More than a decade ago, in a prescient essay for Scientific American, the inventor of the World Wide Web denounced what Facebook and other tech giants were doing to his signature invention. “Why should you care?” Tim Berners-Lee wrote at the time. “Because the Web is yours.” These companies, he warned, were restructuring the web itself, turning an expanse of interconnected websites all built on the same open infrastructure into a series of “fragmented islands” where users were kept hostage.
On Facebook’s island, he wrote, people give over their entire digital life for the chance to connect with their friends, but have no way to transfer their information to any other platform. Once captive, users upload photos, add friends, send messages, click ads, and react to posts, all the while leaving a trail of information from which Facebook can profit. The more they do these things, the harder leaving becomes—so much of people’s digital life is nested in Facebook, rather than in Facebook’s rivals. The logic extends to other tech platforms too. On Apple’s island, Berners-Lee explained, iTunes users can tap into an immense catalog of music but can’t easily share it with anyone. “You are trapped in a single store, rather than being on the open marketplace,” Berners-Lee wrote. “For all the store’s wonderful features, its evolution is limited to what one company thinks up.”
That was 2010. Since then, with the aggressive development of Apple’s App Store, Facebook’s strategic acquisitions of Instagram and WhatsApp, and many other protectionist moves that have made tech’s most dominant companies even more powerful, the web’s fragmented islands—or “walled gardens,” as Berners-Lee also called them—have grown only more secluded.
But lately, a funny thing has happened. As tech giants face mounting antitrust scrutiny and try to navigate the development of generative AI technology, the most powerful companies in Silicon Valley are attempting to signal their open-web bona fides. Meta, the company formerly known as Facebook, has become a particularly prominent voice. Last week, it announced that it was partnering with Microsoft on the release of its latest large language model, Llama 2, which it is making openly available for free. That means that, unlike its nearest rival, GPT-4, which users can pay to license from OpenAI, developers will be able to download Llama 2’s code, tinker with it, and build new things on top of it, dramatically expanding access to generative AI technology—and potentially leaving OpenAI out to dry in the process. Around the same time, Meta announced that its Twitter copycat, Threads, will eventually be interoperable with small competing social platforms such as Mastodon and WordPress.
If any of this sounds like the first step into a wonderful new era of ungated collaboration, it’s not. These gestures toward openness aren’t the product of some sudden, soul-cleansing instinct to cede power. Quite the opposite. Lowering the garden walls ever so slightly works in service of entrenching Meta’s power and ensuring that the company is just as indispensable to the next era of computing as it was to the last one.
In his Facebook post announcing the Llama 2 release, Mark Zuckerberg co-opted the gospel of openness, arguing that it would “unlock more progress” and improve safety for generative AI more broadly by enabling more developers to build these technologies and identify potential problems and fixes. It’s not exactly a novel idea: Some of AI’s most prominent ethicists have repeatedly raised concerns about black-box AI models being concentrated in the hands of just a few multibillion-dollar companies. But Zuckerberg is decidedly not one of AI’s most prominent ethicists. He is, instead, a person who not only runs one of those multibillion-dollar companies, but who is also responsible for building and ruthlessly defending one of the internet’s most infamous walled gardens.
This is the same company that has prohibited academics from scraping data from its products for the purposes of research, neutralized competitors by acquiring them, and actively made it harder for rival platforms to use Facebook’s features. When the video platform Vine wanted users to be able to find their Facebook friends through its app, Zuckerberg personally approved the order ensuring that they couldn’t. When people started posting links to Instagram photos on Twitter, Facebook prevented the links from generating full image previews, rendering them mostly useless for nearly a decade until the policy changed.
It isn’t difficult to understand why Meta would be flirting with a more open approach. For starters, giving away Llama 2 for free will help the company speed up adoption to compete with OpenAI. Besides, offering free access to powerful tools and then figuring out how to make gobs of money from them later is kind of Meta’s thing: It did the same with Facebook before ever charging advertisers a dime. The idea that social media should be decentralized—enabling different networks to plug into one another, rather than hoarding every user for themselves—is also having something of a moment among the extremely online. That’s mostly due to Twitter’s Elon Musk–induced implosion, which has sent former Twitter users on a search for alternatives en masse. If the internet is indeed transforming from a series of fragmented islands into something more like an archipelago, Zuckerberg is already staking his turf.
To Meta’s credit, as Zuckerberg dutifully noted in his post, the company does have a history with open-source development of machine-learning technology. And the long-standing walls around just about every other part of its business, including its billions of users, arguably have offered some benefits beyond profits. At least in theory, those barriers help the company protect user privacy and enforce its standards and policies, thereby taming some of the chaos it has unleashed on the world. Meta’s enforcement record, however, has been inconsistent at best—and utterly lacking at worst.
Despite this already lax policing, what Meta is doing now with Threads and Llama 2 will make enforcing its own rules even more difficult for the company. If Threads becomes interoperable with other platforms such as Mastodon—allowing posts to flow back and forth between servers—Meta will face rule breakers from a whole bunch of other networks, including ones where just about anything goes. (So-called federated social media is already home to an astonishing volume of child-abuse material, according to a new report from the Stanford Internet Observatory.) With Llama 2, Meta has invested in AI guardrails designed to prevent the worst abuses, but as has been the case for other open-source models, once developers have their hands on it, people will inevitably find ways to jump those guardrails, and Meta will have limited control over what those people do on the other side.
All of this can seem like a moderation nightmare for Meta—until you consider that the company might use its newfound excitement about abdicating control as an excuse to abdicate even more responsibility. If Meta’s open to everyone, then maybe its problems are everyone else’s.
Like Berners-Lee, many people deeply believe in the dream of an open web. But the danger right now is that the performance of this belief can easily cover up a canny business strategy. Despite its public framing, Llama 2 isn’t quite what purists would call open source. In the fine print, its community agreement stipulates that although most developers can freely access the model, anyone with more than 700 million monthly users will have to get a license from Meta first. Tough luck for Meta’s competitors such as Snap and Telegram, who just happen to miss the cutoff.