On AI and knowledge problems; (Neoliberalism is dead; Long live Neoliberalism!, pt 3.)

In the dog days of Summer, in a very stimulating Substack post (that is worth re-reading),

Carlo Ludovico Cordasco

(Manchester) made a useful distinction between two features of markets:

What we usually praise as “the magic of markets” is really two distinct engines working in tandem (on this see Hurwicz, 1972; and Maskin, 1999). The first is knowledge discovery: millions of agents, each holding local, often tacit information, reveal slivers of what they know simply by making offers to buy and sell. The second is error correction: when someone guesses wrong (e.g., plants grapevines nobody wants, orders too much saffron, misprices a battery pack) profit and loss push them to revise their plans or exit. Prices are just the interface where those two engines meet.

The first of these is usually associated with Hayek’s (1945) "The Use of Knowledge in Society." (The second also has a Hayekian character but surely predates him.) Now that 1945 Hayek paper was also an argument in what during the 20th century came to be known as the ‘socialist calculation debate.’ (Keep that in mind.) It is also central to what is known as Hayek’s ‘knowledge problem.’ (Keep that in mind, too.)

Carlo, then went on to discuss the potential roles of AI in the two market engines. He concluded, “I think AI could potentially do well in [knowledge discovery] but I’m agnostic as to whether AI will be able to bring about [error correction].”

Now, I have gotten to know Carlo a bit, and on his own website he advertises himself as an “Affiliate Scholar at the Classical Liberal Institute at the New York University School of Law.” So his style of argument surprised me a bit. And on one of the social media websites, I jokingly suggested that he had been anticipated by Lenin (recall this post on the role of cartelized banks in centralizing information processing) and was becoming a market socialist (a la Oskar Lange).

Much to my pleasant surprise and without, I suspect, awareness of my comment,

Andrea Genovese

(Professor of Logistics and Supply Chain Management at Sheffield University Management School), wrote a fascinating response to Carlo, where he immediately got to the point:

What strikes me about Carlo's analysis is how closely it aligns with a market socialist perspective, one that sees markets not as sacred, but as contingent institutional tools that can be supplemented, simulated, or partially replaced if and when better mechanisms arise. In fact, many of the challenges Carlo raises, and the hybrid solutions he envisions, fit comfortably within the framework proposed by Oskar Lange and others who sought a synthesis of planning and decentralised feedback.

It’s very much worth reading the rest of Genovese’s post, too, because he gives a nice history of developments in market socialism influenced by Lange (quite a bit unfamiliar to me), including grappling with the experience of the failures of central planning. (As regular readers know I think very highly of Lange as a theorist — not as a politician, of course.)

But mediated through Carlo, Genovese here grasps, first, something rather important about the new (polycentric friendly) classical liberalism that is developing in our age (recall the first post (here) in this series; and here the second.) This point was also already lurking in the impact of the Ostroms’ Bloomington school on this new classical liberalism; the new classical liberalism is neither anarchist nor market-fundamentalist anymore. Rather, as I noted, alongside state and market it explicitly takes The Third Sector very serious as a site of voluntary collaboration and coordination.

But, also, second, and more important, where to draw the line between market and other forms of coordination and production is going to be context sensitive and subject to (preferably bottom up) social experimentation. The market is, no less than the nature of property, a creature of local norms, rules, institutions, and government(s)s’ activity.

These two points are very explicit in the 2019 book (by Aligica, Boettke, and Tarko) Public governance and the classical-liberal perspective. that I discussed at length in my first post.) So Carlo is here very much in the spirit of the new classical liberalism. (But as I noted, this social experimentalism has in some respects a friendly relationship to features of democratic pragmatism.)

Genovese then notes in his post:

we don’t need to abolish markets to acknowledge that AI can shift the boundaries of where and how they are needed. As Lange anticipated, if informational conditions change, the case for market reliance changes too.

And those conditions are changing. AI-based planning is no longer theoretical: recommender systems already allocate attention and goods in digital media and online platforms; large multinational companies rely on sophisticated OR algorithms and to forecast and optimise production planning, inventory management and routing of goods across global supply chains (Phillips & Rozworski, 2019); smart energy grids adjust production and consumption in real time based on predictive models. These are not thought experiments: they are partial, algorithmically mediated planning systems embedded in market economies.

I do think that at the right level of description there is indeed some convergence between the updated market socialism and the new classical liberalism (and some variants of pragmatism). This is especially so because Genovese has no “nostalgia” for “bureaucratic planning.” This convergence is not so odd because both movements learn from the same historical experience; and because market socialism and the new classical liberalism share a vibe in theorizing in terms of the benefits of epistemic institutions.

Unlike Carlo, Genovese thinks that AI can also facilitate some error correction:

A particularly relevant field of study is algorithmic mechanism design, which explores how multiple AI agents can engage in resource allocation through incentive-aligned protocols. In distributed algorithmic mechanism design, agents interact directly to reach equilibria without centralised control, suggesting that coordination and error correction could emerge from peer-based feedback rather than traditional markets.

Of course, letting AIs gamble with other people’s money is not what the new classical liberalism signs up for! (Again, they are not my team, but I am clearly often in harmony with them.) But in fairness to the market socialists, ordinary capitalism sees Wall Street players do this with people’s savings and pensions all the time (as new classical liberals will tell you at once).

I want to close with three observations.

First, I wouldn’t want to defend the idea that the ‘magic’ of markets is limited to two engines that Carlo emphasizes. I have in mind two ideas: (i) markets, especially when they are competitive, are a means to combat concentrated power. (Of course, there is also concentrated power in some markets and markets can be sources of rents; and some states will terrorize or extort market participants. So I don’t mean to suggest that this is always effective.) But then again the polycentric new classical liberal at least has a consistent story to tell about combatting concentrated power (even if I wish they took anti-trust and limitarianism more seriously); the market socialist does not.

And (ii) market processes and market-like processes also sometimes constitute entities as emergent effects (i.e., prices). This is an idea one can also find lurking in the vicinity of Hayek’s ‘knowledge argument.’ As I have noted (recall) classical liberals seems to have gotten the idea from Renan.

Alas, I am worried that replacing markets with AI in the current property rights regime will actually facilitate concentrated power. I have already suggested that given the massive property thefts in building LLMs, society should not protect such AIs as property (recall). It is an open question whether all the salient emergent properties that market processes produce — not the least social identities — are also available by analogy to LLMs.

Second, and interestingly enough, Genovese speaks of a third idea too: (iii) using the market to ‘discipline’ economic activity and market participants. Some market-friendly liberals (going back to Smith) emphasize this, too (as Foucauldians note critically). I was surprised to see a market socialist embrace it. Genovese primarily wants such disciplining to be directed toward “social needs”—it’s not wholly clear what he has in mind, but I am worried it would be oppressive like the Chinese social credit system.

My final, third thought is this. At the core of contemporary neural networks is a structural opacity (lack of explainability, black-boxing, etc.). As is well known, this generates all kinds of alignment problems. Such structural opacity can — on the Hayekian picture embraced by market friendly liberals — also be found in the market process (at a certain level of description). The analogy (it is no more than that) is not wholly incidental if one is familiar with Hayek’s The Sensory Order. (About that some other time more.) But this reminds us that new classical liberals are not unaccustomed to submit to an opaque process if it can be shown to meet our needs, respect moral equality, and be liberty enhancing.

To be continued.