the deal for twitter buy by Elon Musk has come loaded with opinions, promises and proposals for improvement by the tycoon, and very few certainties.
Among the range of ideas that you have commented on if its acquisition is finalized and, therefore, the passage of the social network to be a private company —an unusual event in the history of the Internet on a platform with the magnitude and influence of Twitter— highlights that of “liberate” and open source your algorithm.
“People can read it and say, ‘Oh, I see a problem here, I don’t agree with this,'” Musk said, explaining his idea in a conversation with the founder of the well-known TED Talks, his first major interview after the deal. shopping. “They can highlight issues and suggest changes, the same way you update Linux or Signal.”
Analogous open source proposals already exist. Mastodon, the social network similar to Twitter to which some of its users have migrated after the purchase of Musk, already is. In the case of Twitter, this week we saw how – it is not known if as a kind of troll— internal by one of its employees, a folder was uploaded on GitHub that claimed to contain the algorithm.
But, What implications would releasing the platform’s recommendation and operation mechanisms have? ‘The Twitter algorithm’, which we will now see is a very simplistic approach, is what causes some tweets to go viral beyond community retweets. That we get recommendations of tweets from people we don’t follow, or how our time-line is organized if we have it in the default version, with which Twitter has been playing in recent years alternating it with the chronological version and now allowing you to choose between the two.
You can’t talk about a single “Twitter algorithm”
First of all, we must remember that Musk’s proposal is based on everyone being able to see the algorithm and correct possible biases it may contain. But that is a first condition. Not everyone, far from it, knows how to read code, and more complex, although without a doubt it would be a transparency mechanism.
What does not seem so clear is that it is as easy to execute as uploading a repository to GitHub. At least for now.
Early versions of social media platforms used to log users’ posts and comments in reverse chronological order, with the most recent appearing first in users’ feeds and timelines. This all changed in 2009, when Facebook began populating news feeds with content that the algorithm determined individual users would be most interested in seeing. The rest of the platforms followed suit and started testing their own versions, citing user experience as their motivation, as did Twitter.
The change is credited with fueling the growth of the platforms and has made them much more profitable, as the ads are easier to sell to engaged readers than bored scrollers.
As in other social networks, It is not known how Twitter is modeled, what visibility it gives according to what messages or according to what factorswhich is a problem that is attributed to all the networks by thinking that they can prioritize controversial or incendiary messages in some cases by generating more interaction or, on the contrary, limiting other conversations according to some guidelines that are not public.
Musk has positioned himself from the beginning with giving almost absolute freedom to say anything in pursuit of freedom of expression, although later he has been nuanced and has set the law as a barrier.
But the first problem is that there is no single algorithm that guides how Twitter decides to raise or not display content, unlike the simple version that Musk seems to offer, which, on the other hand, is surely aware of this complexity.
Besides, Twitter is known to use machine learning (machine learning) to guide many decisions. For example, Twitter trains numerous learning models to help decide which posts to prioritize in users’ feeds based on a dizzying number of factors. These models cannot be inspected like normal code; they should be tested in an environment that replicates the real world as closely as possible.
“You cannot open a machine learning model as if it were a bubble,” said Steve Teixeira, vice president of product at Twitter, who denied the biggest.
And being transparent with the algorithm is not everything
There are also experts who disagree that open source is the panacea in this case and that it can lead to a number of major problems. For example, it could give people with bad interests a better understanding of how to game the system and could fuel the proliferation of bots on the platform, just another of Musk’s crusades. “When something is made open and available to everyone, it can be misused”commented Arun Kumar, director of data and marketing technology at the advertising company Interpublic Group of Cos a Bloomberg.
Besides, a public algorithm does not prevent manual penalties which, in this case, could be given very vertically from the figure of Musk. The use of this type of action is already carried out by Google, for example, when a portal is positioned in its searches but shows content that is considered illegal or inappropriate but has managed to outwit to the algorithm.
There are also questions about how this could hurt Twitter commercially. Recommendation algorithms are proprietary and are the main intellectual property of platforms such as Twitter, Facebook and TikTok. Who would need to invest paying for more reach in a network where the elements that promote that reach are known? Although the latter, advertising investment, does not seem to be one of Musk’s biggest concerns, he has said that he prefers to promote the subscription model for Twitter Blue and other revenue streams.
Bluesky, Dorsey’s proposal that is also related
So: Is there any solution? Twitter could choose to release its most basic algorithm, or at the user or account level, so that the limiting and enhancing factors of the platform can be known for sure
Also that Musk will recover an old aspiration of his friend and former Twitter CEO Jack Dorsey: the Bluesky project. A model that, explained in a simple way, would work as a kind of marketplace for algorithms that developers could upload and contribute and users could select based on what type of Twitter they want to have. Perhaps a very basic one with hardly any algorithmic intervention, to others where the discovery or veto of certain topics is the basis in a much more refined and open way than the current recommendations of topics that the platform follows.
Bluesky was, in fact, a project thought to be extended to all social platforms, such as we explain you in hypertextual it’s been a while here.
We would love to give thanks to the author of this article for this outstanding web content
Elon ‘unleashing’ Twitter: the consequences of making his algorithm open source