Facebook's strange data dunk

Programming note: Today is the first day of Twitter’s “profile subscribe” feature – which allows Twitter users to subscribe to Revue newsletters directly from their creators’ profiles – and this newsletter has the lucky fortune of being one of the featured newsletters at launch. If you arrived here from my profile button, welcome! Have a look around. And if you like what you see, please consider subscribing. It’s free, occasionally insightful, and always much better than my Twitter feed.

Our first order of business today is Facebook’s new “widely viewed content report,” which came out yesterday and represents, among other things, the most time and effort a trillion-dollar company has ever spent trying to make me look like an idiot.

I have already written so, so many words about my two-year feud with Facebook over @FacebooksTop10, a Twitter account I started because I was tired of spamming my Twitter followers with daily updates on the most viral content on Facebook.

I might literally die if I have to tell the whole story again, but the basic gist is:

– The @FacebooksTop10 lists, which capture the sources of the link posts by U.S. pages that get the most reactions, comments and shares, revealed that Dan Bongino, Ben Shapiro and other right-wing commentators were absolutely crushing the competition on Facebook, frequently racking up more engagement on their posts than basically anyone else, including major mainstream news outlets.

– This observation caused extreme discomfort to a few executives at Facebook, mostly civic-minded California liberals who didn’t like thinking of their flagship product as a giant Bongino machine.

– Last year, when it looked like Trump might win re-election (and that Facebook’s right-wing content ecosystem might be implicated in getting him re-elected), those executives decided that discrediting my 30,000-follower alt account was an Urgent Company Priority, and devoted a ton of resources to figuring out how to push back on the narrative that Facebook was a right-wing echo chamber.

– The company eventually landed on a strategy that involved promoting a different metric, known as “reach,” that executives thought would give a more balanced, less embarrassing view of what was happening on its platform than “engagement,” the main metric that is measured by a Facebook-owned data tool called CrowdTangle (and, by extension, by @FacebooksTop10, which uses CrowdTangle as its data source).

– Initially, Facebook’s plan was to put reach data into CrowdTangle so that journalists and researchers could poke around for themselves and see that Facebook was not, in fact, a giant Bongino machine – that, in fact, hyperpartisan political content was a tiny portion of what people actually saw on their feeds. But when they built a prototype, they found issues. One problem: misinformation and conspiracy theories that got a lot of engagement also got a lot of reach. The reach-ranked lists weren’t as Bongino-heavy as the engagement-ranked lists, but they weren’t entirely Bongino-free, either. “Reach leaderboard isn’t a total win from a comms point of view,” CrowdTangle CEO Brandon Silverman wrote in an email to other Facebook executives last fall. (Silverman was subsequently relieved of his duties, and the entire CrowdTangle team was broken up and shuffled into other Facebook divisions, in one of the great shooting-the-messenger feats in recent corporate history.)

– The backup plan, promoted by Facebook CMO Alex Schultz and others, was to start putting out quarterly reports on “widely viewed content.” These reports, which would accompany Facebook’s quarterly community standards enforcement reports, would list the top posts, links and domains of the quarter ranked by reach, rather than engagement. The hope, among some executives, was that these reports would draw attention away from @FacebooksTop10, and successfully convince people that most of what people see on Facebook isn’t right-wing rage-bait, but cute cat videos and pictures of their grandkids. The echo chamber myth would be busted! The journalist would be owned! Reason would prevail!

That is…not exactly how it went down. Almost immediately after the report landed yesterday, people began noticing how strange it was.

Ethan Zuckerman pointed out that the report raised more questions than it answered. In particular, it revealed that some of the most-viewed content on Facebook was literal spam – a link to a Green Bay Packers alumni booking site, for example, was the #1 most-viewed link on the platform in Q2, with more than 87 million views. (Other top links came from similarly obscure domains like Pure Hemp Shop, MyIncredibleRecipies.com and ReppnforChrist.)

Will Oremus wrote that “instead of presenting Facebook as a hotbed of right-leaning politics, the company’s inaugural report presents a far weirder, messier, and spammier picture: the news feed as a junk-mail folder.”

Charlie Warzel also characterized the report as “very weird,” and said that its seemingly arbitrary methods of slicing the data “might also be representative of how internally disorganized the platform is.”

And Brian Boland, Facebook’s former VP of partnerships strategy (who left the company last year over his concerns about transparency) wrote that after reading the report, he “came away believing that this entire effort is a PR stunt.”

I have a lot of thoughts on the report itself, which I might end up exploring at column length. For starters: I’m genuinely glad that Facebook has put more data into the world that might help outsiders better understand its information ecosystem, no matter the motive behind it. I’m already hearing from researchers who are digging interesting nuggets out of the report, and expect that future versions will unearth even more useful information.

But I keep coming back to the origin of the report as the primary explanation for its strangeness. This report didn’t emerge from an organic transparency process, or a data science team deciding that it had interesting information that the public needed to see and making it available in a coherent, user-friendly format. It emerged from a handful of Facebook executives getting extremely mad about a Twitter account, and tasking their teams with figuring out a more flattering way to slice the data.

If the report seems weird, in other words, it’s because it is weird. Trillion-dollar companies don’t usually care this much about single-purpose Twitter accounts. They don’t usually undermine their own data tools, just because those tools allow journalists and researchers to spot problems with their platform. And they definitely don’t usually assign teams of highly-paid engineers and data scientists to spend months producing transparency reports of dubious value, just to throw shade at a journalist’s side project.

Shannon McGregor, one of the smartest academics studying social media’s effects on politics, summarized my general feelings on all of this:

What a weird time to be alive, indeed.

Second, I want to discuss my recent foray into the world of Pudgy Penguins.

You can read that column if you want to understand the whole phenomenon, but the basic story is that crypto fans, unsatisfied with paying hundreds of thousands of dollars for old memes and pictures of newspaper columns, have come up with the idea of “community NFTs.” These are limited-series tokens that feature many different variations on a theme, like overweight penguins, or bored apes, or low-res punks. Some of these NFTs have sold for millions of dollars apiece, and become a luxury status symbol. (Jay-Z, for example, reportedly owns a CryptoPunk.)

I understood the economic appeal of NFTs, and why you might invest in them if you were hoping to turn a quick profit. But I didn’t understand the social appeal until I started hanging out in the Pudgy Penguin Discord server a few weeks ago.

I expected the Discord, like some other crypto Discords I’m in, to be mostly full of crypto shop talk and traders frantically trying to pump their bags. Instead, it was a whole scene! Just thousands of Penguin owners spending all day casually shooting the shit about any topic that crossed their minds.

Here they are, for example, talking about their favorite fast-food restaurants:

And here they are, a few minutes later, debating the best flavor of Mountain Dew:

There’s plenty of NFT-related chatter in the Discord, obviously, along with all the memeing and bag-pumping you’d expect. But what surprised me was how little their collective mood seemed to be related to the price of Penguins, or anything having to do with crypto, really. They were simply enjoying each other’s company, building friendships, exploring their commonalities and differences. In some ways, the penguins were just a bonus.

It’s kind of amazing, if you think about it. Here, in a chat room devoted to buying and selling cartoon penguins, was an example of the classic Zuckerbergian dream: the internet connecting people and allowing them to discuss their shared interests in a clean, well-lighted place. As a social space, the Pudgy Penguin Discord was orders of magnitude more fun than Twitter or Facebook, or anywhere else I routinely spend time online. Being there felt like recreating the sprawling message boards and chaotic ICQ chats of my teenage years, and it made me instantly aware that if I were 17 today, I’d be spending all my free time and allowance money collecting NFTs, rather than buying Magic: The Gathering cards.

Look, I’m as skeptical of excessive crypto hype as anyone. I roll my eyes at the messianic proclamations of crypto-optimists, and I shudder when I hear Bitcoiners talking about how adopting crypto could have saved Afghanistan from the Taliban. (Seriously, that’s a thing people are saying.)

But it’s also hard to spend a few days hanging out in a community NFT Discord server and not see the appeal, on a purely social level.

Does this mean I’m emptying my bank account and becoming a penguin collector? Not even a little. But I understand why a lonely, bored young person might be tempted to take part in a project like the Pudgy Penguins, even if they don’t care much about crypto or blockchain-based collectibles.

Community NFTs, at their core, are a kind of securitized social capital. They introduce you to a group of like-minded people, and give you access to space away from your main social feeds where you can nerd out on your favorite subjects. If you squint, you can imagine the most successful of these projects becoming to Gen Z roughly what country clubs are to boomers, or Soho House is to millennials – a structured, high-status way to break into a new social scene, with some bonus perks (golf, networking, the chance to flip a jpeg for a six-figure profit) to help justify the expense.

Sure, on some level, the penguins are dumb. You probably shouldn’t buy one unless you have a bunch of crypto gains you can afford to lose. But all told, there are worse ways to make friends online.

Part of what bugs me about the mainstream discussion of AI and automation – a gripe I spent a lot of time expanding on in Futureproof – is that the consequential things rarely get attention, and the attention-getting things are rarely consequential.

Case in point: Boston Dynamics, the viral video company masquerading as a manufacturer of advanced robotics technology, came out with a new video this week showing its bipedal robots doing Parkour moves. It’s an impressive-looking stunt, and it garnered all of the usual Twitter reactions and Terminator jokes.

But despite how terrifying its robots look, Boston Dynamics – which was once owned by Google before being sold to SoftBank and is now, confusingly, owned by Hyundai? – isn’t actually anywhere near the cutting edge of AI research. In fact, leading AI companies like OpenAI have been quietly disbanding their hardware robotics teams in recent years, and focusing on software-based projects that have more promising near-term applications.

One of those projects, OpenAI’s Codex, was unveiled last week. It’s an evolution of the organization’s GPT-3 language model, and its main function is turning English-language text into functional code. That sounds like a simple task, but you really have to watch the demo to grasp how crazy it is in practice. The engineers type basic commands like “Make a webpage that says our message” and “Start a web server to serve that page,” and Codex does it automatically, using a neural network that has been trained on tons and tons of examples of functional code.

Codex, which can write code in 12 programming languages, is the type of benign-sounding AI application that will actually have profound implications for our society in the coming years. Untold numbers of junior web developers are currently employed doing the kinds of basic tasks that can now be done by Codex, and while the software’s makers think it will ultimately make coders more efficient rather than replacing them, there’s a non-zero chance that Codex and programs like it could dramatically shake up the labor market for programmers. AI-assisted coding could also alter fields like cybersecurity, by making it possible for bad actors to build customized exploits without much technical knowledge.

So, why do we talk much more about Boston Dynamics’ dancing robots than OpenAI’s programmer-replacing software? I’m guessing it’s because we’ve been educated by decades of sci-fi movies to expect the robot apocalypse to arrive in a humanoid hunk of metal, rather than in the backpropagation algorithms of a neural network.

But anyone who understands AI will tell you that software-based AI is going to be far more transformative than the hardware robots we see in viral videos, at least in the short term. (Even if those robots work flawlessly, which they don’t, their usefulness is limited to physical labor.)

I’m sure we’ll keep oohing and ahhing over the Boston Dynamics videos every time a new one comes out. But personally, I’d like us to start worrying less about robots that know Parkour, and more about robots that know Python.

Did you like this newsletter? You can get regular installments of Futureproof in your inbox by subscribing here. Want something a little heavier? My new book, Futureproof, can be ordered here. Thanks for reading, and see you next time.