additionally, we seek a natural implementation path for the utilitarian optimal algorithm. This is why we advocate for horizontal interoperability, which interoperability compels platforms to compete based on algorithms. In the absence of platform-specific network effects that entrench users within dominant platforms, the retention of user bases hinges on implementing algorithms that outperform those of competitors.