The Hidden Social Aggression of AI Progress
"Violence is present when human beings are being influenced so that their actual somatic and mental realizations are below their potential realizations."
— Johan Galtung
The main struggle of human civilization has always been competition for resources. For the vast majority of our history, all resources were finite and the only way to reallocate them was to take them by force. Technology has expanded the resource pool, and it has given us a new peaceful mechanism for re-distribution. But the ugly truth is that technological cycles still involve a forceful exchange, albeit more subtle and organized.
AI progress, despite its promise to be the ultimate abundance mechanism, is predicated on a tradeoff. In order to make the economic machine more efficient, we need to collectively give up automatable human activities and reorient our productivity towards newly scarce activities. The promise is bold: cure of all disease, reduction of manual labor and prosperous lives for all. But at the core of this re-distribution is the fact that we, the participants, have no real alternative but to comply. The change, no matter how noble its goals, is forced upon us, like many wars have been forced onto our ancestors.
Even if you agree to this new social contract of progress that justifies economic hardship for some, even if you are willing to accept the lopsided economic reality where AI labs capture most of the value, the lack of democratic consent should disturb you. AI is fundamentally ungovernable by society, and its perpetrators not only de-emphasize democratic norms, they are insisting on the necessity to reject them.
I don't envy the position that the labs are in, and for the record I think most of them are good actors, especially Anthropic. The problem is that, for the first time in recent history, economic agents are forced to be the stewards of critically important infrastructure instead of governments. This creates a natural tension where society is at the whim of profit-maximizing institutions, instead of governments incentivized with broad social welfare.
This becomes especially alarming when considering the fact that AI is such a powerful technology that it can be used to coerce action. Any functioning society needs a separation of powers, even if its simply between the ideologists and the executors. The government, for obvious reasons, is the ultimate executor, anointed and legitimized by voters. AI labs should be active participants in shaping the discourse, especially due to their access and insight to the ever-evolving capability frontier. But it gets very dangerous when AI labs can enforce their beliefs through control of the technology, which today is exemplified by the precipitation of ever more dangerous models, while actively curating how those models are being used.
That is why Anthropic's decision to implement "new interventions that limit Claude's effectiveness for requests targeting LLM development" is such an alarming development. Of course, safeguarding AI use is not an entirely new phenomenon. Labs have always restricted model use through explicit limits or ToS. This, however, is the first time that a productive, universally beneficial activity (access to technical education) is broadly and actively forbidden. It's reasonable for Anthropic to protect against competitors and bad actors, but actively limiting the use of this technology for education, especially in the context of releasing admittedly dangerous models, is an infringement on liberty.
The current narrative goes something like this: "The new models are powerful, and they can be used by bad actors to do bad things, but we want to release them anyway, because if we don't someone else will and oh, by the way, we are going to release it only for use-cases we think are valid, and not wait for the government to weigh in". Anthropic is a victim of its own success, and they should be applauded for the technological progress that they have been able to achieve, but actively advocating for the dangers of a model, while at the same time YOLO'ing it out of the door, while being prescriptive about how to use it and without any regard for government is an act of social aggression.
Dario is well aware that his company is crossing a chasm with profound implications for social power dynamics, as shown by his extended apology letter that can be best summarized as "the government wouldn't do it so we did it ourselves to the best of our ability". It is one thing to define the ideological boundaries of reality, and it is a totally different thing to enforce this reality by acting on it. Anthropic is actively creating a problem, then using it to justify a totally new world order, accompanied by a moral system that favors those who are aligned with their own views.
The nature of AI development is what makes this particularly troubling. We have no real mechanism to ensure that access to this resource is evenly distributed to society. A useful lens for viewing AI labs is that of economic agents that acquired access to a scarce resource in the form of AI researchers and produced a good that is not replicable. Monopoly law doesn't view human resources as a scarce good, but had AI labs relied on, say, a critical mineral, it would be very easy to regulate and gatekeep it for the benefit of society.
Despite some sensible solutions proposed by Dario, I, for once, think that the problem is not in the slowness of government action, but rather in the aggressive approach of giving the world technology that it is not yet ready for, and then allocating it according to views of an economically driven enterprise. Labs have become producers of customer-facing applications due to the financial realities of AI development, but it doesn't have to be that way. There is room in this world for labs that develop this technology within closed walls, with a much better mechanism for capturing value, that is both beneficial to society and financially solvent.