People have been running social platforms and media companies for literal decades, after all, all while Musk was busy with cars and spaceships and whatnot. A fair number of lessons have been learned! But Musk — so resolutely convinced of his own genius — has dedicated himself to making old mistakes new again, compressing a lifetime of bad ideas into six short months.1 It’s his most reliable pattern: announce a crazy new policy, preferably on a weekend; face huge blowback from users; reverse the policy, claim you were misinterpreted all along or just pretend it never happened.

So when I saw this tweet on Saturday afternoon, I wasn’t sure whether to laugh or cry.

Rolling out next month, this platform will allow media publishers to charge users on a per article basis with one click.

This enables users who would not sign up for a monthly subscription to pay a higher per article price for when they want to read an occasional article.…

— Elon Musk (@elonmusk) April 29, 2023

Since embeds of his new longer-than-280-characters tweets don’t show the full text, here’s what it says:

Rolling out next month, this platform will allow media publishers to charge users on a per article basis with one click.

This enables users who would not sign up for a monthly subscription to pay a higher per article price for when they want to read an occasional article.

Should be a major win-win for both media orgs & the public.

Fiiiiiiiiiinally, Elon turns his attention to micropayments. (Pretty sure this is in the Book of Revelation somewhere.)

The idea of news publishers charging readers by the article is not a new one. At least once an hour, someone tweets about “why hasn’t anyone figured out how to let me buy one article????” Literally dozens of micropayments-for-news startups have come and gone; dozens of publishers have run tests of various models; none have gained much traction.

Even today, well into the 2020s, you can find people saying the dream is an “iTunes for news” that — as the iTunes Store did 20 years ago — allows you to buy a single song (an article) rather than the full album (a subscription). (They say this despite the fact that approximately zero people still buy MP3s that way; instead, they pay a monthly subscription fee to Spotify or Apple.)

I’ve long been a micropayments skeptic. Not because I have any philosophical issue with the idea; I’m all for publishers making money and readers consuming news. My skepticism is driven by it being a strategy that sounds appealing but works poorly in practice. Others have written about the problems with micropayments at great length, but here are, to my mind, the most significant:

Friction at the story level.

What do people do when they hit a news site’s paywall? We have some data on that question, from a Gallup/Knight Foundation survey last fall. They asked American adults: “Suppose you were trying to access a news story online and had to pay to keep reading or watching it. Which ONE of the following would you be most likely to do?”

48% said they would “try to access the information elsewhere for free from a different news outlet.” 28% would “move on to something else or to a different news story.” 7% would “try to find information about the news story on social media.” 4% would “sign up for a free trial if available.” 3% would try to “get the story through friends or family who already have access.”

A measly 1% would “pay for access to the story or outlet.”

In the overwhelming majority of cases, a person faced with the need to pay a news site money will say “no, thank you.” You can view that as an artifact of subscription models, or you can view it as evidence of how transient most news stories are in people’s information lives. It’s hard to evaluate how much an individual article is “worth” before you’ve actually consumed it — and there is always free competition available, either on the same topic or in the broader universe of “things to click on in my feed.”

Friction at the payment level.

If an individual publisher sets up their own micropayments system, getting money will require readers setting up an account, attaching a credit card, and all the usual stuff that moving money online requires. Not many people will do that to read a single news story.

So maybe they sign on to one of the many micropayment startups that want to create an industry-wide network of news sites using a common payment platform — either as part of a pan-publisher subscription or on a pay-per-article basis. Unfortunately, none of them have the scale to be appealing or the appeal to build scale. (“Just sign up with your NewzBux account!” isn’t much of a pitch to your readers if they’ve never heard of NewzBux, or InfoCents, or FactCoins, or whatever.) And the companies that might be able to start with scale (Google, Facebook) are not ones that publishers trust with their money. And whoever owns the pipes, they’ll want their 30% cut.

Most paywalls aren’t that hard.

In a digital universe where every news story is behind a hard paywall — one impenetrable to the non-paying reader — then a micropayments model might make sense. But that’s not the digital universe we live in. The number of completely paywalled sites is low and typically either hyperlocal (a county-seat weekly with no competition) or high-end (think The Information or Politico Pro). Nearly all news sites will let a random web user read a story (or two, or five) for free. It’s only after a given number of clicks that the wall goes up.

If you want to think of that as “news sites already offer micropayments for those first five articles — they’ve just set the price at $0,” be my guest. And for those times when someone really wants to read just one article, that free allotment allows all the paywall workarounds that the savvy digital news consumer knows about. (We’re all adults here; we can talk about incognito windows.) If most paywalls aren’t that hard, there’s little pressure for a paid product to get around them on a single story.

No one agrees on what micropayments are.

Is a micropayment 10 cents for one article? That was the number Elon Musk was thinking about in this video from November, when he complained that he should be able to pay 10 cents to read an especially good Philadelphia Inquirer story despite not living in Philadelphia.

.@elonmusk describes the issue of reading articles with paywalls. In clips never shown before in our interview he alludes to the value of what Twitter can bring. He also talks about how the media wants to own their subscribers. @kgee @EliBurton_ @klwtts pic.twitter.com/WsA2lUHo0S

— Tesla Owners Silicon Valley (@teslaownersSV) November 1, 2022

If there is a sustainable price for journalism, it isn’t 10 cents an article. A large scale data analysis from Medill found that digital news subscribers don’t even visit those news sites on most days. For small local news sites, the typical subscriber visits once every three days. At larger sites, it’s once every five days. Those visits can include consuming multiple articles, of course, but the point is 10 cents an article would be a radical price reduction for most subscribers — and thus a radical revenue reduction for most publishers. Price points will have to be higher — and thus less appealing to fly-by readers.

Publishers don’t want to cannibalize subscribers.

It’s not at all unusual for a business to insist on their product being purchased in a particular quantity. Try to go to the grocery store and buy one peanut M&M, or one tablespoon of ice cream, or a single Corn Flake. They’ll look at you funny, because the businesses that manufacture those consumer goods have been structured around selling bags, pints, and boxes of them, respectively. Go ask the people at Tesla if you can buy a Roadster that’s only for the weekends — at 2/7ths of the price. The economics of information goods (like news) aren’t identical to those of physical goods, but they both require sustainable business models, and for most quality news sites, that requires paid subscriptions.

And that’s the root problem, from publishers’ point of view: If you sell subscriptions for $15 a month, but you sell individual articles at 15 cents each, you’re telling any subscriber who reads less than 100 articles a month they’re an idiot and should give you less money. There aren’t enough payment-willing fly-by customers to make up the difference for even a few lost subscribers. You’re encouraging your best customers to think of you as an occasional treat rather than a service you pay for — and to pause before every headline they click to estimate its worth in cash. It shouldn’t be surprising than “we’ll charge you $10 a month until you tell us to stop” is more appealing than “we’ll charge you 10 cents now and maybe you’ll come back again someday.”

As Tony Haile once smartly put it, news subscriptions are like gym memberships. Imagine a gym that charges $50 a month for a membership — but also lets anyone pop in for a single workout for two bucks. Why would anyone pay for a membership again? “If you would take the micropayments version of a gym membership, it would be like, ‘I can turn up and I can pay a couple of quid, and I can go into the gym whenever I want to use it.’ No gym works like that.”

All that said — these problems are not insurmountable. Smart people might come up with solutions, even if they haven’t so far. Indeed, I’ve long believed that if anyone could create a micropayment system for news that worked, there were only two real possibilities: Apple and Twitter.

With iPhones, iPads, and Macs, Apple controls the devices that most paying digital news consumers use. They have hundreds of millions of users’ credit cards already on file and attached to your identity. And with Apple Pay, they have a nearly frictionless payment platform that has already been integrated into countless apps and websites. If they decided to offer a “Read With Apple Pay” button for news sites, the technical problems of micropayments would mostly go away. (Along with 30% of publishers’ revenue, no doubt.) And Apple News+ is the closest thing to an all=news subscription that currently exists.2

Twitter, meanwhile, is the center of the digital news universe. There is no place online with more news-curious users clicking links to new-to-them news sites. And it showed interest in the subject, buying Tony Haile’s Scroll and integrating its network of ad-free news sites into Twitter Blue and teasing some sort of paywall integration on the way.

But that was the old Twitter. One of Musk’s first decisions after taking charge was killing off the remnants of Scroll — the closest thing to a foundation for a pan-publisher revenue model anyone had.

Unless you are one of the few Twitter Blue subscribers, Twitter doesn’t have your credit card number. It has no ready payment platform for publishers to integrate into their sites. Twitter would likely only be interested in a payment system that goes through Twitter, not via links that go to a publisher site from Facebook, Google, or elsewhere.

But let’s be honest: The biggest problem is Elon. What mainstream publisher would trust Elon Musk with their money right now? The guy who refuses to pay the rent on his corporate HQ? The guy who has spent the past six months dumping on the media, banning reporters, declaring their work a “relentless hatestream” from “media puppet-masters” that you “cannot rely on…for truth“? This is the guy who says he has a “major win-win” for publishers? The same guy that complains “media is a click-machine, not a truth-machine” thinks the answer is tempting people to pay with a single headline?

(Not to mention that Musk has no deadline cred remaining, and saying that micropayments will “roll out” later this month could mean this summer, late 2024, or never.)

Maybe someone will figure out micropayments for news someday. I think it’s unlikely at scale — but I could be wrong! But I am quite confident the man who has spent the past half-year destroying the news media’s favorite online space won’t be the one to do it.

1. I believe it was Techdirt’s Mike Masnick I first saw using this metaphor for Musk, specifically around content moderation.
2. Pro tip: Apple News+ now includes, along with roughly all the magazines, The Wall Street Journal, the L.A. Times, The Times of London, The Globe and Mail, and the metro dailies in Charlotte, Dallas, Fort Worth, Houston, Kansas City, Miami, Raleigh, Sacramento, San Antonio, San Diego, San Francisco, plus a few more. If you run into a random local-news paywall, there’s a pretty decent chance that searching for the headline in Apple News might find it. It’s now a much better product for newspapers than it was at launch.

The changes coincide with Twitter officially shutting down its free API, and also with Twitter inaccurately labeling NPR as “state-affiliated media” (a label also given to propaganda outlets like Russian broadcaster RT and China’s People’s Daily newspaper).

On Thursday, Twitter made it impossible to embed tweets in Substack posts. Paste a Twitter link in a Substack post and it simply doesn’t work, giving you this pop-up message:

Team is working on it. We're not the only ones experiencing problems with the Twitter API today.

— Chris Best (@cjgbest) April 6, 2023

Twitter is also not allowing users to take actions on tweets that contain substack.com links — as of Friday morning you can’t like, reply to, or retweet them. (Quote-tweeting still seems to work.)

oh wonderful — I see Elon has made it impossible to RT, reply, or like tweets containing a link to a Substack article. What's next, a SpaceX launch site right outside my condo?

— Aaron Rupar (@atrupar) April 7, 2023

The block on RTs/likes/replies also doesn’t appear to apply to tweets that include Substack sites with custom domains:

Okay, seems like the custom domains are still exempt from this goofy change in Twitter's end. (For now.)

— Parker Molloy (@ParkerMolloy) April 7, 2023

A current workaround is using a link shortener so “substack.com” doesn’t appear in the link you’re sharing.

Looks like you can get around twitter’s apparent ban on substack links by just using a URL shortener lol https://t.co/Bt7HveFV1I

— Ken Klippenstein (@kenklippenstein) April 7, 2023

Substack’s statement:

Statement from the founders of Substack on Twitter’s decision to restrict engagement with Substack content pic.twitter.com/HbQJk5v0jn

— Max Tani (@maxwelltani) April 7, 2023

Twitter’s move against Substack isn’t totally unprecedented; Instagram and Twitter squabbled in pre-Musk times, though more recently the relationship appears to have mended.

1. Substack, at least until recently, was also burning money. The Information, referring to recent SEC filings, reported Friday that the company “Substack’s expenses skyrocketed as a result of its expansion in 2021, causing enormous losses,” and that year “Substack reported negative revenue, which is unusual.” You can look at the SEC filings here.

Maybe you could categorize them all by their financial impact — from low (asking all Twitter engineers to print out their code for his personal inspection) to medium (launching a “verification” system that immediately exploded into market-moving chaos) to high (unbanning enough Nazis to drive away most of your biggest advertisers).

Maybe you’d group them by their degree of self-humiliation, from high (“Based on current trends, probably close to zero new cases in US too by end of April”) to extreme (yelling “I’m rich, bitch!” while being booed off a San Francisco stage) to otherworldly (accusing an employee with muscular dystrophy of faking his disability).

Or maybe you’d order them by the extent of disrespect they show Twitter’s users. Changing Twitter’s logo to the doge Shiba Inu isn’t offensive per se, but it is a persistent reminder that Musk considers the site his personal playground, not anything larger or more important to the world.

The question awaits its own Linnaeus. But before any more ecosystems of shame are discovered, let’s remember one in particular that hasn’t gotten enough attention. That would be Musk’s tossed-off decision to kill off a huge chunk of ongoing research into the internet and how we all interact on it. That includes work by everyone from academics to hobbyists, businesses to journalists. Elon Musk decided to kill it, and soon it will all be dead.

It was September 2006 — only six months into its existence — when Twitter launched the first version of its API. It was free to use and turned Twitter from a website you visited into a data source for an entire ecosystem of apps, tools, art projects, and experiments. Over time, the API’s capabilities expanded and a few paid products were added for those with specific needs. But for the vast majority of cases, the Twitter API was a free window into what people were talking about in the internet’s public square.

That made it catnip for researchers. While Facebook was a much larger and more consequential social network, it was purposefully opaque, limiting outsiders’ view into its inner workings. Twitter’s generous API, meanwhile, made it the most popular choice for academics researching social media and anyone else wanting to understand online discourse. There’ve been a few hiccups along the way, of course, but overall it worked well for a decade and a half. “Twitter is the most (over-)studied social media platform precisely because it offers relatively open data access,” George Washington University’s Rebekah Tromble wrote two years ago.

So of course Elon decided to kill it. In February, desperate for anything that smells like revenue, Twitter announced that it would be eliminating free API access to tweets.1 Instead, it would offer an API level much worse than what was free before at the price of $100/month. (The old free plan for academics allowed access to 100 million tweets per month. The new $100 plan allows access to just 10,000.) To do any meaningful level of research, you’d need to move up to one of their enterprise API plans, which start at $42,000/month. (That’s right, they start there — and go up to $210,000/month.) Been working on a longitudinal research project for five years? Sorry, your data’s about to be shut off.

Because this is Elon’s Twitter, the switch to this new regime has been delayed repeatedly.2 But final pricing details were released last week, and they’re as bad as feared. On Monday, a group of researchers called the Coalition for Independent Technology Research released an open letter decrying the change, saying they “will devastate public interest research” and break more than 300 ongoing research projects that they know of. “Twitter must be held accountable to the public it impacts,” the coalition writes. “And if Twitter is to be held accountable, independent research must continue.” The new rules are set to take effect by the end of the month.

I’m sure there are a few corporations who’ll be willing to pay these outrageous sums, but a Ph.D. student isn’t going to find a spare $42,000 in their couch cushions. The research just won’t get done, and we’ll all be a little bit dumber.

This is the part of the piece where I’d like to list an idea or two for how you could try to help fix this situation. (“Write your member of Congress!”) But the reality here is that the only path to change runs through one man. A man who has spent the past year setting his reputation on fire, reply-guying his way into a laughingstock. One of the richest human beings in the history of the species is unilaterally killing off the work of thousands of researchers, all to chase a buck he won’t end up getting. We may need a new category in the taxonomy.

Here’s the Coalition for Independent Technology Research’s letter.

April 3, 2023

On March 29, after weeks of delays and uncertainty, Twitter’s Developer team announced new pricing tiers for access to the Twitter API that will devastate public interest research.

Over the past decade, researchers across the world have relied on Twitter’s API to study the impact of social media on democracy, the role of social media in strengthening public health, how social media has been used to amplify marginalized voices, and much more. With free API access, researchers could systematically and reliably collect public tweets posted by public figures, gather information about network dynamics, investigate bots and other inauthentic activity, or analyze conversations around specific topics. The knowledge from this research has been shared with journalists, policymakers, and the public, enhancing understanding of issues vital to society.

Free API access also allowed researchers to build public tools like Botometer and Hoaxy that detect social bots and visualize the spread of misinformation. Thousands of users, journalists, and public servants have used these tools in their daily lives and work.

Twitter’s new system to monetize and dramatically restrict access to its API will render this research and development impossible. Unless they can pay, researchers will not be able to collect any tweets at all. The Basic tier costs $100 per month but allows researchers to collect only 10,000 tweets per month — a mere 0.3% of what could previously be collected for free in one day. The Enterprise tier, which ranges from $42,000 to $210,000 per month, is unaffordable for researchers.

Yet even these outrageously expensive Enterprise tiers provide inadequate access for systematic, large-scale research into the impact of Twitter on society. Previously, Twitter provided researchers with low-cost access to its Decahose, a real-time sample of 10% of all tweets. As of March 2023, that equated to roughly a billion tweets per month. The most expensive Enterprise tier would cut that by 80% at about 400 times the price.

In response to a questionnaire fielded by the Coalition for Independent Technology Research, public interest researchers listed over 250 projects that would be jeopardized by ending free and low-cost API access, including research into the spread of harmful content, (dis)information flows, crisis informatics, news consumption, public health, elections, and political behavior. Under the new pricing plans, studying the communications and interactions of even a small population — such as the 535 Members of the U.S. Congress or the 705 Members of the European Parliament — will be unfeasible. The new pricing plans will also end at least 76 long-term efforts, including dashboards, tools, or code packages that support other researchers, journalists, first-responders, educators, and Twitter users.

Though Twitter’s Developer team claimed that they “are looking at new ways to continue serving” academia, they provided no specifics—merely stating that “in the meantime” academic researchers could make use of the pricing plans described above.

What precisely is Twitter “looking into” for academia? How long might academics have to wait for these new options to appear? And what are Twitter’s plans for non-academic public interest researchers, including civil society organizations serving communities around the globe?

Twitter has not answered any of these questions. Indeed, to date, the company has failed to engage with the research community in any meaningful way. Twitter has had mechanisms for dialogue with our community readily available — including via the European Digital Media Observatory, shared working groups tied to the European Union’s Code of Practice on Disinformation, and the company’s own Academic Research Advisory Board. Twitter has not used any of these channels, rolling out these changes without substantive input from public interest researchers whose work will be shut down.

The Coalition for Independent Technology Research will continue to support the research community in the face of these challenges. To date, we have provided mutual aid to nearly 50 projects, focusing in particular on assisting under-resourced and junior researchers. We have supported the National German Library’s German-language Twitter archive efforts, and two groups associated with the Coalition have offered data storage support for researchers.

Our mutual aid efforts are ongoing and will persist as long as possible. Mutual aid is available to all researchers across academia, journalism, and civil society — members and non-members alike. (To request mutual aid or contribute to these efforts, please complete the mutual aid section of this form.)

Going forward, the Coalition will also help organize researchers who wish to explore alternative data-collection and data-sharing mechanisms. And we will continue our discussions with policymakers and regulators around the world.

Twitter must be held accountable to the public it impacts. And if Twitter is to be held accountable, independent research must continue.

That is why — no matter what barriers technology companies erect to public understanding of their services — the Coalition for Independent Technology Research will advance, defend, and work to sustain the right to ethically study the impacts of technology on society.

The Coalition for Independent Technology Research

Executive Board:
Rebekah Tromble
Alex Abdo
Susan Benesch
Brandi Geurkink
Dave Karpf
David Lazer
Nathalie Maréchal
Nathan Matias
James Mickens

Mutual Aid Committee:
Megan Brown
Josephine Lukito
Kai-Cheng Yang

1. After criticism, Twitter announced there would be a free API product, but it’s write-only — meaning it can post tweets but not read existing ones. That makes it useless for research purposes.
2. It was supposed to take effect one week after the announcement. We’re now looking at about three months. Thanks for being unable to meet deadlines, Elon, I guess?

Elon Musk, as he is wont to do, challenged that conventional wisdom when he made Twitter’s blue “Verified” check — previously evidence of actual verification — into a paid product. Verification was initially intended as a confirmation of identity, the sort of small mark that makes a platform sliiightly more trustworthy and secure. But it became some weird marker of status to some of the internet’s worst people, and so it became an $8 SKU.

This conversion — this shift from a “Trust and Safety” feature to a consumer product — had the results everyone predicted, a rash of impersonations, brand danger, and other malfeasance.

But last week, Musk-era Twitter went a step further and said only $8/month customers will be allowed to use SMS for two-factor authentication — a basic layer of security frequently used by journalists, celebrities, officials, and others who fear being hacked. The company tried to explain it as a matter of security (“we have seen phone-number based 2FA be used — and abused — by bad actors”) — but apparently the threat is only to non-paying customers, since Twitter Blue subscribers can keep on using it forever. There will be other ways to use 2FA for Twitter, but they’re not available worldwide and are not without their own risks.

Basic security features going behind a paywall — not good. So it was even less encouraging to see Facebook follow Musk-era Twitter’s lead:

Meta’s testing paid verification for Instagram and Facebook for $11.99 per month on web and $14.99 per month on mobile. In an update on Instagram, CEO Mark Zuckerberg announced that a “Meta Verified” account will grant users a verified badge, increased visibility on the platforms, prioritized customer support, and more. The feature’s rolling out to Australia and New Zealand this week and will arrive in more countries “soon.”

“This week we’re starting to roll out Meta Verified — a subscription service that lets you verify your account with a government ID, get a blue badge, get extra impersonation protection against accounts claiming to be you, and get direct access to customer support,” Zuckerberg writes. “This new feature is about increasing authenticity and security across our services.”

On Facebook, Zuckerberg engaged in some limited back-and-forth with users over the change. (“Call me crazy but I don’t think I should have to pay you guys to take down the accounts impersonating me and scamming my followers.” “This really should just be part of the core product, the user should not have to pay for this. Clearly it’s known by Meta this is filling a need, why profit additionally from it?”)

One user argues that “direct access to customer support is the real value, much more so than the blue check mark.” Zuckerberg: “I agree that’s a big part of the value.” And indeed, a hotline to Facebook customer service is likely the most valuable piece of the package here. But it doesn’t feel good to see features like identity verification — basic stuff for running a trustworthy platform — put behind a paywall.

For Twitter, there’s a certain mad sense to the move. Elon Musk has set the company on fire, from a cashflow perspective, and he’s desperate for all the user revenue he can generate. If 63% of your best advertisers drop you, you grab at whatever dollar bills you see floating by. (Not many seem to be floating Elon’s way.)

Facebook, meanwhile, is still pulling in more than $30 billion a quarter in ad revenue. But various headwinds, whether economic or Cupertino-driven, have demanded a “year of efficiency,” which includes chasing money from users too.

We’re seeing an addendum to that old conventional wisdom about social networks. You can’t charge most of your users — but you can charge some. Few would be bothered by a subscription product that offered additional features — ad-free browsing, say, or custom icons, like the old Twitter Blue. But it’s sad to watch basic security features put behind a credit card charge.

Musk has said he wants to make Twitter an arena for free speech. It’s not clear what that will mean, and his statements have fueled speculation among both supporters and detractors. As a corporation, Twitter can regulate speech on its platform as it chooses. There are bills being considered in the U.S. Congress and by the European Union that address social media regulation, but these are about transparency, accountability, illegal harmful content and protecting users’ rights, rather than regulating speech.

Musk’s calls for free speech on Twitter focus on two allegations: political bias and excessive moderation. As researchers of online misinformation and manipulation, my colleagues and I at the Indiana University Observatory on Social Media study the dynamics and impact of Twitter and its abuse. To make sense of Musk’s statements and the possible outcomes of his acquisition, let’s look at what the research shows.

Political bias

Many conservative politicians and pundits have alleged for years that major social media platforms, including Twitter, have a liberal political bias amounting to censorship of conservative opinions. These claims are based on anecdotal evidence. For example, many partisans whose tweets were labeled as misleading and downranked, or whose accounts were suspended for violating the platform’s terms of service, claim that Twitter targeted them because of their political views.

Unfortunately, Twitter and other platforms often inconsistently enforce their policies, so it is easy to find examples supporting one conspiracy theory or another. A review by the Center for Business and Human Rights at New York University has found no reliable evidence in support of the claim of anti-conservative bias by social media companies, even labeling the claim itself a form of disinformation.

A more direct evaluation of political bias by Twitter is difficult because of the complex interactions between people and algorithms. People, of course, have political biases. For example, our experiments with political social bots revealed that Republican users are more likely to mistake conservative bots for humans, whereas Democratic users are more likely to mistake conservative human users for bots.

To remove human bias from the equation in our experiments, we deployed a bunch of benign social bots on Twitter. Each of these bots started by following one news source, with some bots following a liberal source and others a conservative one. After that initial friend, all bots were left alone to “drift” in the information ecosystem for a few months. They could gain followers. They acted according to an identical algorithmic behavior. This included following or following back random accounts, tweeting meaningless content and retweeting or copying random posts in their feed.

But this behavior was politically neutral, with no understanding of content seen or posted. We tracked the bots to probe political biases emerging from how Twitter works or how users interact.

Surprisingly, our research provided evidence that Twitter has a conservative, rather than a liberal bias. On average, accounts are drawn toward the conservative side. Liberal accounts were exposed to moderate content, which shifted their experience toward the political center, while the interactions of right-leaning accounts were skewed toward posting conservative content. Accounts that followed conservative news sources also received more politically aligned followers, becoming embedded in denser echo chambers and gaining influence within those partisan communities.

These differences in experiences and actions can be attributed to interactions with users and information mediated by the social media platform. But we could not directly examine the possible bias in Twitter’s news feed algorithm, because the actual ranking of posts in the “home timeline” is not available to outside researchers.

Researchers from Twitter, however, were able to audit the effects of their ranking algorithm on political content, unveiling that the political right enjoys higher amplification compared to the political left. Their experiment showed that in six out of seven countries studied, conservative politicians enjoy higher algorithmic amplification than liberal ones. They also found that algorithmic amplification favors right-leaning news sources in the U.S.

Our research and the research from Twitter show that Musk’s apparent concern about bias on Twitter against conservatives is unfounded.

Referees or censors?

The other allegation that Musk seems to be making is that excessive moderation stifles free speech on Twitter. The concept of a free marketplace of ideas is rooted in John Milton’s centuries-old reasoning that truth prevails in a free and open exchange of ideas. This view is often cited as the basis for arguments against moderation: accurate, relevant, timely information should emerge spontaneously from the interactions among users.

Unfortunately, several aspects of modern social media hinder the free marketplace of ideas. Limited attention and confirmation bias increase vulnerability to misinformation. Engagement-based ranking can amplify noise and manipulation, and the structure of information networks can distort perceptions and be “gerrymandered” to favor one group.

As a result, social media users have in past years become victims of manipulation by “astroturf” causes, trolling and misinformation. Abuse is facilitated by social bots and coordinated networks that create the appearance of human crowds.

We and other researchers have observed these inauthentic accounts amplifying disinformation, influencing elections, committing financial fraud, infiltrating vulnerable communities and disrupting communication. Musk has tweeted that he wants to defeat spam bots and authenticate humans, but these are neither easy nor necessarily effective solutions.

Inauthentic accounts are used for malicious purposes beyond spam and are hard to detect, especially when they are operated by people in conjunction with software algorithms. And removing anonymity may harm vulnerable groups. In recent years, Twitter has enacted policies and systems to moderate abuses by aggressively suspending accounts and networks displaying inauthentic coordinated behaviors. A weakening of these moderation policies may make abuse rampant again.

Manipulating Twitter

Despite Twitter’s recent progress, integrity is still a challenge on the platform. Our lab is finding new types of sophisticated manipulation, which we will present at the International AAAI Conference on Web and Social Media in June. Malicious users exploit so-called “follow trains” — groups of people who follow each other on Twitter — to rapidly boost their followers and create large, dense hyperpartisan echo chambers that amplify toxic content from low-credibility and conspiratorial sources.

Another effective malicious technique is to post and then strategically delete content that violates platform terms after it has served its purpose. Even Twitter’s high limit of 2,400 tweets per day can be circumvented through deletions: We identified many accounts that flood the network with tens of thousands of tweets per day.

We also found coordinated networks that engage in repetitive likes and unlikes of content that is eventually deleted, which can manipulate ranking algorithms. These techniques enable malicious users to inflate content popularity while evading detection.

Musk’s plans for Twitter are unlikely to do anything about these manipulative behaviors.

Content moderation and free speech

Musk’s likely acquisition of Twitter raises concerns that the social media platform could decrease its content moderation. This body of research shows that stronger, not weaker, moderation of the information ecosystem is called for to combat harmful misinformation.

It also shows that weaker moderation policies would ironically hurt free speech: The voices of real users would be drowned out by malicious users who manipulate Twitter through inauthentic accounts, bots and echo chambers.

So whether intentionally or not, the creators of the text-generating algorithm GPT-2 played the PR game brilliantly in February when they announced that, well, it just may be too powerful to release to the general public. That generated a wave of global publicity that is, shall we say, uncommon for new text-generating algorithms. (Elon Musk is involved, you’ll be shocked to learn.)

In any event, now, nine months later, the folks at OpenAI have apparently decided that the infopocalypse is not right around the corner and released its secret superweapon GPT-2 into the wild. They say they have “seen no strong evidence of misuse so far” from more limited releases of the technology.

We're releasing the 1.5billion parameter GPT-2 model as part of our staged release publication strategy.
– GPT-2 output detection model: https://t.co/PX3tbOOOTy
– Research from partners on potential malicious uses: https://t.co/om28yMULL5
– More details: https://t.co/d2JzaENiks pic.twitter.com/O3k28rrE5l

— OpenAI (@OpenAI) November 5, 2019

The alleged threat is not, as some journalists have feared, that this machine is going to eventually cover city council meetings and aggregate viral tweets more efficiently than a human reporter could. It’s that the ease of generating semi-convincing output could make it efficient to pump out thousands or millions of individually tailored misinformation, spam emails, or some other text-based form of grift.

I suppose that’s something legitimate to worry about, but my experiences playing around with the GPT-2 has mostly matched this guy’s: Creating readable prose, sensical prose, and believable prose are still pretty different things.

The full GPT-2 model is finally out.

I dunno, guys. pic.twitter.com/k5dP8V2sa5

— Tomer Ullman (@TomerUllman) November 5, 2019

That something sounds/reads like it was generated by GPT-2 is an interesting new kind of an insult.

— Andrej Karpathy (@karpathy) November 6, 2019

To test out its capabilities, I thought I’d feed it the ledes of a few Nieman Lab stories and see what it autogenerated from there. For each of these below, the actual human-written text is in italics; everything after that is “by” GPT-2. (Note that GPT-2, as used on Talk to Transformer, often ends its output mid-sentence.)

Original story: “Meet The Salt Lake Tribune, 501(c)(3): The IRS has granted nonprofit status to a daily newspaper for the first time,” by Christine Schmidt, published Nov. 4.

Verdict: This is in the broad ballpark of news prose style. It’s interesting it chose to contradict the lede so quickly — specifically saying it rejected 501(c)(3) status. The real Fraser Nelson wouldn’t like that quote, I imagine. Call it a B+.

Original story: “Watch your language: ‘Data voids’ on the web have opened a door to manipulators and other disinformation pushers,” by me, published Nov. 1.

Verdict: Choppy short sentences without much flow or sense, despite sticking in the general topic area. I was sorta hoping it would throw in some more Spanish. I’ll give it a C-.

Original story: “The Daily Wire and black salve show that Facebook’s takedown policies have one thing in common: Inconsistency,” by Laura Hazard Owen, published Nov. 1.

Verdict: Pretty realistic at the sentence level, and maybe the self-contradicting “The problem is that the right-wing site has a left-leaning bias” is actually a galaxy-brain commentary on our current politics. A solid B.

Original story: “Why I’m starting a company to build out a new model for local news in the 2020s,” by Ken Doctor, published Oct. 25.

Verdict: Um, no. There’s no mention of Germany in Ken’s piece and I can’t quite figure out what would make the algorithm want to jump straight to Snowden and Merkel. The fractured lines would make this easy to spot. F.

And finally, to go full ouroboros, let’s feed the lede of this story right here into GPT-2.

Original story: This one, silly

Verdict: I’m sure it was right about to pivot back to text-generation algorithms in a couple more grafs. But this is a very reasonable continuation of the lede (other than that first sentence). B.

GPT-2 is not coming to take the jobs of journalists, as some have worried. Paid reporting jobs generally require a certain level of factuality that the algorithm can’t match.

Is it coming for the “jobs” of fake-news writers, those Macedonian teens who until now have had to generate their propaganda (gasp!) by hand? Probably not. Whether your intention is to make money off ad arbitrage or to elect Donald Trump as president of the United States, the key value-add comes in knowing how to exploit a reader’s emotions, biases, preconceptions, and other lizard-brain qualities that can make a lie really hit home. Baiting that hook remains something an algorithm can reliably do. And it’s not as if “lack of realistic writing in grafs 3 through 12” was a real problem limiting most misinformation campaigns.

But I can see some more realistic impacts here. This quality of generated text could allow you to create a website will what appear to be fully fleshed out archives — pages and pages of cogent text going back years — which might make it seem more legitimate than something more obviously thrown together.

GPT-2’s relative mastery of English could give foreign disinformation campaigns a more authentic sounding voice than whatever the B-team at the Internet Research Agency can produce from watching Parks & Rec reruns.

And the key talent of just about any algorithm is scale — the ability to do something in mass quantities that no team of humans could achieve. As Larry Lessig wrote in 2009 (and Philip Bump reminded us of this week), there’s something about a massive data dump that especially encourages the cherry-picking of facts (“facts”) to support one’s own narrative. Here’s Bump:

In October 2009, he wrote an essay for the New Republic called “Against Transparency,” a provocative title for an insightful assessment of what the Internet would yield. Lessig’s argument was that releasing massive amounts of information onto the Internet for anyone to peruse — a big cache of text messages, for example — would allow people to pick out things that reinforced their own biases…

Lessig’s thesis is summarized in two sentences. “The ‘naked transparency movement’…is not going to inspire change,” he wrote. “It will simply push any faith in our political system over the cliff”…

That power was revealed fully in the 2016 election by one of the targets of the Russia probe: WikiLeaks. The group obtained information stolen by Russian hackers from the Democratic National Committee and Hillary Clinton’s campaign chairman, John Podesta…In October, WikiLeaks slowly released emails from Podesta…Each day’s releases spawned the same cycle over and over. Journalists picked through what had come out, with novelty often trumping newsworthiness in what was immediately shared over social media. Activists did the same surveys, seizing on suggestive (if ultimately meaningless) items. They then often pressured the media to cover the stories, and were occasionally successful…

People’s “responses to information are inseparable from their interests, desires, resources, cognitive capacities, and social contexts,” Lessig wrote, quoting from a book called “Full Disclosure.” “Owing to these and other factors, people may ignore information, or misunderstand it, or misuse it.”

If you wanted to create something as massive as a fake cache of hacked emails, GPT-2 would be of legitimate help — at least as a starting point, producing something that could then be fine-tuned by humans.

The key fact of the Internet is that there’s so much of it. Too much of it for anyone to have a coherent view. If democracy requires a shared set of facts — facts traditionally supplied by professional journalists — the ability to flood the zone with alternative facts could take the bot infestation of Twitter and push it out to the broader world.