AI capabilities are advancing at a faster rate than our institutional response. Soon, a transformative technology will crash into the capitalist political economy, presenting risks to humanity’s survival, autonomy, and democratic order. Especially concerning is the possibility that vast amounts of power and wealth are concentrated into the class that owns the AI systems, or uses those systems to override democratic control. Silicon Valley talks flippantly of a ‘Permanent Underclass’ - and they intend for you to belong to it.
Upon the automation of most, and then plausibly all, jobs, we will face an unprecedented crisis of technological unemployment, with gains from a technology trained on the corpus of humanity’s data concentrated into the hands of a small class of shareholders. Large language models are not conjured from nothing. They are trained on vast quantities of text, images, and code scraped from the internet and digitized archives. The models are, in a meaningful sense, derivative works of collective human effort. They are also built on publicly-funded infrastructure: the internet itself, the universities that educated AI researchers, the government grants that supported foundational research, the public datasets that enabled early breakthroughs. DARPA funding seeded neural network research. Public universities trained the PhDs. The data came from all of us. This is a commons being enclosed.
Given their lack of interest in the welfare of humanity as a whole, the AI companies have resisted pressure from civil society and government to abide by safety standards, even as they recognize potential for catastrophic risks. Clearly, the most transformative technology since agriculture should not be left in private hands. The default trajectory is extreme concentration of benefits but socialized risks.
Thus, the Norwegian Government Pension Fund Global ought to purchase OpenAI and manage it for the benefit of the international community.
GPF-G is valued at over $2 Trillion, and OpenAI’s valuation hovers around $800 billion. Formerly, OpenAI was a non-profit, with capped profits and a windfall clause guaranteeing an intelligence explosion would benefit all of humanity. With the approval of the California Attorney General, this commitment has been scrapped, expropriating the non-profit and allowing OpenAI’s transformation into a for-profit corporation. This is an impossibly deep loss for the human community. With the capital of Norwegian Social-Democracy, we can return the lab to public hands. Though Norges Bank Investment Management would have to liquid 40% of its portfolio and violate the GPF-G’s mandate, accelerating AI capabilities in private hands is an emergency that all actors must approach with increasing flexibility towards existing rules and norms.
To those worried about concentration of power in the hands of a government, Norway is one of the world’s most stable democracies, with a decades-long track record of collectively managing wealth. The Kingdom is deeply embedded in international institutions, with a moral commitment to cosmopolitanism not shared by a private corporation or the Republican Party. The GPF-G even divested from several Israeli firms collaborating with the occupation of the Palestinian territories. After the takeover, the GPF-G ought to transfer management of the lab to an international multilateral institution with authority over development. This is a desirable end point, but there must be some mechanism of transfer, thankfully Norway’s deft management of its oil revenues provides egalitarians with an agent capable of doing so.
It is one of the few countries to consistently meet or exceed the UN target of 0.7% of gross national income devoted to foreign aid, often reaching 1%. It has served as a mediator in conflicts from the Middle East to Sri Lanka to Colombia. It hosts the Nobel Peace Prize. It houses the Svalbard Global Seed Vault, a backup of the world’s agricultural biodiversity held in trust for humanity. Norway’s foreign policy identity is explicitly cosmopolitan, as a small country it must attain security and legitimacy through international institutions rather than unilateral power.
Such an intervention would clearly be blocked by the United States government, though if the Nobel Peace Prize Committee awards Trump the bauble then perhaps there will be an opening of political possibility. Perhaps this is an option that should be left open for a future administration, should Congress refuse to bring the labs into public hands themselves.
The gap between the magnitude of the AI transition and the policy imagination of the left is impossibly vast. Proposals to ban data center construction are a reflexive response, foreclosing the benefits of living in an automated, but collective, future in favor of naive localism. The Overton Window on AI governance is narrow to the point of absurdity. If the ownership structures of the labs are not changed, and democratic steering not guaranteed, the default future is the disempowerment of the vast majority of the human population. We need policy proposals that push the boundaries of plausibility, ones that will be dismissed as impossible or unnecessary. We need optionality. In the future, we might not have much time to respond, and such imaginative plans will be in the wings, available to actors with a prosperous and egalitarian future in mind.