Zuckerberg's manifesto: Do Meta's promises and actions align?

How Zuckerberg sees the future of AI

Zuckerberg describes the current AI discourse among AI developers as "filled with doom." He says fears that modern AI threatens to take jobs and spawn large-scale cybersecurity chaos have created a widespread belief that the only way to control AI risks is to create an "extreme concentration of power" by restricting access to the most powerful AI models to a handful of businesses.

Zuckerberg offers a different take on what AI can become. In his telling, AI, if developed and disseminated in the right way, will usher in "a new era of personal empowerment."

According to Zuckerberg, the key to unlocking this future is to ensure that everyone has access to advanced AI technology. In his manifesto, he frequently refers to democracy and "democratic values." However, he stops short of saying we should "democratize AI," perhaps because that would imply that everyone should have the freedom to develop the technology, not just access it. He also writes about the importance of "putting power in people's hands." History shows, he says, that technology creates the greatest societal benefit when it's distributed and available to all.

Meta's AI commitments

The manifesto doesn't go into much detail about the role Zuckerberg sees Meta playing in ensuring everyone has access to the best AI technology, but it does mention a couple of notable initiatives.

One is the introduction of what Zuckerberg calls "personal agents" or "personal superintelligence agents." It's not clear how these will work, but the manifesto says they'll help to "improve your relationships, health, career, finances, home management, hobbies, and more." It also mentions that the agents will come with "strong privacy and security options," although again, it doesn't go into detail about them.

The other key Meta contribution he mentions is the release of open source AI models. Meta began debuting open source AI models in 2023, but it seemingly abandoned the effort earlier this year in favor of proprietary models. But according to the manifesto, Meta plans to "resume releasing some open source models soon."

Meta's AI strategy: Vision vs. reality

On the surface, it's hard to take issue with the vision for AI's future that Zuckerberg articulates. Who doesn't want to ensure that the best AI technology benefits everyone and is safe and secure? But it's also hard to have deep confidence in Meta's willingness or ability to deliver on this vision.

So far, Meta has lagged in developing cutting-edge AI technology. As the New York Times writes, Meta's latest proprietary model, Muse Spark, "does not perform as well as leading A.I. models from Anthropic and OpenAI on measures such as coding, reasoning and writing."

It's possible that Meta will catch up on the AI front. But at present, it seems unclear that the company is in a position to create AI models or agents that will perform as well as those of the AI labs that Zuckerberg accuses of creating the "extreme concentration" of AI power.

Another, bigger question surrounds how committed Meta is to democratizing access to AI. Although Meta released a series of AI models it describes as "open source," some segments of the open source community, like the Open Source Initiative, have taken issue with its approach.

The criticism centers on the limited transparency of how the models work. It also involves Meta's decision to create an unusual custom license for its open source models instead of using an established open source software license. This effectively bars other large technology companies from using the models without Meta's permission. These terms undermine the spirit of open source software, which prioritizes transparency and free access, even for competitors.

To date, Meta hasn't said how its approach to open source AI models could change. The company might decide to offer models that are more "open" in terms of transparency, and to rethink its licensing terms. If it doesn't, it's hard to believe Meta's so-called "open source models" will move the needle toward making AI open and flexible for everyone. Users and businesses who want true open source models are more likely to turn to alternatives like OpenClaw, whose core is genuinely open source and governed by the MIT License, a widely used open source license.

It's also worth noting that although the manifesto states that Meta plans to "offer free versions [of Meta's AI tools] that will be accessible to billions of people," it also mentions a "dynamic auction mechanism" that will let people pay for access to extra compute. Given the limited information on how the technology will work, it's hard to say which benefits this extra compute will confer. This could be Meta's way of saying "everyone can use our AI technology, but deep-pocketed consumers and companies can pay for and access more capable versions."

New business challenge, old Meta strategy

The depth of Meta's commitment to ensuring everyone can benefit equally from AI is also uncertain, given the company's long history of positioning itself as a champion of the underdog in its self-serving marketing initiatives.

In 2020, Meta ran ads declaring, "We're standing up to Apple for small businesses everywhere." This was a response to new features in Apple products that restricted ad tracking and personalization -- a direct threat to Meta's revenue. It was hard to believe that Meta cared about small businesses as much as it worried about the sustainability of its ad-based business model.

Another Meta ad campaign touted its open source AI models, promising "Open source AI: Available to all, not just the few." That sounds remarkably similar to the promises Zuckerberg made in his recent manifesto. But just as it's hard to point to much that Meta did during this campaign to deliver true open source technology, it's also difficult to believe that Meta's personal AI agents or renewed commitment to open source AI models will change much now.

A cynic might interpret Zuckerberg's manifesto as an effort to position Meta as an outlier in the AI ecosystem, not because the company is committed to open source or the democratization of AI technology, but because it's falling behind other hyperscalers in the AI race. This is a PR move above all. Don't bet the future of AI on Meta right now.

Chris Tozzi is a freelance writer, research adviser, and professor of IT and society who has previously worked as a journalist and Linux systems administrator.