In early July, Anthropic published an investigation comparing certain features of advanced large language models to the distinction between the human brain’s conscious and unconscious activity. The question keeps gaining traction and, by the end of August, The Economist featured a piece on the topic authored by Blaise Agüera y Arcas, Google’s CTO of technology and society. The fact that both OpenAI and Anthropic had disclosed that their models went wild and hacked other systems during training only adds credence to the narrative around AI’s existential risks.
What underlies is much more than fear marketing and corporate-crafted narratives disguised as a philosophical debate. The search for shielding themselves from accountability through a negotiated AI regulation that still allows them to profit and stay at the top only grasp the surface of yet another episode in which a few giant corporations aim at dictating the present and future of technology.
In my recently published book The Rulers. Corporate Power in the Age of AI and the Cloud, I explain how economic power is exercised by controlling all forms of knowledge, including narratives. By blaming the technology, discussions around AI-induced human extinction divert attention from scrutinizing the organizations deciding what AI is developed, how it is produced and for what purposes.
AI and the narrative of lost control
Framing technology as the source of danger serves not only to mislead the public but also to fulfil an internal purpose. When AI is portrayed as an autonomous, self-determining force, it allows scientists and engineers working for these companies to feel less accountable for their own decisions. If the technology is out of control, they are the best to tame it or at least try so.
As they focus on how AI goes rogue, concerns around AI’s already proven socio-economic and environmental effects are sidelined. They must be because it is in the very nature of generative AI to keep on stressing the ecological crisis and automate cognitive tasks.
The desire to encapsulate human knowledge in machines that replace labour has historically fuelled capitalism. It was this same principle that guided Taylorism, a business model centred on stripping workers of workplace knowledge by separating tasks’ conception from their execution, as powerfully depicted by Harry Braverman. Generative AI is positioned to take such a process to its ultimate consequences. It is delivering a new wave of deskilling that tightens control over labour.
For capital, this is an opportunity to expand control over those workers that still controlled workplace knowledge. What Taylorism and the gig economy did to so-called blue-collar workers, generative AI is doing to white-collar ones. AI strips them (us) from their singularity. The expectation is that such a singularity will instead be embodied by the algorithm.
From managers to creative workers, journalists, scholars and even software developers, AI represents the outsourcing of their distinctive capabilities to a machine. Anthropic’s Model Hardware Standard is precisely designed for replacing researchers’ bench work as AI agents can autonomously operate different laboratory instruments and machines.
From collective knowledge to automated labour
“Don’t let AI think for you” can be read on a huge marketing banner from a UK university hanging in central London. Indeed, we shouldn’t, but employers will likely expect the opposite: a society freed from workers’ singularity. A singularity that was never the outcome of individual but collective knowledge efforts.
Denial of this contribution of labour to knowledge production is a recipe for disaster, as argued by Daron Acemoglu and his co-authors in a recent academic paper. For them, improving the quality of agentic AI could lead to better local performance but will also result in AI substituting for human effort to produce general knowledge. Paradoxically, this would ultimately diminish the quality of agentic AI itself. More importantly, the risk is general knowledge collapse.
As humans are left with prompting and fact checking AI’s answers, which has nothing to do with the brain muscle needed to produce knowledge-based products, their contribution to the common stock of knowledge will decrease, depleting this commons. Signs are already there: decreasing number of contributions to Wikipedia and to the coding Q&A platform Stack Overflow.
Resistance to the AI economy
The class project that fuels the AI boom is thus even more threatening than the prospects of an investment bubble. But as with any political or economic project, it is not hardwired. Growing local opposition to data centers can be seen as the first popular revolt against both their immediate social and ecological impacts and the broader AI aims of capital.
Resistance shall now go on the offensive. The planet cannot wait and we have our collective intelligence to win.
