That is a tightrope to walk. Or as a friend tweeted: “Awwwkwaaard!”
The problem that Facebook faces, as we have discussed in many previous issues, is to reconcile its role as dominant media platform with its objective of serving engaged users to advertisers.
Engagement (which ultimately drives strategic financial value) is easy to enumerate and target.
Truth, diversity, serendipity, civil discourse, reason, empathy, all things we might want our civic discourse to manifest are just much harder to make explicit. And so they are much harder to build. And it’s not clear they lead to more dollars from Unilever, American Express or Nestle compared to the current strategy of optimising content to drive user engagement.
Short-term incentives are misaligned with the long-term benefits of civil society. Worse, the gains are taking by the producers, in this case, Facebook’s owners, while the costs are borne by society. (Seem familiar?)
Could Facebook tune the newsfeed another way? Yes, of course, it could. It could optimise its newsfeed by allocating more weight to a user reading a story from sources that had broad and diverse (and long-standing) trust signals. Or it could reward the algorithm for showing me some stories on topics read by people on the other side of the interest graph from me. Or liked by people who sometimes, but not always, disagreed with me. Or it could put greater weight on items read or promoted by people based on their Scientific h-index
Facebook chooses not to. Is that the right decision or the wrong one? We can discuss that. But what seems certain is that these choices are Facebook’s to make.
Note: PeerIndex built an algorithmic curation engine back in 2009 to its acquisition in 2014 and we contended with many of the issues of trust, credibility, filter bubbles, serendipity and discovery. We learnt that sorting signals across 300m+ users per month was not an easy technical task, but that it was possible. But ultimately meta-editorial judgement was required to determine our specific technical implementations (what objective you target, what features you evaluate, how you weight them and what errors you allow.)