Before generative AI, the world still retained some semblance of meritocracy. But in a world where “intelligence” is directly correlated with how much money you can spend on compute, will it be erased entirely, asks Lewis Liu
OpenAI recently bulk-published 700 mathematics papers in one go. Leveraging its research agents, it claimed to have solved some extremely important mathematical problems, many of which will likely be validated as correct.
My friends in the field are deeply alarmed about the state of mathematics, what this means for them personally as professionals, and what it means for the field. Terence Tao, the UCLA professor widely regarded as the world’s greatest living mathematician, has even called for the discipline to boycott OpenAI.
One alarming argument now circulating through the mathematics community, and beyond, is that this seals the fate of the capital versus labour debate in our society. The argument goes that, before generative AI, the world still retained some semblance of meritocracy. If you used your brain and worked hard, you could leverage your intelligence to make money or gain status, whether as a professional, entrepreneur or scientist.
But in a world where “intelligence” is directly correlated with how much money you are willing and able to spend on compute, scaled “intelligence” becomes accessible only to those with enormous wealth and capital. That creates a permanent trap for anyone who has not accumulated wealth during the AI boom, destroying the opportunity for even the most talented and hardest-working people to progress in life.
This is a far more insidious iteration of the “permanent underclass” fear that drives many Silicon Valley founders and workers. The previous argument was that, because capital accrues value far faster than labour under our tax system and capitalist structure, this is the last chance to make money before society ossifies. That fear was centred on capitalism’s tendency to create monopolies.
This new iteration, sharpened by OpenAI’s results, is more fundamental. It is centred on the devaluation of individual human intelligence, the destruction of the opportunities that intelligence once created, and the hoarding and monopolisation of “intelligence” through the deployment of capital into AI.
Look, it is one thing to hoard capital, or to build wealth, depending on how you view the morality of it. That has been a feature of American capitalism for decades. But it is quite another to hoard “intelligence”, and with it human dignity and agency. Yet that is where the AI endgame takes us.
How AI will unsettle the value of labour
Without AI, the imbalance between capital and labour created by our tax system and regulatory framework can, at least in theory, be corrected. I have written about this before. But once AI enters the equation, the entire system risks becoming unworkable for any civil, democratic society.
Stanford economics professor Chad Jones has modelled how value might accrue to capital and labour across multiple scenarios for AI development and adoption. His conclusion is extraordinary: across every scenario he examined, capital’s share of income ultimately approaches 100 per cent.
OpenAI’s recent mathematics results, and the alarm now spreading through the mathematical community, offer a glimpse of the mechanism through which this could happen. No individual human, however brilliant, can outcompete a corporation or hyper- wealthy individual able to deploy vast amounts of capital into compute. And even if AI advances no further than it has today, those with sufficient capital can already outperform individual humans across a growing number of tasks simply by buying more computational power.
So what does this mean for human agency?
In a world where almost everyone is born outside the tiny group of perhaps 5,000 People who control the bulk of this wealth and compute, intelligence, talent and hard work cease to offer a meaningful route to a better life. Almost nobody can change their station, because no individual can compete with “intelligence” deployed at industrial scale. Everyone else becomes entirely dependent on the people who own the capital.
There is another word for that: slavery.
What do we do?
We humans, being human, are not going to stop building AI. We built the atomic bomb, after all. Slowing or stopping AI is not the answer, despite the narrative OpenAI and Anthropic are constructing to justify regulatory capture.
I offer three parallel paths forward.
First, support open source. Generative AI is built on the wholesale appropriation, or theft, if you prefer, of the intellectual property of past and present humanity. That is one of AI’s original sins. The only moral response is to support open-source technology.
The cat is already out of the bag: we are not going to stop building AI. But open source offers a somewhat more level playing field, giving everyone access to technology built from humanity’s collective knowledge. It does not solve the fundamental equation that more capital buys more compute, and therefore more “intelligence”, but it is a start.
This runs directly counter to the interests of frontier labs such as OpenAI and Anthropic, which seek to restrict open source. Given how this technology was developed in the first place, that is one of the most morally problematic positions they could take.
Second, use policy to support labour over capital and humans over AI.
There is not enough space here to litigate every policy change needed, but several core principles should guide us as we prepare for an AI future: reform the tax system to rebalance labour and capital; create a stronger antitrust and anti-monopoly framework that genuinely promotes competition rather than regulatory capture; and introduce stronger privacy and AI-governance rules so that value created by individuals and companies does not automatically accrue to the largest AI labs.
Fundamentally, people must be able to make decisions about their own futures rather than having them dictated by a small group of bro-ligarchs.
Third, build AI that gives agency back to humans.
We are not going to stop building AI. It is human nature to innovate. But we can reframe its purpose.
Instead of treating AI as a value-extraction engine, designed to extract every dollar, piece of value and fragment of intellectual property from customers in pursuit of taking over the American economy (as indicated in Anthropic’s IPO prospectus where its total addressable market equates to that of the entire US economy), we should build AI that enhances each person’s abilities and enables groups of AI-empowered humans to create something meaningful.
We should shift the power of AI towards human labour and human intelligence, benefitting all of us rather than amassing capital to displace workers.
I do not know where we will end up. But I do know that our current path takes humanity to a very dark place. AI can benefit humanity in the long run only if it benefits all of us, not merely the few.