When Elon Musk announced his intention to buy Twitter, he said he plans to “make the algorithms open source to increase trust.” Twitter has a little bit of a trust problem. Conservatives worry that they are being censored, while progressives worry that Twitter has become a vehicle for disinformation and abuse. Paradoxically, Twitter’s most avid users routinely turn to their followers for advice about everything from good restaurants in a new city to how to cope with personal crises. So some Twitter users trust their followers, but not the platform writ large. And because every user sees a different timeline, it’s difficult to know just how Twitter, the platform, shapes the flow of information. But the dream of a more open, more transparent Twitter — or any tech platform — isn't as simple as publishing some code on Github (Musk’s proposed destination for Twitter’s source code). “Just releasing the source code doesn’t necessarily tell you how the machine learning system is working in practice,” said Jeff Allen, a co-founder of Integrity Institute, a nonprofit network of tech integrity professionals. So-called “machine learning” algorithms behind a lot of AI are essentially complicated rules for making statistical inferences. In the case of Twitter, the algorithms look for patterns about what types of tweets have induced people to stick around by liking, retweeting, or otherwise engaging with a post, and then aim to show users more of the sorts of things that have interested in them in the past. But the algorithm itself is only part of the story — access to the data on which an algorithm is trained is also crucial to understanding the behavior of the system. “It’s only after machine learning models have been trained on the data that you can actually see what are the most important features and how the models actually work,” Allen said. However, answering questions about whether or not Twitter’s algorithms are in some way “biased” might become easier were they open source, especially since some of the underlying training data is already publicly accessible. For instance, much, though far from all, of Twitter’s archive, can be downloaded in bulk. And Congress is pushing to get more insight into platforms’ datasets. On Wednesday, a Senate Judiciary hearing will weigh whether platforms need to give outside researchers access to information on how they use the massive amounts of data they collect. If you combine open-source algorithms with a viewable dataset, it might become possible to start to make real determinations about fairness. If all that sounds complicated, it is: Anyone, or any agency, seriously interested in policing Twitter for fairness would still need formidable machine-learning skills. As critics and tech leaders spar over just how much of their platforms to open up to public scrutiny, another movement is arising: Decentralized social networks, which aim to build those sorts of communities of trust in a way that is not subject to central control. One open source Twitter alternative, called Mastodon, has benefited from Musk’s planned buyout of Twitter. The open source protocol makes it possible for different servers to talk to one another, but the variety of servers makes it possible for each one to tailor its rules to suit its members. Eugen Rochko, Mastodon’s founder, wrote in a 2018 blog post that it gives the power back to the user, who gets to pick their preferred regulator. Mastodon is tiny compared to Twitter — it has about 500,000 active monthly users, compared to 40 million. But 200,000 of those users joined in April alone. The decentralized site, started in Germany, is run by thousands of individual servers, each of which has its own content moderation policy. The network boasts: “Your feed is chronological, ad-free and non-algorithmic—you decide who you want to see!” One Mastodon user, Sasha Costanza-Chock, a media and design researcher, touts the benefits of a decentralized platform where "decisions about how we interact with each other aren't made by billionaires." But, as this Wired article points out , it’s unclear if Mastodon’s virtues can scale. Arguably it works as a space where freedom of expression and comity coexist only because it is small. The question facing Musk is now the same question that Facebook has wrestled with for many years: Can you really make a community the size of the whole world?
| 
