There is a passage from Elon Musk’s interview at Ted2022 that few have noticed and highlighted. It is when Tesla’s CEO, speaking of the improvements he would have made to Twitter if he had had the opportunity to buy it (no, he has not yet succeeded) spoke of “opening the algorithm”. In other words, to make the code that hierarchizes the information on the open source platform, available to everyone for consultation. A novelty, this yes, which would be disruptive for the world of social networks. But is it really possible? To understand this, we need to take a few steps back.
Social
Why Elon Musk wants to buy Twitter
by Arcangelo Rociola
The algorithms of social networks and the magic formulas of our feeds
When we talk about social network algorithms or web platforms we refer to those systems called to organize our online experience. Basically, when we enter a digital service (from Twitter to Facebook to Netflix) what we see is ordered based on how much an algorithm deems that content suitable for us.
Each recommendation system (they are called so, to be precise) does exactly this: it studies the behavior of users on a given platform, compares them with those of all the others and, based on a series of criteria, selects the contents with which that specific person has a greater chance to interact. The work of the algorithms is oriented towards the achievement of objectives. In other words, there are content classification criteria, but it is the artificial intelligence, based on user data, that works out ways to get to the desired destination (usually, the increase in time spent on the platform). The formulas, if not for some indications, are secret. After all, they are a bit like the Coca-Cola recipe of digital companies.
Social network
Twitter has decided: it will use the Poison Pill to block Elon Musk’s takeover offer
by Emanuele Capone
Can algorithms be “opened”?
In summary, each algorithm has three fundamental components. First, the user data, which trains the machine and allows the AI to “know” individual people. Secondly, there are a number of content rating criteria: for example, on TikTok or YouTube a video is considered to be of greater value when viewed in full. Last, but not least, there is artificial intelligence, which allows the system to be dynamic, to adapt to circumstances and to any changes in users’ tastes.
When we talk about open source algorithms, we encounter at least two of these fundamental components. The first hurdle is artificial intelligence. Let’s imagine that, tomorrow, any social network publishes the formula of its online recommendation system. First question: who would be able to understand something? Few, no doubt. And even those who were able to move within those lines of code would not have certain answers on individual decisions. AI, in fact, is optimized to achieve a goal, but to do so, it can take unexpected paths. Twitter knows this well. In 2021, the bird’s social network released a study explaining how tweets from right-wing organizations received more distribution from the platform. Because? Twitter engineers don’t know, they don’t have an answer.
And then there is the data. Because making public the formula of a recommendation system would mean making public the data that is used to train it: billions of content from different users, their reactions, their viewing times of videos and posts. A move that, according to experts, would be suicide from the point of view of privacy.
On Twitter (almost) nobody wants Elon Musk to buy Twitter
by Pier Luca Santoro
The solutions, from the laws to the algorithm marketplace
In short, the open source of algorithms does not seem to be a viable path. Having said that, it is important to imagine alternative paths of transparency, which can make our relationship with the digital world more aware. In this sense, there is some legislative effort, such as the Chinese one, which aims at informing users on the use that platforms make of their data and on the prohibition, for the latter, to spread “addictive content”. Or the European approach, which aims to consider AI as a product that must satisfy a series of requirements based on the potential impact on the population.
While waiting for a not simple legislative intervention, starting from Musk’s words, a particularly interesting debate on the subject has opened. According to journalist Ryan Broderick, author of the Garbage Day newsletter, what platforms should be called upon to do is release a list of criteria on the basis of which content is evaluated and hierarchized. What if the algorithms were too complex to understand? “It shouldn’t be legal,” Broderick tweeted.
Another interesting proposal came from Nathan Baschez, co-founder of Every, a collective of journalists dealing with technology. Baschez wrote an article in which he proposes the possibility for users to choose the most suitable algorithm for them. A sort of marketplace for recommendation systems, starting from a suggestion that the then CEO of Twitter Jack Dorsey had already explored in 2021. Let’s imagine entering a social network and being able to choose our algorithm: do we prefer to see longer content? Or just videos? Or could they make us angry? According to this proposal, we would have a choice.