Twitter's unacceptable hidden defect

image: github analysis of Twitter follower network (link)

Twitter's paroxysms in the past two weeks have been unsettling. But suddenly, I am coming to believe that many of us have misunderstood what we were getting into when we got involved in Twitter in the first place. We had made assumptions about the advantages that Twitter could bring to us and maybe to our academic and cultural communities -- unfettered ability to hear what a fairly lengthy list of "people to follow" and "people who follow" have to say about subjects that the individual user is interested in. We users thought the platform would offer a stimulating sharing of ideas, with a bit of synergy -- new perspectives on old issues and topics. Stuff comes in, we think about it, and stuff goes out. The platform doesn't "make" the message.

But let's look at the situation more closely. Our initial assumptions about Twitter presupposed a process: Person A follows {F1, F2, F3, ..., Fn} and is followed by {G1, G2, G3, ..., Gm}. The platform is simply a conveyor: it serves up messages from source to followers, all through the network of initiators and followers. Whenever any of the Fi people tweet, person A gets a chance to read the message and interact with it. And anything that A tweets is seen by everyone in the follower list Gi. There is usually some overlap between the F's and G's, but they aren't usually exactly the same sets of people.

But notice that this summary of the Twitter process makes an assumption: that the platform is an automatic, neutral, and mechanical server; it simply delivers messages from A to all of Gi, and it delivers all the messages created by the Fi crowd to A. (Whether A is overwhelmed by the volume of messages is a different matter; that's why it makes sense to have a reasonably small set of F's, so that it is possible to pay attention to the messages originating from the people one follows.)

What we are now learning, however, is that the platform (Twitter) is not neutral and automatic. Here I'm focusing on the personalized feed that each Twitter user receives through the app. There are two settings on the HOME / WHAT'S HAPPENING? tab. The user can stick with the default, the curated list, the "top Tweets first" list; or the user can select "Latest Tweets". The implication is that the second choice gives the user an unprocessed feed from all the accounts he/she follows, in reverse chronological order. But my own experience in the past week or so indicates that this is not the case today, if it ever was. Today, for example, 23 of the first 25 tweets in my feed are from "blue-check" accounts; whereas fewer than 10% of the people I follow have blue-check accounts. So blue-check items are vastly over-represented in my news feed, even on the "Latest Tweets" option. Twitter made a trial-balloon announcement last week that "premium accounts" will receive priority in the news feed, and it appears unavoidable that this has been implemented. And that is flatly unacceptable to me.

It is unacceptable for two reasons. First, Twitter is a communications system for me; and the system should not decide which messages I get to see -- anymore than the phone company should decide which phone calls to put through. I want to have routine, unbiased access to the tweets published by the people I follow; I don't want those messages to be buried at the end of a stream of several hundred messages that have been given priority. The whole value of having a Twitter account is having direct access to the ideas, observations, and messages of these people whose opinions I respect, and I want to have an unbiased access to those messages.

But second, the implications of a "curated" feed are quite horrible when you think them through. Any sort of bias can be built into the curation algorithm, emphasizing one kind of message over another, and building a "thought world" for the individual user that is the construction of the algorithm. Like counting votes, the only way to avoid that bias is to mechanically serve up the messages in the order in which they are published. I had long presupposed that this was the way that the feed worked. But plainly it does not work that way today -- even on the seemingly "automatic" setting of "most recent".

This is a deficiency that we probably recognized more readily in the case of Facebook, where the Facebook news feed is plainly a selective "curated" list of items drawn from the agents one has "friended" and other news sources. But on Twitter, this curation bias wasn't evident to me until this week. The reality of how the system works is important: by subscribing to Twitter and reading or scanning the news feed, we are giving the platform an incredible amount of discretion in deciding what we see and what we don't see. Right now the selection algorithm seems to be centered on Twitter's effort to incentivize users to select a paid plan (Twitter Blue, blue check, verified), by offering the advantage that the paid plan messages will get priority in the feed. That all by itself is unacceptable to me as a user, because it opens up the possibility that a disproportion of the messages that I receive are boring and irrelevant. I don't want more messages from Elon simply because he's a blue check user. But the selectivity can be even more harmful than that, since an algorithmic feed can be tuned to political purposes as well. For example, we could imagine an algorithm that gives priority to messages casting doubt on the value of US support for Ukraine and low priority to messages that emphasize the importance of US support for Ukraine.

Consider a fairly dystopian fantasy that sheds light on what I'm getting at. What if a new mass email provider offered a new service. "We will deliver your email in nano-seconds; but even better, we will automatically correct the spelling and grammar, and we will screen incoming and outgoing messages for statements you might later regret." This would be algorithmic "processing" of email communications. And it would be nightmarish. I say nightmarish, because when I send a message, I formulate it in the way that best expresses my meaning and intentions; it is the job of the email carrier to blindly and neutrally deliver the message without algorithmic review and without editing. In a small way, Twitter is messing with my social communications -- both incoming and outgoing -- by imposing an algorithmic "prioritizing" weight on different messages that means that some messages sent in my direction have a much greater probability of being read by me than others. That is a kind of soft censorship. In the present case it is censorship based on "membership status"; but the same routine could prioritize messages according to a measure of their place on the political spectrum. Not good.

I began this post by saying we didn't really understand what we were getting into when we joined Twitter -- thirteen years ago in my case. We thought, without reflection, that it was simply a cool communications platform, a basis for communicating with individuals and groups in many parts of the world and in many different disciplines. But upon reflection, it is something different than that. It is a system in which we have given up control of who sees our messages -- and when -- and which messages from our interest group we are able to see. Those determinations are being made by the algorithm, not by a simple "first-in, first-out" process that guarantees that every message will be delivered. And the algorithm is tweaked according to the business and political interests of the corporation. And Elon Musk has made it plain this week that he is entirely OK with using Twitter's system for his own political purposes. That is not acceptable.