Elon Musk once said he would make changes on Twitter when someone takes it over, and one of the things he suggested was to give people access to its code. Last week, Twitter allowed that to happen by posting their “For You” recommendations programming on GitHub.
Soon after Musk’s tweets were noticed, some people started examining its code. It didn’t take long for them to spot something amazing – a special ‘category’ was made just for Musk’s tweets! Twitter engineers admitted that this was meant for tracking stats, which other people also confirmed later on. However, the section of code related to this was removed quickly after being published. This still makes people think that maybe the company pays extra attention to Musk’s posts and has done something to boost his tweets even more.
Since Twitter released their code, not much about how the algorithm works has been revealed. And for those hoping that this code would show more details about how Twitter’s algorithm works will probably be disappointed. That’s because despite releasing their code, a lot of info was missing which experts who have looked at it noticed.
Twitter gave us a special code which was not complete, as noted by Sol Messing, an expert from NYU’s Center for Social Media and Politics who used to work at Twitter. This special code did not have all the information necessary to figure out how its recommendation system works.
Twitter is not giving anyone the code that deals with advertisements, and trust and safety systems because it doesn’t want people to abuse these features. The company said this in a post last week. More importantly, Twitter will not share the underlying models it uses to train its algorithm so that user privacy is guaranteed. This means that the most important part of the algorithm which nobody knows about yet remains unknown.
Elon Musk wanted companies like Twitter to make the algorithm used for censorship open source so that everyone could see any changes made. He believed that by doing this, people would be able to tell if the company was secretly altering anyone’s content or options on their platforms. This would prevent any behind-the-scenes manipulation from taking place. Musk stated this during an appearance at TED last April when he revealed his plans to take over the company.
Twitter recently released a code, but it doesn’t tell us very much about the potential bias or hidden things that Elon Musk wanted to expose. An expert called “Messing” said that although the code looks like transparency, it doesn’t tell us how the algorithm works and why some tweets get downrated while others get uprated.
Twitter recently changed the way their API works, which means that most researchers can’t access the data they need to do research properly. This makes it hard for them to figure out how the algorithm of Twitter works. As one researcher said, “Twitter released this code but at the same time, made it really hard for research to check its accuracy.”
Alex Hanna, who is in charge of research at the Distributed AI Research Institute (DAIR), had a chat with us last year after Elon Musk mentioned his plan to make Twitter’s algorithm available to public. Just like Messing, Alex was uncertain if releasing the codes on GitHub will really make it more obvious how Twitter works.
Hanna said that if you want to make sure something like a Twitter algorithm is properly watched then we need to use more than one technique so that it can be monitored.
The code found in GitHub gave us some new information about Twitter’s algorithm. Data scientist Jeff Allen discovered a kind of “formula” which tells us how much priority the algorithm gives to different types of engagement. According to this finding, a like (or “fav” on Twitter) is worth half a retweet. On the other hand, a reply is equivalent to 27 retweets, and if someone responds to the tweet from the original author…it’s going to count as 75 retweets!
It’s not easy to get a full understanding of the situation, but we still have an incomplete picture. Elon Musk recently made it really hard and really expensive for academics to study what’s going on, because if they want to do that now they need to be granted at least half a million dollars each year just for data.