He Is Right, And He Is Also The Last Person Who Should Be Saying It

Zuckerberg's warning about concentrated control of AI is correct, and it arrives from a company that would benefit from fewer constraints on its own releases.

Sign up for my Substack daily AI newsletter here.

See my AI Training course portfolio for corporate Business Leaders here.

Article link: https://johanosteyn.substack.com/p/he-is-right-and-he-is-also-the-last?r=73gqa&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true

Follow me on LinkedIn: https://www.linkedin.com/in/johanosteyn/

There is a particular difficulty in reading an argument you agree with, advanced by someone who profits from your agreeing with it. Meta’s chief executive published a 6,500-word manifesto this week making a case about concentrated AI power that readers in this country have every reason to accept. The awkward part is who is making it.

CONTEXT AND BACKGROUND

The essay, titled The Future is for Everyone, argues that the biggest risk in artificial intelligence is not any specific capability but that one government, company or institution ends up with too much power over the technology. Zuckerberg told Axios that he is personally more worried about centralisation than about any of the specific risks others are discussing, and that putting the technology in everybody’s hands achieves both individual empowerment and checks and balances. He wrote that most other labs are focused on building AI for companies, governments or other institutions, so if those labs lead, the balance of power will favour larger institutions over individuals. Meta simultaneously released the weights of a new model called Muse Glimmer and promised to open a version of Muse Spark 1.2, alongside a billion-dollar fund for American communities hosting its data centres. The manifesto also argues that policy slowing American model releases, even by a month, could add significant risk to American leadership while letting foreign models race ahead.

INSIGHT AND ANALYSIS

Take the central claim seriously, because it is correct. Concentration of AI capability in a small number of American firms is a real structural risk, and it is one this continent lives inside rather than reads about. Now notice the shape of the remedy. Matt Lane, senior policy counsel at the digital rights group Fight for the Future, called Zuckerberg annoyingly out of touch on his best day while agreeing he is right about open source, then made the distinction that matters: only Meta benefits if everyone uses Meta-designed AI systems, even when those systems are fully open source. Open source AI, he argued, needs to be supported beyond these corporate interests, and unlike Zuckerberg, they hope for as much competition as possible.

Anthony Aguirre of the Future of Life Institute put the other objection, noting incidents in which AI models, including Meta’s own, hacked other companies on at least three separate occasions, and asking what makes Meta or anyone think they could control something exponentially smarter and more capable. Distributing a technology is not the same as distributing power over it, and a world in which everyone runs one company’s freely available models is not obviously less concentrated than the one being warned about.

I have previously written about this in an article on the MIT risk study, where I argued that the parties best placed to reduce AI risks are not the parties most likely to suffer the consequences, and that South Africa sits squarely in the second category.

IMPLICATIONS

For South African organisations, the practical questions are narrower than the philosophy. Open weights genuinely matter here, because a model that runs on modest hardware reduces exposure to dollar-denominated inference costs, and local enterprises already face fluctuating operational costs because of rand volatility against the dollar. Accept that benefit without accepting the framing. A model released freely is not a commitment to release the next one, the terms of an open licence are set by the party granting it, and dependence on a single supplier’s generosity is a different arrangement from dependence on its invoices only in the manner of the exposure, not in its existence. The deeper issue is that the argument about concentration is being made for us rather than by us. It is a Global South argument; it is correct, and we have left it to be advanced by a company with a commercial stake in the conclusion.

CLOSING TAKEAWAY

Both propositions hold. Centralised control of the most capable systems is a genuine danger, and a manifesto calling for looser guardrails serves the firm that published it. Holding both requires more discipline than agreeing or dismissing, and it is the only honest position available. South African boards evaluating open-weight models should take the benefit where it is real and price the dependency accurately. South African policymakers should notice that the most prominent case for distributed AI capability now belongs to Menlo Park. The argument deserves better custody than that, and the people best placed to provide it are the ones who would gain most from winning it.

Johan Steyn is a prominent AI thought leader, speaker, and author with a deep understanding of artificial intelligence’s impact on business and society. He is passionate about ethical AI development and its role in shaping a better future. Find out more about Johan’s work at https://www.aiforbusiness.net