These data points soon fed into some other comments Musk made last week about apparent changes to the X algorithm. That if “far more credible, verified subscriber accounts” mute or block your account compared to those which like your posts, “your reach will decline significantly”. The tech mogul has always used the site to optimise what he calls “unregretted user minutes”. This is something economists would recognise as a minimax function: a “satisfice” (satisfy/suffice) equation. It’s an attempt to maximise something within a constraint, in this case maximising happiness subject to minimising unhappiness.

One way to do this would be by dense logical flowcharting. But as a good engineer, Musk is instead treating the problem as a black box — remaining agnostic as to what constitutes a good time, and what constitutes regret — and focusing on its outputs.

X already does this to some extent. The algorithm pulls up a selection of tweets from your network based on the probability you will interact with them. Different types of engagement are then “scored” differently, and this is fed back into your personalised feed. “Probability the user opens the tweet author profile and likes or replies to a tweet” has a weighting of 12. A simple “like” is weighted at 0.5. And “Probability the user replies to the tweet and this reply is engaged by the tweet author” rates at 75. In the negative category, “Probability the user will react negatively (requesting ‘show less often’ on the tweet or author, block or mute the tweet)” scores -74.

Via many such variables, a score is built for each tweet, and the feed orders the highest-ranked scores first. In that sense, down-regulating users who are frequently blocked is in line with the general trend. But this comes with its own complications. “High block and mute rates from verified users” implies that there might be accounts a user likes which are also widely blocked and muted by others, and which would as a result increasingly be sidelined. This will naturally happen on both sides of the political aisle.

Yet if fewer controversial accounts make it through the net, Musk risks stultifying his audience. It’s a dilemma, but one that may be unavoidable. The H-1B row has perfectly illustrated this. There were plenty of salient graphs about the visa regimes relative to GDP, as well as insider anecdotes from recruiters and software engineers about disastrous in-sourcing of Indian engineers. But there were also outright racist posts from those who just wanted an excuse to dunk on Indians.

Unlike Twitter’s previous owners, Musk’s view is that sensible adults can view things with which they disagree, and not become hypnotised by them. Yet in terms of his minimax function of unregretted user minutes, it might be that the site is approaching an upper bound, as accounts which pander to baser feelings receive a boost from the algorithm. Musk would be content if the ecosystem he entrained stopped neatly at the classical liberal station he prefers — but the train has its own momentum. He is now learning that to let a thousand flowers bloom is also to become a constant gardener.

---

Gavin Haynes is a journalist and former editor-at-large at Vice.

@gavhaynes

---