AI Belongs to Everyone, as Long as Mark Controls It

I devoted some time to reading Meta's article, signed by Mark in his capacity as philosopher. I laughed, and then, sipping my ice-cold tonic water, I declared the time spent reading it officially wasted.

The manifesto is an epideictic text produced by a publicly traded company, and I expected nothing else, so why pass it off as some kind of philosophical treatise? There is a deep performative tension in being told, by the boss of one of the largest concentrations of communicative power in history, that the problem is the concentration of power.

The (far too) long declaration of intent culminates in policy recommendations that all conveniently converge with the commercial interests of the man writing them: no restrictions on training data, legitimization of distillation, export controls for rivals but no regulation for oneself, no delays in releases.

The entire edifice of Mark's argument rests on a premise (never defended), that superintelligence is a coherent concept, that it is imminent, and that it is inevitable. None of this is wrong per se, but it would deserve at least a minimum of argumentation. Instead, nothing. Our interim paladin of digital freedom doesn't even bother to define the term superintelligence. An essay built on predefined corporate axioms, to be accepted as if they had been carved into rock by God, or whoever was standing in for Him that day.

So what does it mean, for Mark, to deliver this undefined superintelligence to everyone? Why, obviously, that billions of people get access to systems designed, trained, filtered, updated, and revocable by Meta. But that doesn't look much like a decentralization of power to me. It looks a lot more like the distribution of a service.

The three thought experiments (the lawyer, cybersecurity, the entrepreneur) are what in Italy, and particularly in Sicily, we call an "americanata", an overblown piece of American theatrics. Yet another paternalistic argument, stained by the same old vice, it takes for granted that universal distribution produces equilibrium, which is true only if the technology is symmetric between attack and defense, and only if the game is not zero-sum. In other words it's not true in the terms Mark argues.

And then, after having crowned Trump president, could the usual sermon on democracy possibly be missing? Pity that eliminating every form of institutional oversight, as our good Mark envisions, is not remotely democratic. The text deploys a Madisonian lexicon while proposing to weaken the only institutional framework available. Meta opposes any policy that would slow the release of American models by even a month, opposes any rigid process or review timelines, preferring instead an alleged (and never further specified) proactive voluntary collaboration between labs and government. But voluntary, self-administered checks and balances are not checks and balances. They are self-regulation, that is the absence of precisely the constraint that the democratic analogy would require.

Proactive voluntary collaboration. Yes. Of course. Now, if you'll excuse me, I'm off to pick bananas on Uranus.