Rumors have been afloat that Twitter may be making a significant change to its service: moving away from the reverse-chronological timeline in favor of an algorithmically tuned news feed. And Zeynep Tufekci's critique of this prospect made the rounds, in waves, through my Twitter stream.
I must confess, my initial reading of Tufekci's article (as a recommender systems researcher and developer) was somewhat knee-jerk. I latched on to this statement:
An algorithm can perhaps surface guaranteed content, but it cannot surface unexpected, diverse and sometimes weird content exactly because of how algorithms work: they know what they already know.
This statement strikes me as overreaching in its claims. ‘Cannot’ is a strong claim to make with a high evidentiary bar, and I think we just don't know enough about the capabilities and limits of algorithms to capture user interest in order to say what they cannot do.
But I let an issue with a supporting statement color my view of the rest of the article, even though I agree with the immediately preceeding sentence (‘I honestly doubt that there is an algorithm in the world that can reliably surface such unexpected content, so well’). (You have a glimpse into how the sausage of my thoughts and opinions gets made, and sometimes it involves non-careful readings of an article that get corrected through vigorous argument with Jennifer.)
I believe strongly that, as researchers and developers of recommender systems (and pretty much any other technology), we must consider the how our systems will affect their users. This is a complicated process, because we often do not know in advance what will happen in response to a proposed development. Part of that process involves hearing critiques. This is why I enjoy reading Cyborgology, for example: it is a regular source of thoughtful criticism of various aspects of technology from a perspective and background that doesn't show up a lot in much of my other reading.
Upon a closer reading, I now agree with Tufekci's article on several key points:
Recommending diverse users to follow is a promising, underdeveloped place for algorithmic enhancement of the Twitter experience.
The cost/utility functions for which the algorithms are likely to optimize may well be at odds with what users want from the system. Building good optimization goals is very hard, doubly so when money gets involved. I think that many tech companies — including, to a large extent, Twitter — understand that their long-term interests are best served by meeting their users' needs and keeping the satisfied with the service, but even without the temptation to let short-term business concerns override user experience, the ways in which user satisfaction is measured will themselves introduce certain biases and reward particular behaviors. ‘Contribution to the Quality of the World’ is probably an uncomputable function. And Twitter has been trending more and more towards centralized control and being a ‘media company’.
An increase in influence inequality is a likely (though not inevitable, if Twitter decides the algorithm should favor less-influential or rising users) outcome of tweet-level filtering.
The specific example she uses to illustrate her concerns, identifying the importance of Keith Urbahn's tweet about Osama bin Laden, is something that would be incredibly difficult for any algorithm to do. It depends on knowledge that requires a substantial feat of AI. The algorithm would have to know Urbahn's historical connections to the subject matter of the tweet; this is not necessarily impossible, especially since his bio contains the necessary social graph data, but is very, very difficult.
The inscrutability — and gameability — of the algorithm change the fundamental logic of Twitter, and do so in a way that substantially alters the relationships between users and their followers/followees.
This last point is the core of my personal objections to the idea of a filtered Twitter feed. Currently, the basic structure of Twitter is simple and, more importantly, transparent: if you follow someone, you see their tweets in their timeline. If you post something, all your followers who are looking at the relevant time window of their feed will see what you posted.
This is the crux of Tufekci's penultimate paragraph: following someone is an expression of trust in their judgement. It says ‘I want to read what you have to say’. A filtered feed would change that to ‘I want a recommender to select some things from your posts for me to read’. There are times when I want that — if I've been away from Twitter for a few days and have a substantial backlog, getting just the highlights would be nice — but making that the primary way to interact with Twitter would be a substantial change to the service. It would make the service fundamentally different, in ways that both her and I expect we would find less useful than the Twitter that we currently use. I also think suppressing some of the messages users want to send to their followers would not respect the user's voice, violating one of Twitter's core principles.
Recommender systems are an important and powerful tool, with the potential to improve the Twitter experience in a number of ways that aren't yet fully realized. But do we need to filter the feed?
There is also an issue, that needs more study in the recommender systems literature, of positive versus negative filtering. Asking a recommender to find me thing, such as movies or people to follow, seems like a different question than asking a recommender to automatically filter out less-important things from my feed. The type and level of trust the user needs to have in the algorithm may differ, and a greater level of transparency may be needed, in order to avoid ‘what am I missing?’ problems. Assuming the users even know they are seeing a filtered view (which is a crucial transparency point in itself).
It may be that I am in a small minority of opinionated users, and that a filtered feed would be a better experience for many users. In which case, if it starts having filtered feeds as a primary feature, Twitter may not be as worthwhile of a service for me. But the judgement that it is better — and the criteria by which that betterness is determined — are worthy of close scrutiny.