From today everyone can read the code, a revolutionary move but totally ignored by the world press. Who should keep the distrust of social media?
- by Marco Hugo Barsotti8 April 2023, 5:55
- 1.8k Views 0 comments
Listen to the article
We all remember how the Cambridge Analytica scandal served as a pretext for the world’s press to explain Donald Trump’s unexpected victory in the 2016 presidential election: the electorate, especially the undecided, would be influenced and manipulated via Facebook, Twitter, Snap and other social networks.
True or not, all this will no longer be possible, at least not on Twitter, thanks to a decidedly revolutionary move and totally ignored (or perhaps misunderstood?) by the world press. On Friday, March 30, 2023, Elon Musk released the Twitter algorithm. We try to understand what this means, leaving at the end of the article some technical details reserved for the most adventurous among our readers.
Why did Trump win
As we know, the 2016 elections were to be won by Hillary Clinton, a progressive, female and politically experienced. Instead Trump prevailed: a hunt for the culprit was needed and this was identified in social networks, which obviously have the fault of not being controlled by traditional media.
What is the mechanism? Not so much the possibility of posting fake messages (concrete, as well explains the strip of the series Pearls to pigs by Stephan Pastis), but the fact that these are (or would be) systematically amplified and distributed to specific segments of the population, manipulating and inducing her to planned behavior by enemy forces.
From Pearls to Pigs by Stephan Pastis
Hard to share the accusation 100%. While it is actually possible - right up to this move by Musk - to manipulate social algorithms (some details here), more difficult to prove that this is enough to change a voting orientation. At least we’ve never seen scientific studies that prove it.
The point is that these algorithms, the systems that decide which posts, messages, photographs and videos are presented to us and in what order they have always been an absolute industrial secret. All those who propose methods or services to emerge from the mass, to "be first in Google searches" or "earn followers" or "maximize our visibility", do nothing but guesswork: they speculate how the system works and sell us their ramblings as if they were the Word.
But they cannot provide any evidence of what they claim. And often - especially those who do Instagram Marketing - perhaps they do not even have the knowledge necessary to understand what they are talking about, since it is necessary to have studied several recent programming languages and how neuronal networks work.
Algorithms are kept secret because they are considered the key to success (in terms of permanence time, called "engagement") of social media. And since it is in human nature to dwell on scandalous messages, extreme opinions, videos of cats or images of girls doing squats, these contents are the ones that most often are "served" and can serve as a trap (as the Cambridge Analytica case has shown).
The move of Musk
But Musk has another idea of how social media should work, or rather two. And he repeated them at least four times during the "Twitter Space" (a kind of online radio broadcast) that accompanied the publication of the algorithm: they are called "Trust through Transparency" (trust through transparency) and "Unregretted User Minutes" (no regrets about the minutes spent on Twitter).
Making the algorithm public gave us the key to understanding the rules used by Twitter to select what we see. A complex system, given that every day about 500 million messages are produced, while in a typical session we have time to consult maybe fifteen.
More importantly the second thing: since it is at the helm of society (a period that is defined without any proof "chaos at Twitter" by the New York Times and by the "democratic" press and podcasters), Elon is trying to modify it in order to propose content not of the type "click-bait" (designed to generate clicks), but of real interest for individual users so, precisely not to regret the time spent on the platform.
It is a long and complicated journey, as those who write complex software well know: you must first understand the code written by others (maybe... just fired) and often seemingly marginal changes cause unexpected problems or completely block the system. Not surprisingly, the entrepreneur has repeated that perhaps a restart from scratch will be necessary.
The answer is in the code
But the point is this: everyone can now read the source code, the instructions given to the system (note well: from the previous management to Musk and still active) to generate what we see. There are no mysteries, there are no "front page" gurus to hold: the answer is all here.
The traditional media do not talk about it - we think - because they are interested in keeping alive this kind of distrust towards social media, which would imply (in their opinion) a greater relevance of their dying newspapers.
If even Meta will follow the example of Musk we can finally go to an era where social media will offer us reliable news (see about the "Community Notes") and the revenues of position of the usual commentators, Those who sometimes do not even care to deepen the news about which pontificano, will finally set.
The algorithm of Twitter
We promised a short technical segment for the more daring: here it is. The algorithm is divided into three phases, partially (but not totally) based on AI (Artificial Intelligence). It takes into account the contents of two groups of users: In Network, those who follow or follow us and Out of Network, those whom we do not follow.
The first phase, called "Candidate Sources" takes into account in an equal way (50/50) tweets made by people we follow and by strangers, with the aim of selecting 1,500 (0.0003% of the total).
The messages of our contacts are selected based on the probability of an "engagement" (a reaction, such as a like, a comment or a retweet) by the individual user. A regression based on past reciprocal interactions is used. Those of whom we do not follow are instead chosen according to the "social graph", essentially the same criterion as before but applied "to friends of friends".
In the second phase, a numerical representation of individual user interests and tweet content is generated. This way you can calculate the affinity of the available tweets with our interests based on the distance between the two numbers.
In other words, if a person with whom I have common interests (say: listen to a press review called "the sects"), speaks of a cat named "Judith", then his tweet will appear prominently in my timeline.
If instead the same person always wrote about a cooking recipe then his message would appear equally, but very low. If finally to talk about the recipe was a famous chef who does not follow then his message would remain for me forever a mystery.
Order of arrival
Finally, the order in which the messages are presented to us is decided by a small neuronal network (in other words, an artificial intelligence) composed of "48 million parameters": by comparison ChatGPT has about 170 billion.
Finally, a heuristic filter is applied (a non-scientific method that uses common sense and simple rules) to further select the contents, avoid seeing too many by the same people and insert some advertising in the final mix. It’s as simple as that.