Zuckerberg's two futures – from the metaverse to personal superintelligence

On August 10, Mark Zuckerberg published The Future is for Everyone — an AI manifesto of roughly 6,500 words. The reactions came fast. Russell Brandom wrote in TechCrunch that the manifesto is exactly why people don't like AI. Elizabeth Lopatto concluded in The Verge that Zuckerberg doesn't seem to understand how to live.

Both pieces are worth reading. But the most interesting comparison is a different one: between this manifesto and the vision of the future the same man was pushing just a few years ago — the metaverse. That comparison says something about both visions, and it exposes the question the manifesto never quite answers.

What the manifesto actually says

It is worth reading the manifesto on its own terms before weighing the criticism, because its core thesis is not "AI will do everything for us." It is: the capability must be distributed to individuals, or the power concentrates in institutions.

From that thesis follow a few ideas that are stronger than the headlines suggested:

• Empowerment over automation. Zuckerberg distinguishes between AI that automates people's existing work and AI that expands what individuals are able to do. He explicitly argues for the latter — "invention superpowers" is his phrase.
• Balance of power as the safety model. His lawyer thought experiment captures the principle: if only one party in a lawsuit has a superintelligent lawyer, they hold an enormous advantage; if both do, the balance is restored. AI risk becomes not just a question of absolute capability, but of the difference in capability between actors.
• Pluralistic alignment. Humanity, as he puts it, is not a monoculture. There is no single set of values a central superintelligence could be optimized against. His alternative: my AI should serve my values and yours should serve yours, within legal and safety limits — and many systems should balance one another, roughly the way markets and democratic institutions do.

That last idea is essentially the same principle much of Western society rests on. We do not try to engineer the perfect human; we build institutions in which imperfect humans check each other's power. The manifesto transfers that principle to AI. It is a considerably deeper thesis than "Meta wants to give everyone an AI assistant."

The criticism — and what it misses

Brandom's objection is essentially that Zuckerberg is selling a techno-optimist future to a public that never asked for it, acknowledging the risks only to move briskly on to why everything will be wonderful anyway. Lopatto's objection cuts deeper: that the vision rests on a narrow idea of what a good life is, in which efficiency and goal completion take over everything from hobbies to relationships. Often the effort is the point — someone learning to play a song does not want a perfect recording in three seconds.

That is an important objection to a real tendency in the AI industry. But it lands less cleanly on this particular text than you might expect. Zuckerberg's own example with his daughter is not the AI baking for her — the agent plans recipes and orders ingredients while they bake together. And the manifesto states outright that there is no objective answer to what a good life is, and that people must decide for themselves what matters to them.

Where the criticism does land is on trust, and that is harder to read away: the company promising a personal AI loyal to the user's goals is the company whose existing products have drawn years of criticism for their actual effects on people. No 6,500 words resolve that.

We never stopped knitting

Part of the criticism also rests on an assumption that history does not support particularly well: that if a machine can do something for us, we stop doing it.

We do not need to bake bread, grow tomatoes, knit sweaters, build furniture or run ten kilometers. People do all of it anyway — because the activity carries a value separate from its economic output. What automation changes is often not whether we do something but why. Before industrialization, knitting was production. Afterwards, the same activity became craft, identity, community. The sewing machine did not kill sewing, and Spotify did not kill guitar playing.

So I think "AI liberates humans from cognitive work" is the wrong description. More likely, AI liberates us from certain cognitive production requirements. The difference shows in two sentences that superficially describe the same activity:

I write because I want to think.

I write because the organization requires a fourteen-page monthly report.

If AI takes over the second, people do not have to stop thinking. They may get more time for the first.

Load is not development

There is one more distinction the debate rarely makes: between cognitive load and cognitive development. Processing a hundred emails, moving information between systems and sitting through status meetings loads the brain — but it does not develop it much. Learning an instrument, building something, or understanding a new field also loads the brain, but with an entirely different quality.

The service economy has done something strange besides: we automated away most of people's physical work and then built administrative systems that keep them seated for eight hours a day — after which they go to the gym to recreate the movement the economy just optimized away. Meanwhile, it is well established that regular physical activity supports memory, attention and learning.

That opens an almost ironic possibility: if AI absorbs a larger share of the administrative sitting-still, and people move more, they may get better at the cognitive activities they actually want to pursue. "AI does more cognitive work" does not have to mean "humans think less."

The question technology cannot answer

Here is the manifesto's real gap, and it is not technical but economic. Suppose AI can perform half of today's administrative and cognitive paid work. Society can then, in principle, choose: the same working hours and dramatically higher output — or the same output and considerably shorter working hours.

Technology does not decide which one we get. Ownership, tax systems, wage formation and politics do. That is how industrialization played out too: the productivity gains were eventually taken out both as higher living standards and as the shorter work week and longer vacations. AI reopens the same negotiation, and the manifesto — which happily uses the agricultural society as its analogy — never really enters it.

Add to that the assumption the entire construction rests on: that distributed superintelligence actually produces balance rather than an arms race. That is not self-evident. Some technologies favor the defender, others the attacker. And when the most powerful capability requires data centers costing hundreds of billions, the question becomes how "superintelligence for everyone" can mean decentralized power, when the physical infrastructure is owned by a very small number of actors.

Still: a more mature document than the metaverse

Even so, I find the manifesto a markedly more mature document than the metaverse vision — and the difference is instructive.

The metaverse was a destination. There was a specific picture of the future: we would work, socialize and play in virtual environments, and Meta would build the world we moved into. The technology tried to prescribe a human behavior for which no strong demand yet existed.

The manifesto starts from the other end. It assumes that nobody — Meta included — knows what people will want to do, and concludes that they should get the tools and decide for themselves. Platform-centered has become individual-centered. It is also strategically stronger: Meta no longer has to guess right about the next big human activity.

And there is a small irony here. If the new vision succeeds, the metaverse may eventually emerge after all — but as one of thousands of things people choose to build, not as the future a company tries to move them into.

The next question: digital delegation

If the personal AI becomes real, a question follows that the manifesto only touches on, and that I believe will be decisive: representation.

An AI that knows my goals and works on my behalf will sooner or later act in my name — attend a meeting I cannot make, negotiate a draft with someone else's AI, hand over information I have approved. At that point, solving digital identity — the question of who you are — is not enough. We also need to solve digital delegation:

What is your digital representative allowed to do in your name — and how is that proven?

My AI may be allowed to book a trip but not sign a contract. It may discuss technical solutions but not express opinions I do not hold. This is fundamentally the same problem as powers of attorney, authorization and federation — fields with decades of architectural groundwork to build on, and where most of the practical work remains.

The metaverse asked us to move into someone else's world. The more interesting future is probably the inverse: each of us carrying our own — and learning to govern what it may do for us.