I think we should view AI progress and the deployment of AI systems through this lens: when technical advances unlock new capabilities from already scraped data or when institutions begin collecting data to ingest into AI R&D pipelines, these acts should be seen as forcibly conscripting more of humanity into contributing inputs to the production function of a quasi-public information good. By "quasi public good", I'm referring to something that is currently a privately managed club good, is distributing utility in a way that kind of looks public goods-ish, and has the potential to either (1) maintain this shaky equilibrium, (2) become fully enclosed or (3) become fully public.
When we try to imagine AI's impact on tasks and jobs or try to identify generalizable insights from historical parallels (factories, the transition from horse-based transit, the mechanical loom, etc.), we should also consider some thought experiments that center digital "quasi public goods". What if all of humanity spent a year working to make every piece of software open-source (but managed by a private entity)? Or, what if every medical case was uploaded into a StackOverflow-style Q&A platform (that was managed by a private entity)? Or, what if every company in the world forced their employees to conduct all knowledge work by interacting through a Q&A forum? Before I expand on the value of these thought experiments, let me first clarify the claim about everyone being "conscripted into the production of a quasi public good".
The quasi-public good I'm referring to here is the collective model weights of all frontier labs and the actionable knowledge and models-of-the-world that are compressed in these weights. To put it another way: because of progress in information processing, any work act now has the potential to contribute to the tweaking of what -- being a bit grandiose -- we'll call the Grand Weights. But the Grand Weights are being governed by private organizations with incentives to try to capture some of the economic winnings from the goods (though there's a real chance that AI operators struggle to capture these winnings).
In short: the AI field is acting to create a pooled informational asset from diffuse human contribution, and the knowledge compressed into them has public-good-like properties -- and the potential to, with public subsidies, produce a public good.
Information goods are often easier to implement as public goods compared to physical goods because of the properties of information. The reason that I think we should see frontier AI models as quasi public goods is because the information in model weights is mostly non-rival (though there are some rivalrous elements that come up in alignment -- people may try to alter an AI model to meet their preferences in a way that lowers utility for others), but model weights require great capital expenditures to "turn on", and depending on how various battles in AI governance and data flow play out, it may or may not be easy for model weights to be effectively enclosed (i.e., make AI more excludable than it might otherwise be).
Returning to our thought experiments around pouring resources into, or conscripting workers into the production of public goods: we should imagine likely economic impacts that might come about just from applying massive amounts of resources to digital commons themselves. For instance, what would happen if everyone on Earth had dedicated a full year to support StackExchange? Almost certainly this would affect certain jobs in complex fashion: at first, it might become much easier to get started as a domain-specific programming consultant (e.g., a specialist in the web platform Django helping businesses fix issues with their web apps) -- but when the Q&A platform got really good at some point you would lose your business!
Of course, humans are adaptable and so you might pick another domain to work on. But recall, in our thought experiment -- the entire planet is pouring resources into making this public good excellent; soon, the second domain you picked might also become saturated, and you'd have to move on again.
This is, to me, a much more apt metaphor for what AI is doing to knowledge work now (albeit with a lot of jaggedness and rough edges, and serious potential for feedback loops that do actual harm to collective knowledge in certain domains).
I think this perspective can help surface and clarify certain tensions that are left unstated when people debate AI's economic impacts and especially the moral status of advancing AI capabilities or arguing for a certain data paradigm. For instance: sometimes putting knowledge into a public commons can reduce economic leverage for other people who have some similarity to you (e.g., if all your friends are Django consultants and you spend a year making the best Django wiki of all time, you've done an amazing thing for humanity but all your friends might have to find new jobs).
This perspective can also help explain why well-informed and well-intentioned actors on both sides of the debate (e.g. folks more aligned with an "AI is theft and working on models under the current paradigm is anti-social" take versus those aligned with a "working directly on frontier models is extremely moral and pro-social" take) disagree so much. The ultimate moral impact will be driven by how the quasi public good-status resolves. If we do disrupt all the Django consultant jobs, will we end up with a thing that looks like a true public good or will the enclosing walls go up?
I think this is an extremely high-stakes question, and the discussion of how to create a healthy resolution of the quasi-public-good-situation is critical. I'll end this post here, briefly noting that my preferred solution space is still heavily focused on supporting collective bargaining for information as a source of friction and countervailing force in the meantime.
I plan to keep iterating on this post to try to make the core point more concise: I'm feeling fairly confident that this perspective is important (and legitimately undervalued -- though I might be missing discussions, perhaps some in the economics space, where these ideas are coming up) and I want to try to socialize this idea. Let me know what resonated or what you disagree with!
See also this longer blog post with a lot more detail: https://dataleverage.substack.com/p/the-paradox-of-reuse-in-2026-a-case