Saving Twitter—A Roundtable

I. Musk could improve Twitter, but the early signs are not encouraging

Bo Winegard is Associate Editor at Quillette. You can follow him on Twitter @Epoe187.

Elon Musk’s purchase of Twitter has excited many on the Right and antagonized many on the Left. On the Right, Musk offers hope of more freedom and less bias against conservatives. On the Left, he provokes fear of an anarchic platform rife with misinformation and “hate speech.”

Although the Left’s anxieties are probably exaggerated, the early returns from Musk’s tenure suggest that divisive provocation and self-righteous anger will play an even larger role in Twitter 2.0 than they did in its prior iteration. The most likely consequence will be to drive up engagement while further degrading discourse. Musk’s behavior in the last week has caused me to revise my (cautiously hopeful) expectations about the platform’s prospects under his stewardship.

Since taking over, Musk has engaged in a variety of trollish behaviors, employing the argot of the-very-online-Right and attacking their favorite targets. “My pronouns,” he announced to his 121 million followers, “are Prosecute/Fauci.” In other tweets, he has attacked the company’s former head of trust and safety, Yoel Roth, by appearing to suggest that Roth’s dissertation offered a defense of child sexualization. Unsurprisingly, this led to a flood of “groomer” tweets and accusations of pedophilia. It has since been reported that Roth has had to flee his own home.

Concurrently, Musk has released a curated tranche of documents detailing behind-the-scenes machinations at Twitter 1.0 to sympathetic journalists including Matt Taibbi and Bari Weiss. Many of the details cast the former administration in a deeply unflattering light and demonstrate that it was, as many had long suspected, biased against conservatives. The predictable—and surely intended—consequence has been to inflame outrage on the Right and a seige mentality on the Left.

As a public official, Anthony Fauci is a perfectly legitimate target for criticism. Twitter 1.0 appears to have engaged in unjustifiably hypocritical behavior and hidden a byzantine and arbitrary system of shadow banning which clearly merits debate. And people’s public dissertations are fair game for scrutiny, debate, and even denunciation. Nevertheless, the personally vindictive nature of Musk’s campaign and the breathless reporting of his chosen conduits suggest a vengeful agenda. This is an ominous development if one believes that anti-establishment nihilism and uncharitable attacks on political opponents can only exacerbate existing divisions in an already polarized society.

One of the most pernicious things about Twitter is the way it rewards self-righteous, inflammatory, and uncharitable behavior. A thoughtful, fair-minded, and informative tweet is less likely to be shared than one that elicits outrage and self-righteous indignation. People who agree with a tweet will become excited and share it. But people who disagree with that tweet will become angry and also share it, usually as evidence of the fatuousness and treachery of their opponents. Derision and “nitpicking” are popular, careful consideration is not; and users chasing engagement might not care why others see their tweets—attention, good or bad, is still attention.

If Musk is to curb the flood of angertainment on his platform, he will need principled wisdom and courage, since angertainment almost certainly drives up engagement, which is good for business, at least in the short term. So far, Musk has preferred to feed right-wing grievances and rile up the mob instead, and he has aggravated the problem by joining the fray as a cultural warrior. One does not need to agree with hyperbolic claims that Musk is some kind of proto-fascist to find this dismaying.

Under such circumstances, can an optimist retain hope for Twitter 2.0? Musk has expressed support for two reforms that might actually improve the platform if he decides to pursue them. First, Twitter should develop a system of user authentication. This does not mean that users could not tweet anonymously—they would still be free to choose a pseudonymous identity. But registration should require a series of steps designed to greatly reduce the number of bots and make nihilistic trolling more costly and rare.

Second, Twitter should update its policies and guidelines and apply them fairly. One of the most consistent complaints from the Right has been that the company’s moderation policies are disfigured by obvious political biases. Until now, most of Twitter’s staff have been politically left-leaning, and enforcement of its rules and terms of service have consequently reflected this imbalance. Conservatives were therefore punished more frequently and harshly than progressives.

Some have argued that this is because they have shared misinformation more often, but this claim is hard to assess since the very concept of “misinformation” is itself open to political bias. It is hard to take any definition of “misinformation” seriously that does not include risible claims that police systemically hunt black people or that men can get pregnant etc.

Twitter would not become a Wild West of “free speech,” since the result would be an infernal world of insults and trolling that would alienate too many users. Even Elon has recognized this. Some kind of moderation is therefore inevitable. But it should strive to be fair to most users, irrespective of political or ideological inclination. Furthermore, it should be predictable. From 2018–21, moderation and expulsion often appeared to be capricious, especially on controversial topics related to gender and race. Some conservatives simply refrained from tweeting about these issues because they did not understand the rules and were afraid of getting kicked off the platform.

But these changes will be ineffectual if Twitter fails to address the problem of perverse incentives. For the time being, Musk seem unlikely to do so since he having too much fun contributing to it. But were he so inclined, there are options. For example, Twitter has already experimented with the introduction of a “downvote” or “dislike” option so the algorithm can account for disapproval as well as endorsement.

This idea should be revisited—it could help to make polarizing tweets less visible and less likely to go viral if the ratio of likes to dislikes is low. Alternatively, a “useful” and “not useful” marker on tweets, in addition to the like and retweet options, could perform a similar function, discouraging deliberately unhelpful and provocative tweets, and encouraging more constructive discourse. Of course, the 280-character limit constrains nuance and complexity by its very nature. Nevertheless, “These are complicated issues. See my article here…” is vastly more edifying than “Democrats are groomers” or “Republicans are transphobes.”

In my more pessimistic moments, I have compared Twitter to cigarettes. It provides an immediate boost of self-gratification, but in the long run, it is a social carcinogen that corrupts discourse and sucks up time that could be better spent. In which case, the best solution is to quit, and Musk would best improve Twitter by destroying it. But it is still possible (although less likely than I originally hoped) that Musk will implement radical changes that make Twitter a more pleasant and constructive platform. This might require a short-term drop in engagement, though, because controversy and righteous rage are perversely appealing.

A less toxic Twitter will necessarily be a slower and less exciting Twitter. Musk’s apparent preference for rabble-rousing might impel him to continue taunting the establishment. But in the long run, what could be bolder and more innovative than reconfiguring incentives to produce a more edifying social-media platform that discourages trolling and encourages charitable engagement?

II. It’s about strategy

Jim Rutt is a tech entrepreneur, executive, and investor. He is also the former chairman of the Santa Fe Institute, and host of The Jim Rutt Show podcast. You can follow him on Twitter @jim_rutt.

If Elon Musk is to improve Twitter, he needs to be clear on his strategic intent and begin executing it soon. What is “strategic intent”? Tesla has the strategic intent to make electric vehicles that are reliable, long-range, and cool enough to sell at scale. SpaceX’s strategic intent is to reduce the cost of space launch via reusable boosters, modular design, and economies of scale. Twitter needs a comparably crisp and executable strategic intent. If Musk can get that right, he might improve Twitter and improve the world.

So what is Twitter’s strategic intent? So far, Musk’s moves suggest muddled thinking. At times, he behaves as though Twitter is a startup that can pivot its way to success. He has spoken of turning Twitter into a WeChat-style super-app that he calls “the X-App.” This would add a wide array of functionality like messaging, payments, e-commerce, a mini-app store, and more. But this would be a big departure from Twitter’s current business model, and transforming it in this way may not be feasible.

At other times, Musk acts as though he is a private-equity buyer intent on squeezing as much cash out of the company as quickly as possible. Terminating half the staff in the first week and jamming through a poorly thought-through $8-a-month blue-check system suggest such a priority. His recently announced goal of rapidly obtaining 50 percent of company revenue from subscriptions alone also looks like evidence of a cash-squeezing agenda.

At the strategic level, these two aims are incompatible. If Musk wants to pivot to an X-App model, he needs a mass audience, which calls for a focus on rapid (and likely free) user growth. If he wants to squeeze the business for cash via subscriptions, that implies a smaller, more focused audience. Pursuing these two goals together is incoherent. And neither opportunistic app-pivoting nor cash-squeezing strikes me as worthy of the attention of the richest guy in the world, whose amazing track record has relied on building and operating audacious, world-changing businesses.

Fortunately, Musk has shown some signs that he could be open to a world-changing strategic intent, after all. He should declare that his strategic intent for Twitter is to become humanity’s foremost platform for collective intelligence, and then relentlessly focus the activities of the Twitter team on realizing that goal. Such a strategic intent would not only be bold, it would also be consistent with Musk’s stated reason for buying Twitter: “to try to help humanity, whom I love.”

Humanity’s collective intelligence—our ability to make sense of the world, organize action, and make good collective decisions—is failing us. We are presently stumbling towards disaster: clown-show elections, an incompetent pandemic response, the unbridled and unthinking advance of exponential technologies like AGI and genetic engineering, an inability to devise realistic plans to address climate change, the return of nation-state warfare and the threat of a nuclear conflagration, and a mental-health crisis among young people. Not only is our current networked info-sphere not helping, it is part of the problem.

A collective-intelligence platform would be much more than a “town square” in which people can yell at each other. It would be a place in which people can self-organize into communities of interest, and where they can individually and collectively discover and curate quality information sources and reach conclusions. In such an ecosystem, quality thinkers could find their respective audiences, and teams of people could coalesce and collaborate on addressing the questions that might improve our lives. I’m not just talking about “big problems” like climate change or the risk of nuclear war—a good collective-intelligence platform can help us better address questions and problems at any scale: Should I quit my job? How do I fix the chronic water leak in my basement? What’s a good design for a 10-square-meter vegetable garden? Getting the right information and connecting to the right people at the right time will be vital.

Twitter is well placed to adopt this mission. Unlike most of its competitors, Twitter is already open-ended and self-organizing. This is a good start, but much more work is needed. A platform for collective intelligence would need to provide its userbase with a set of accessible network capabilities and behind-the-scenes processes that let us see high-quality information and ideas, post queries and questions in search of meaningful answers, work together to extend and improve what we see, and sometimes find the right teams of people to begin the process of putting ideas to work.

Twitter currently lacks the discovery capacity for us to find what we are interested in while curating for quality. Imagine a browsable diagram of users with similar interests to your own that highlights those who are most active, knowledgable, and connected. These could be color-coded with a measure of quality and importance that users can select and tune. Twitter could use machine learning and artificial intelligence to identify communities of users who are talking about the topics you follow.

“Twitter,” Musk has said, “needs to become by far the most accurate source of information about the world. That’s our mission.” Taken at face value, this is encouraging. Accurate information is critical to good collective intelligence. While Twitter is often very good at rapidly surfacing information, it is also erratic, with axe-grinders, trolls, and bots showing up for every significant news event. Community curation is one means of improving accuracy in a scalable way. Imagine a Twitterpedia (analogous to Wikipedia) in which a community of user-editors collectively makes sense of the incoming signals in near-real time, filtering out the junk and amplifying the good stuff. The curation of community-sourced information is a famously difficult problem. I hope that Musk makes solving it a priority.

Our attention is our most valuable resource but it is also limited. Focusing each user’s attention on what is good and important must be a key part of a collective-intelligence platform. Today, Twitter has weak tools for prioritizing what we see. There are just two available options for organizing our Twitter feeds: the default “Home” algorithm which shows us what it thinks we want to see, or the chronological “order posted” option that employs no filtering or prioritization. The first is imprecise, opaque, and vulnerable to manipulation, and the second is incredibly inefficient.

When Musk was first maneuvering to buy Twitter, he proposed open-sourcing the feed algorithm. That could help to build confidence that the feed is operating in good faith and accelerate its improvement. Better still would be the creation of a marketplace for open-sourced feed algorithms to which users could “subscribe” for a few dollars a month, yielding an ecosystem of innovative ways to manage our attention.

A useful feed algorithm idea would give users the ability to limit the number of tweets they see per day, and predictive algorithms could be used to select those tweets which the user would likely find most interesting or important. This could be made tuneable by button-click feedback. Another useful feature in a feed algorithm would be the addition of items I might not usually expect to see. Think of it as a dash of novelty. The novelty rate could increase when my feed is highly focused—a sort of an automated anti-bubble mechanism.

For any kind of serious collaboration or deep discussion, the current threading functionality on Twitter is incredibly bad. Better tools are needed to facilitate structured cooperative conversation. It would also be useful to develop tools for users to create links between conversations as a way for communities of interest to emerge and connect. The addition of a more flexible “Group” feature for creating membranes around communities of interest would also be helpful. Something more dynamic than what Reddit or Facebook currently offer. A micropayment or similar infrastructure would be great for letting us compensate valuable sources and collaborators. An ecosystem of good incentive signals can produce amazing results.

Identity on Twitter is its own can of worms but there is much room for improvement. Twitter should support varieties of authentication such as verified pseudonymous accounts and restrictions that permit only “one account per human.” Real-name accounts could be verified at lower levels by credit card, and at higher levels by bank-style Know-Your-Customer processes or even in-person verification by Network Notaries. Twitter should also support high-fidelity third-party identity services. To make identity really work for achieving collective intelligence, we need to be able to use the type and strength of identity to filter our feeds and searches. Such a system could be augmented by softer forms of community-based mutual vouching and rating for users and sources.

These are just a few examples of ways in which Twitter could move towards becoming a platform for human collective intelligence. There are plenty of other possibilities to research. My friend Jordan Hall has laid out a more radical plan that includes many other ideas worth considering. Other thinkers like Tristan Harris and Daniel Schmachtenberger are producing interesting ideas on collective intelligence that could be applicable to Twitter. Musk should reach out to the best thinkers in this domain and solicit their thoughts and suggestions.

Building a platform for achieving collective intelligence is a different challenge to those Musk has previously confronted. It is not at all like engineering a solution to a semi-static physics and economics problem, such as designing a rocket to get to Mars. The new discipline of emergent engineering provides some useful ways of thinking about these challenges. What’s required is an empirical and experimental mindset capable of acknowledging that emergences are powerful but not entirely predictable. It will require a smart incremental “feeling of the way” towards collective intelligence rather than a rigid engineering approach. I’d urge Musk to get his Twitter product people up to speed on emergent engineering.

Finally, it seems to me that Musk made a serious mistake by including so much debt in the acquisition capital structure. This likely caused some of the upheaval in the early days of his ownership, and it is not congruent with the strategic intent of building humanity’s collective-intelligence platform or any other audacious intent. If you look at Twitter’s cash flow statements for the last three years, it seems unwise to have loaded up the company with $13 billion of debt and more than $1 billion in annual interest payments. The company has never made vast amounts of cash and trying to generate that much free cashflow quickly is not in alignment with evolving Twitter in a direction to “help humanity.”

Enabling good collective intelligence will take time and require the addition of significant amounts of new talent and AI/ML capacity. I’d strongly suggest that Musk reconsider the capital structure of Twitter. He should replace the debt layer with equity. He’s the richest guy in the world, after all, so he can afford it. And if we believe the capital markets news, he could probably repurchase the debt right now at a substantial discount.

Whether Twitter will become a world-changing force for good by adopting a clear and powerful strategic intent is in Musk’s hands. He didn’t make his fortune by ignoring focused and audacious strategic intent. He shouldn’t make that mistake now.

III. There needs to be a shift in our internet architecture

Cody Moser is a PhD Student in Cognitive and Information Sciences at UC Merced. You can follow him on Twitter @LTF_01.

Seven days after he took over as CEO of Twitter, Elon Musk tweeted, “Because it consists of billions of bidirectional interactions per day, Twitter can be thought of as a collective, cybernetic super-intelligence.” To which, he added, “…with a lot of room for improvement.”

Much of the discussion about possible improvements to Twitter skews towards the internal regulation of its speech norms, the problems of marketing and revenue, and the performance and user-interface features of the site. An under-discussed option is a complete overhaul of the site’s recommender system and the structure of its information network. This could lead to a massive improvement of our online space. As a researcher in the field of collective intelligence, I entirely agree with Musk’s view that our “discourse” is, in essence, the product of an intelligent system, and that such an overhaul of Twitter would have great global benefits to both individual users and the website as a whole.

My argument here is counter-intuitive: contrary to received wisdom, less connected networks are better than more connected ones. Collective decision-making is, at bottom, a process of consensus-building. In a roundtable of 50 people, the loudest, most charismatic voice will dominate and squeeze the conversation to a single fixed point. But in 10 roundtables of five people, 10 very different conversations will develop, with unique ideas and consensuses.

In a less connected Twitter, then, independent cliques are shielded—at least somewhat—from the global cacophony, which allows them to dive deeper into the arcana of their specialist topics. Cheese fanatics can see tweets about cheese and not be bombarded with tweets about Ye’s appearance on InfoWars or Will Smith slapping Chris Rock at the Oscars. On a more integrated Twitter, on the other hand, a few loud voices dominate the conversation. Connectivity then becomes one of the drivers of polarization on the website, and as everyone converges on the same topics, the most antagonistic and controversial voices will dominate the conversation, forcing spectators to either disagree with someone’s horrible take on the situation or agree with the person antagonizing them.

When Twitter was founded in 2006, all tweets on a user’s timeline were displayed in chronological order (you can still switch to a less raw version of the website’s home timeline, with some limitations). In 2015, Twitter began employing a recommender system to show users “more relevant” tweets, which the website still refers to as “Top Tweets.” The purpose of this change was to improve engagement. By promoting Top Tweets, Twitter brought its users together in shared topic space. Consequently, the tweets appearing on your timeline are more similar to those seen by other users on the website, regardless of who users are following.

While this change brought individuals closer together in our spheres of interaction, the trade-offs have been mass polarization, the seeping of extremist beliefs into the discourse, and the disappearance of any serious specialist discussions online. These costs are all related to the increasing connectivity of the Internet as a whole. As our networks have become closer, the information to which the network is exposed has become less diverse, while disagreements have increased in number.

The similarity of the content to which users are exposed, irrespective of who they follow, has made the timelines between people who follow the same sets of people somewhat different. In the terminology of network science, the ability to form individual “cliques” or sub-units of the Internet’s broader network has gradually diminished. As networks have become larger and more connected, they have become more “efficient”—there are fewer links between a user and people further from that user than there were in years past, and less and less information flows between middlemen. People therefore have more direct access to each other’s information.

“More information” was one of the promises of the digital information age, and it seemed to be an obvious good that ought to be maximized. But models of collective intelligence have repeatedly shown that fully connected networks drive rapid consensus and decrease the diversity of information; less-connected networks, on the other hand, are slower but increase the diversity of information available to the network. The increasing connectivity of the Internet is therefore a serious problem. In its nascent days, the worldwide web served as a niche information-sharing device for hobbyists with particular, not general, interests.

In early episodes of The Simpsons, the web was a tool that the “Comic Book Guy” used to gossip about movie spoilers with other nerds, but that view is now antiquated. As of 2022, Reddit—a site that has capitalized, popularized, and fatally centralized nerd culture—has 52 million active users worldwide and is widely derided as a termite mound for plebian opinions. The increased connectivity of our social-media networks (and the Internet in general) has decreased the intelligence of our collective, cybernetic super-intelligences by increasing the speed of our conversations and decreasing exploration of alternative viewpoints.

The means by which we can improve the structure of our internet and bolster our collective intelligence are already available. Unfortunately, these require that we give up some of the critical means by which Twitter drives engagement, such as finding and forcing common interests and ending algorithmic augmentation as ways of bolstering the intelligence. Given the recommendation algorithm’s influence in shaping who Twitter’s users talk to and what they talk about, any serious proposal for improving its collective intelligence must begin there.

Several such proposals are available to Twitter. Engineers could introduce an algorithm which induces specialist echo chambers, thereby driving dissimilarity in timelines. They could introduce multiple timelines for people to switch between, similar to a Reverse List, producing an effect whereby an individual can communicate differently with different groups of people (producing a more diversified version of what we are seeing happening in Twitter Circles). Or Twitter could simply remove its computationally expensive recommender algorithm entirely and force everyone to adopt a chronological timeline to mimic pre-2015 Twitter, pre-2016 Instagram, and pre-2013 Facebook.

In each of these cases, the cliquishness of Twitter’s global network would increase, as there would be less crosstalk between individual users’ timelines and the formation of more specialist groups. In this way, we would bolster the diversity of independent conversations rather than encourage the intense, centralized cacophony exhibited by Twitter’s current network. While these networks would, for the most part, engage in independent exploration, this independent exploration of different problems means that when this information does come together, we will have had more epistemic diversity. Like the parable of the blind men and the elephant, we will have painted a more solid picture of a problem by first focusing on its component parts.

Whatever the solution, it needs to produce a shift in our internet architecture. The increasing centralization and connectivity of the Internet is paradoxically driving both rapid homogenization and polarization over issues which have not had the proper time to settle in the collective brain. The speed with which Twitter generates, elaborates, and spreads information makes it one of the most unique playing fields for testing our human collective intelligence. If Elon Musk takes his cybernetic vision seriously, this makes it the ideal place to begin with a new vision of how things could be.