This week, Elon Musk revealed that he has purchased a 9% stake in Twitter, and has joined the company’s board of directors. Media coverage has focused on Musk’s pro-free-speech views, which will probably shape how he tries to influence Twitter. But there are also many other ways that we might hope to tweak Twitter for the long-term benefit of humanity!
Purchasing a prestigious, tastemaking institution (like a social media site, newspaper, university, or scientific journal), has repeatedly been proposed as an “EA megaproject”. The common theme is:
1. We could buy influence over the “commanding heights of culture”, then use that influence to either:
2a. Directly promote the effective-altruist worldview, like by publishing EA-flavored newspaper editorials.
2b. Generally reform and improve the rationality/functioning of those institutions, like by improving the practices of a scientific journal. (As a neutral public platform, Twitter seems best suited for this approach, rather than direct EA promotion.)
Elon Musk seems sympathetic to effective altruism, so with him on Twitter’s board, we could consider Step 1 of a Twitter Megaproject partially accomplished, and get started on brainstorming specific potential reforms that Twitter could make. Personally, I think it would be cool for Twitter to add features that familiarize people with decisionmaking mechanisms like prediction markets and approval voting. But I’m sure there are other great ideas out there—I know there have been several rationalist efforts (including this very Forum!) to design social media sites that promote especially thoughtful, productive discussion. What’s your take on what Twitter could do for the long-term betterment of civilization?
Federated Censorship.
I’m very excited by the prospect of a more free speech freindly twitter, as I think their (and other firms) use of aggressive censorship has done a lot of harm over the last few years.
It would be good to achieve this while still offering more pro-censorship users the experience they want. This should make the change policy change less likely to be reversed, and help Twitter’s commercial success.
One method to achieve this might be to allow users to choose a form of censorship from a menu. Different groups would produce different censorship algorithms, and users would only see content that was compliant with their chosen algorithm(s). You could have a wide variety of different algorithms for different user preferences, e.g.:
ChildSafe, for parents to protect their kids from sex, swearing etc.
SFW, for people logging in on their work machines.
LGBTQIA+ Ally, blocking anything that seems bigoted.
NoSports, for those who don’t want to read about sports.
Libertopia, which allows everything.
Index Titiatum Prohibitorum, protecting Catholics from seeing unpopish things.
These filters could compete to produce the best user experience.
This way, people would be able to protect themselves from ‘toxic’ tweets without burdening others who want to learn and debate in an unencumbered manner.
Fwiw, there are tons of tweets that are genuinely toxic on Twitter—personal attacks, threats, and so on—that Twitter in my view currently does too little rather than too much about. Twitter may sometimes be too strict but the opposite issue is very much a problem as well. And I think that that view shouldn’t be called “pro-censorship”, which seems to me an unnecessarily value-laden term.
Twitter could implement a play-money prediction market just like metaculus or manifold markets—they could even consider buying one of these teams. Ideally, starting or voting on a prediction market would be as easy as running a Twitter poll. (Reddit recently did something similar.) Having large, metaculus-style prediction markets on newsworthy events might directly help important online conversations become more productive, more reality-based, and less polarized. And in the long run, familiarizing people with how prediction markets work might also encourage/legitimize the further adoption of prediction markets as information sources to inform decisionmaking.
Do we know why the Facebook prediction platform failed?
I personally don’t know (here is a not-super-informative article about it), although Facebook seemed to launch their project as a totally separate app unconnected to Facebook-the-website? I’d guess they simply struggled to attract users to the new app.
I think that one ultimate goal of prediction markets is to break out of the “enthusiast crowd” of rationalists and Phil Tetlock types who are excited about prediction markets for their own sake, and instead become a wider social norm among journalists, experts, politicians, etc, that if you aren’t willing to take a stand on prediction markets, you might be full of BS. The enthusiast crowd might flock to apps like Metaculus and Polymarket, but the wider “pundit crowd” might have to have prediction markets somewhat forced upon them, as something they must engage with if they want to build/maintain their reputation.
Since Twitter is already such a dominant platform for journalists, breaking news, political debates, and celebrity figures like politicians and founders, Twitter seems like a natural fit for this purpose. People already care deeply about Twitter metrics like their follower count and bluecheck status, much more than they care about karma points on a random new app like Forecast. If you could earn your way to a bluecheck by doing well enough on predictions, or if all the cool smart pundits showed their prediction score on their profile, that could provide a lot of motivation to get normal people thinking more rationally about political debates and current events.
Ideally it might be best if Twitter could try to focus people’s attention to a smaller number of higher-volume prediction markets in a more centralized, Metaculus-y way, rather than “everyone can start their own market” in the style of Facebook’s Forecast, Reddit’s Predictions, and Manifold Markets. For example, imagine if instead of fighting Covid “misinformation” by directing people towards a bunch of official CDC proclamations, they instead directed people towards a page more like Metaculus’ fortified essays, with CDC bullet points mixed in with live prediction markets on questions like “how much do the vaccines reduce covid risk?”. Similarly, Twitter could host high-volume markets on breaking news events like the invasion of Ukraine. But a big downside of this might be that by centralizing the prediction markets more, Twitter would face more political pressure and controversy. And of course it would be a difficult task to constantly create new markets on breaking-news topics.
Yeah I discuss this ToC a bit in this thread, under “Forecasting as a general epistemics intervention.” I do think it’s plausible that this is great, but I don’t (currently) think it is the highest impact outcome of forecasting, relative to talent recruitment, early warning forecasting, or prediction-evaluation setups to improve research or grantmaking.
Quartz wrote a good piece on it, can’t remember off the top of my head if it covers it: https://qz.com/2069284/facebook-is-shutting-down-its-experimental-app-forecast/
iirc, the project lead left for a startup.
It’s funny, I came with the same idea and implemented it in the social network I’m building https://nmkbs-aaaaa-aaaam-aadfa-cai.ic0.app/post/740#main.
I’m also planning to make it EA aligned at the core.
Hell yeah! Glad to see experiments with this kind of thing. I think one of the potential routes to impact here is (of course) becoming a kind of rationalist-twitter for EA discussion. Another path to impact worth keeping in mind is just that, by testing and proving out a bunch of innovations, your project could help influence actual Twitter (and other social media sites) to adopt similar features.
Twitter could create an easy-to-use, secure voting infrastructure for use by student groups, nonprofits, small businesses, unions, and other relatively low-stakes situations where you mostly just want to get a reasonably trustworthy voting system up and running easily. Twitter could use this platform to advertise the merits of designs like approval voting and quadratic voting, boosting interest in those types of voting and building legitimacy for them to be adopted in higher-stakes contexts.
We’re planning to test these kind of voting in our governance system.
Instead of the current, AI-based system of content moderation, Twitter could experiment with different methods of community governance and judicial review.
Imagine a system where AI auto-censorship decisions could be appealed by staking some karma-points on the odds that a community moderator would support the appeal if they reviewed it. Others could then stake their own karma points for or against, depending on how they thought the community moderator would rule. An actual community moderator would only have to be brought in for the most contentious cases where the betting markets are between, say, 30% and 70% -- this would make the system more scalable since most appeals would get resolved by the community without ever escalating to a moderator.
You could then have multiple levels of appeals and judges, creating another market on whether some kind of Twitter Supreme Court would uphold the moderator’s decision. (The above idea is ripped directly from Robin Hanson but I can’t find the exact post where he describes it. It also resembles the dispute-resolution mechanism of the UMA crypto coin.)
Making nuanced, human-based judgement scalable in this way could both directly improve the quality of twitter discourse, and help familiarize people with an innovative new social technology. Also, by creating a system of community governance instead of AI-based censorship, it might offer a superior middle path compared to the current “AI-based censorship vs 4chan anarchy” debates about social media content moderation.
The “Stratchery” newsletter proposes a sophisticated scheme to split Twitter into two companies, with core Twitter retaining control of the social graph and underlying infrastructure, but relaxing their control of the end-user’s UI experience, advertising, and content moderation. Those endpoint presentation services would be provided by numerous companies competing in the free market. Ultimately there is a vision for Twitter to evolve into essentially an internet standard for notifications, supporting many uses that sometimes look nothing like today’s Twitter.
I’m not sure if it would be altruistically good for the world to loosen control in this way and open up Twitter via APIs (although it would certainly help avoid undue censorship on the free exchange of ideas). But it’s an interesting analysis of Twitter’s business situation.
In our case we’re planning an open graph architecture, with users able to fork/spawn subnets.
A list of fixes requested by Tyler Cowen of Marginal Revolution:
1. A better organization of DMs, including functional search. And why do some of my DMs seem to disappear?
2. End-to-end encryption for DMs.
3. Available blue checks for more people.
4. Lately they have begun serving me up “popular” tweets from major tweeters multiple times. I hate this.
5. Eliminate the quote tweet function, to limit pile-ons.
6. Sometimes my “scroll down” function gets stuck. Unstick it.
7. I don’t myself prefer Promoted Tweets, and in theory yes I hate the bots. But in practice, viewed only selfishly, neither has been a major problem for me. I am not doubting they may be problems for others.
8. Longer-run, when AI is better and cheaper, how about a button “You didn’t subscribe to these tweets, but we think you really might like them.” But apart from the main flow and screen.
9. I wish for a slightly smarter list of trending topics. Yes I am greatly interested in the war in Ukraine, but I really don’t need “Zelenskyy says Russia is trying to hide ‘guilt in mass killing’ as the war in Ukraine continues”.
A year later, it seems like Elon does not value sane discussion on the Birdsite. If he does, he can’t change Twitter too much, because he is under a lot of pressure to make money. This is the fundamental problem with modern ad-supported social media—The business model is not aligned with users, and there’s not really an easy way of escaping this.
How do you evaluate community notes? Multiple times they have given me fairly informative context on some viral tweets, and it seems like they were introduced under Musk.
Community notes seem like a genuinely helpful improvement on the margin—but coming back to this post a year later, I would say that on net I am disappointed. (Disclaimer—I don’t use twitter much myself, so I can’t evaluate people’s claims of whether twitter’s culture has noticeably changed in a more free-speech direction or etc. From my point of view just occasionally reading others’ tweets, I don’t notice any change.)
During the lead-up to the purchase, people were speculating about all kinds of ways that Twitter could try to change its structure & business model, like this big idea that it could split apart the database from the user-interface, then allow multiple user-interfaces (vanilla Twitter plus third-party alternatives) to compete and use the database in different ways, including doing the federated censorship that Larks mentioned in his comment. The database would almost become the social version of what blockchains are for financial transactions—a kind of central repository of everything that everyone’s saying, which is then used and filtered and presented in many different ways.
But instead, the biggest change so far has been the introduction of a subscription model. Maybe this is just Step 1 of a larger process (gotta start by stabilizing the company and making it profitable)… but it seems like there is no larger vision for big changes/experiments like this. With a year of hindsight, it seems like Elon’s biggest concerns were just the sometimes aggressively left-wing moderation/norms of the site, and the way that the bluecheck system favored certain groups like journalists. It seems like now he’s fixed those perceived problems, but it hasn’t resulted in a transformative improvement to the platform, and there are simply no more steps in the plan.
So, that’s unfortunate. But I am still optimistic that Twitter is interested in experimenting and trying new things—even if there isn’t a concrete vision, I guess I am still optimistic that Twitter will eventually find its way to some of these interesting ideas via small-scale experimentation and iteration.
Community notes are great, even though they are (still?) restricted to the US. The good thing is that they seem to work fast enough so most tweet impressions are actually annotated.
Twitter needs to clamp down hard on disinformation and extremist content, especially from or on behalf of authoritarian regimes.
Categorizing one’s favorites (or putting them in folders) so one doesn’t have to scroll through them all to the beginning.
Twitter, as a fairly open platform, could be immensely valuable (because social networks are valuable) for humanity if it was more geared towards its users.
My sense is that people have felt for years that Twitter experiment with features that no one asked for rather than making it a nicer place for existing users. (Not sure how true this is, although I’d personally agree.) This is often noticeable in e.g. this way: https://twitter.com/scifiagenda/status/1328804296436006912 (See thread for interesting comments on how Twitter product design is an echo chamber of toxic positivity.)
If a social network of Twitter’s size and popularity could be slowly shifted towards an open feature development process informed by its users, it could start to grow into something else than what its most frequent users endearingly call a “hellscape”. At some point, it would be so usable and useful that its users might even consider paying (a small fee) to keep their account or create a new one. With such a change, the bot problem would probably be significantly reduced automatically. By then, a Stack Overflow-style reputation system is in place that allows self-moderation of the entire platform. Maybe it ends up truly enjoyable again.
Nothing about the above is too hard for capitalists to stomach, making it much more realistic than asking for hyper-niche features that only EAs and rationalists would like/use. Whatever you want to do to Twitter, there has to be a business case for it, sadly.
Yes, I rant about this regularly for some strange reason that still eludes me. On Twitter.
I made a similar question yesterday without being aware of this one https://forum.effectivealtruism.org/posts/ccZCJcqBtWxFLCN2t/how-does-an-ea-aligned-twitter-looks-like