Streetwise Professor

Back when AI began to emerge as the Next Big Thing, I asked myself “how long until some idiot comes along and says AI will permit efficient, rational central planning because it solves the Knowledge Problem.” Well, the day has arrived and found its idiot: Martin Sandbu in the Financial Times.

Sandbu at least correctly frames the issue in the context of a long-ago debate that is largely forgotten (something that he acknowledges): the Socialist Calculation Debate of the interwar years (that extended into the 1940s: Hayek’s seminal paper that largely settled the controversy was published in 1945).

In many respects the debate was a very abstruse one that was rooted in the technical economics of the era. Starting with Leon Walras, it was shown that under certain technical assumptions (some of which I will get back to), prices in a competitive economy would lead to an “efficient” (i.e., Pareto optimal) allocation of resources. But there are many Pareto optimal allocations, with very different distributive consequences (c.f., “the Edgeworth Box”).

This was held out as a demonstration of the truth in Adam Smith’s idea of an “invisible hand.” Years later the mathematical rigor of the idea was strengthened greatly by Arrow and Debreu and the theory of general equilibrium that led to the First and Second Theorems of Welfare Economics, showing the duality between competitive equilibrium and Pareto optimality.

The socialist economists-most prominently Abba Lerner* and Oskar Lange-tried to flip this on its head. Walras’ modeling artifice was what came to be called the “Walrasian auctioneer.” The auctioneer would call out prices. If the markets didn’t clear at those prices, he would adjust them via a “tâtonnement” (“groping”) process until he found the price vector (the vector of the prices of everything in the economy) at which all markets cleared, i.e., the quantity supplied of everything equaled the quantity demanded. This market clearing allocation would be Pareto optimal (i.e,. “efficient”).

Abba Lerner.

Lerner and Lange (which sounds like a Tinpan Alley duo or a pair of movie moguls) seized on this idea, and said that a central planner could set the prices. These prices could achieve a Pareto optimum, and what’s more, could achieve some more socially desirable (by whom?) allocation of goods. A visible hand could do better than the invisible hand. Hence, central planning was not only feasible, it was superior to market outcomes. Alternatively, the central planner could choose Pareto optimal quantities (corresponding to the desired allocation) and let markets clear at the prices implied by them.

Enter von Mises and Hayek. They argued that the modeling abstraction originating with Walras was totally divorced from reality. Most importantly, the information the central planner would require is vast and unobtainable. It is dispersed among millions of economic agents in the form of their preferences (which are different for everybody) and the technical transformation possibilities in the (global!) economy. Much of this information is highly localized. Much of it is tacit and not articulated, but only embodied by “human action” (human choices based on information and their environment).

Frederich von Hayek

In von Mises and Hayek, emergent market processes elicited this information, and market prices summarized this information. For example, by choosing how much to buy at a particular price a consumer reveals information about his preferences. By choosing how much to sell at a particular price a supplier reveals information about cost (which can depend inter alia on tacit knowledge about his production technology). Interactions in markets produces prices at which market clear. These prices aggregate and reveal vast amounts of information about preferences and technologies that no central planner could hope to collect and then act upon.

This pretty much put paid to the socialist calculation debate. The epic failures of communism and socialism (not just in the USSR or Cuba, but in places like India) provided definitive empirical evidence.

Enter AI, and Martin Sandbu. AI can scrape vast amounts of information about prices and quantities. What’s more, “AI offers the promise of transcending price information altogether by predicting who wants, needs or can produce what at what cost, better than market prices could ever summarise.” Spooky behavioral algorithms are evidence, he says, of the uncanny power to uncover individual preferences.

Welcome to the Brave New World!

There are so many factual and intellectual errors in his essay-and indeed, even in the single sentences I reference-to render its conclusions utterly risible, and which make it an impossible labor even to list in detail. So I will just hit a few salient points.

First, even the vast amounts of data AI can collect and organize and presumably use to make decisions represents only a small fraction of the information required to make efficient decisions. For example, yes, we can collect information that is relevant in predicting some individual decisions, e.g., I might be interested in buying a new car, and based on my income, buying history, where I live, etc., AI might be able to predict which cars I prefer. But what about a very fundamental choice: my preference for leisure vs. consumption of goods and services? That is a very personal, internal thing. What’s more, it might change from day to day. And there are many such preferences.

Arguably things are even more complicated on the supply side. In a highly complex modern economy production technologies are incredibly intricate, so intricate that even for a given firm it is a challenge to articulate them. And these two change stochastically. A machine breaks. Someone figures out how to use the machine better. A technical innovation may render the machine obsolete, but the innovation’s actual effects can only be determined through use-“learning by doing”-with said learning being largely tacit, unarticulated, and embodied in the minds of the doers.

Information flows like a river. AI is bound by history-the data (information) it was trained on. It cannot possibly keep up.

The whole literature on the economics of organization is based on the understanding that information in organizations is (a) highly diffuse, and (b) private. Employees know a lot that their employers don’t. Accounting systems and compensation schemes, to mention just two organizational functions, are attempts to elicit information and mitigate the effects of information asymmetry. This testifies to the ubiquity of the issue and the fact that information isn’t just lying about to be collected and collated by a bigger central entity than a firm. If firms struggle to obtain it, why would AI find it a breeze?

Same goes for financial markets. Again, the concepts of private information asymmetric information are fundamental in finance, both in corporate finance and asset pricing. The “price discovery” role of financial markets basically means that trading in financial markets determines prices that aggregate vast amounts of highly dispersed and often tacit pieces of information. Ai trading algorithms may be able to extract some of that information from prices (an “inverse problem”), but absent markets, it cannot access it to determine optimal allocations of capital (a “direct problem”) because it cannot collect this information unless it is produced through human action. It therefore cannot use this information to direct human action.

The problem here goes beyond financial markets. One of the things that Sandbu claims AI can do is “scrape prices” from vast numbers of transactions in order to extract information about preferences and technology. Er, if AI is making resource allocation decisions, those transactions will never occur, and AI can never get that information.

So no Marty, AI will not “bury Hayek.” Yes, it has more information than Gosplan did in the day, but it is still only a sliver of the information that humans use in deciding how to utilize the scarce resources at their disposal.

Electricity provides an interesting illustration. In the past 30 years or so, electricity systems around the world have adopted methods that resemble in crucial ways the type of “market socialism” central planning of Lerner and Lange, and the auctioneer mechanism of Walras. In the US, “Regional Transmission Organizations” or “Independent System Operators” operate centralized auction markets whereby market participants enter offers to supply power and bids to buy it. These are highly granular, given the complexity of electricity. They vary by hour of the day (sometimes down to 5 or 15 minute intervals) and location. The RTO/ISO then put these bids and offers into a mixed linear integer programming algorithm to determine generation schedules and loads. Since fundamentals can change minute-to-minute, the RTO/ISO updates “dispatch” which differs from pre-established schedules.

This is kind of feasible in electricity because the available resources are well known, as are the physics of the grid and the behavior of consumers (e.g., load goes down when the Super Bowl ends because millions of people switch off their TVs). But even here the central planners’/operators’ systems have been plagued by periodic failures, sometimes on an epic scale (e.g., California 1998-2000). Moreover, they face challenges because (a) they still lack much information dispersed among millions of market participants, (b) there are technical issues, like “non-convexities” (one of the issues I alluded to above) that require kludges to mitigate, and (c) the systems can sometimes be gamed.

This last point deserves emphasis. Central planners are in a game with human beings. Human beings will try to figure out and predict–perhaps using AI!–the central planners’ decision rules. They will they see how they can manipulate the rules to defeat the central planners’ purposes.

Put differently. Electricity represents the best case for market-like centralized allocation mechanisms. And even there it poses huge challenges.

Goodhart’s Law also comes into play here. Once the AI measurements become used as inputs to targets, they cease to become effective measurements. That is, their informational value declines. The socialist calculators implicitly assume information is exogenous. It is not, it is endogenous. And changing the system changes the information.

Finally there is a huge issue that Sandbu touches on only indirectly, when he talks about the political aspects of AI central planning. Specifically: determining a better, or more “rational” allocation of resources requires an objective function, a social utility function if you will. In the context of economic history, which of the Pareto optimal allocations in the Edgeworth Box will AI choose? Who determines that objective function?

Rationality in economics involves the choice of means to achieve given ends. Whether a given action is “rational” depends on how you evaluate the ends it achieves.

Which brings us to another fundamental problem with AI Utopia–even assuming the difficulties above could be surmounted (which they can’t, but still). That is: what is the “social welfare function” that a presumed panopticon AI will maximize?

Here Ken Arrow makes another appearance, via the “Arrow Impossibility Theorem,” which proves that there is no such thing as a social welfare (utility) function that meets several basic axioms. Crucially, one of the axioms is that the function is “non-dictatorial.” Yeah, you can have a social welfare function–the dictator’s.

So who gets to be the dictator? Do you want to be subject to his (or her) whims? Or maybe the “dictator” is a democratic polity. Yeah, those are totally “rational.” And then the central planning machinery would be the focus of political battles and rent seeking. And given the stakes, the battle would likely result in violence and oppression.

Oh, and this dictator would have to be global in order to achieve the promised utopia. The UN in charge of all powerful AI. That would be just swell, I’m sure.

I knew–metaphysically knew–that it was inevitable that AI would be advanced as a solution to the Knowledge Problem, and therefore the ultimate innovation that would make socialism possible. (As I recall, early on Daron Acemoğlu mused along these lines-so much the worse for him). All this demonstrates is the superficial understanding these people have of the Knowledge Problem, and of the nature of information-knowledge-in the economy.

The information necessary to allocate resources efficiently is far deeper and far more extensive than AI can collect, let alone analyze. Moreover, even beyond heroically assuming that problem can can be overcome, it creates a fundamental political problem. What will an AI overlord use that information to maximize? What will be its criteria for choosing how to use that information? Whoever determines that will inhabit the pinnacle of power. And we know that struggles for that power are inherently violent and repressive.

* Weird connection here. Lerner taught at Roosevelt University in Chicago. My thesis advisor, Lester Telser, was a Lerner student as an undergraduate in the 1950s. Lester went on to get his doctorate at Chicago, and was definitely NOT a Lernerite. As a grad student I taught some courses at Roosevelt, also weird because it was still a leftist hotbed.

I had two immediate reactions to the news of Jason Arday’s apparent suicide. The first was sadness, especially for his two children, wife, and parents. The second was to start a countdown to when the left would blame his decision to end his life on those who had uncovered his serial academic misconduct and fabulism. (Uncovering is not quite the right word: it was there in plain sight all along. Having the temerity to point it out is more accurate).

The countdown was a short one. The blaming started almost immediately.

But the blamers should look in the mirror. They are responsible for Arday’s unwarranted rise and are thus responsible for his inevitable fall. They are a necessary condition for the whole sorry tale.

The British academic establishment is the guilty party here. His thesis advisor(s) who granted him a doctorate based on a shoddy thesis: even putting plagiarism aside, its intellectual vacuity, borderline literateness, and sheer sloppiness (apostrophes go wild!) disqualify it even in a notoriously weak field like education. The universities that hired him, and used him to virtue signal despite his obvious lack of scholarly ability and flagrant fabulism: Cambridge is the most guilty here, because the gap between its scholarly reputation and Arday’s scholarly unfitness was the greatest. The university adjacent in media (the BBC in particular) and politics. Supposedly reputable publishers who advanced him seven figures for his Baron von Munchausen-esque autobiography.

Think of the likely effects of this on someone like Arday. What else could he conclude from his treatment other that there was no penalty for academic misconduct; that schlock research on racism got him treated like an academic superstar; and that there was an audience that not only suspended disbelief but avidly gulped down his tall tales? Given that feedback, the natural thing to do is more of it.

Which eventually and inevitably attracted the skeptical attention which brought it all crashing down. His failings were so immense that even his most ardent promoters found them indefensible.

Unless he was totally delusional, Arday must have awakened every day asking: “Is this the day I get called out?” I can only imagine the psychological toll that took. And when that day came, he decided he couldn’t go on.

Of course Jason Arday is ultimately responsible for his choices. But the “elite” enabled those choices, when at every step they should have given him a reality check. And they enabled not because they cared about Jason Arday. It was all about them. To advance their agendas. To validate their intellectual conceits. To feed their vanity.

Decent people would consider their role in Jason Arday’s rise and fall, and repent. But instead, they blame those who would not countenance misconduct–misconduct that they enabled, and indeed encouraged.

It’s a story as old as time. An allegedly genius investor has a winning streak, which burnishes his genius image even more. He starts to believe it. Levers up. Then the wave he was riding crests and falls, the iron logic of leverage exerts itself. He becomes illiquid-if not insolvent-and either goes bust or gets bought out. And he’s no longer a genius, but an excuse maker.

One particularly famous example of this is Long Term Capital Management. LTCM had bona fide geniuses, Nobel Prize winners Myron Scholes and Robert Merton. Their genius apparently paid off for quite a while, with LTCM showing stellar risk-adjusted returns. Then, you know, stuff happens. Stuff like Russia effectively defaulting on its debt. Then your supposedly uncorrelated trades become suddenly correlated, and all start going in the wrong direction.

All that leverage that amped your returns now turns on you with a vengeance, and amps your losses. What’s more, your success has allowed you to grow to the point where you effectively are the market, which (a) moves the markets strongly against you when you try to liquidate some positions to meet margin calls and/or reduce leverage, and (b) pushes up the correlations even more.

Then, you are transformed from Golden Child to Systemic Problem Child. In LTCM’s case, it was so large that the Fed had to intervene to coordinate a liquidation of its positions into the tender loving care of the likes of Goldman.

Fast forward 28 years. The Golden Child and putative genius is Leopold Aschenbrenner, head of the oh-so-ironically named Situational Awareness hedge fund. The soi disant “Nostradamus of AI” implemented the picks-and-provisions strategy in the AI space, investing heavily in stocks of companies that provide the inputs to data centers and other AI-related activities. (SA also has private investments in AI firms like Anthropic).

He did time his entry right, and rode prices upwards. He (sound familiar?) amped his returns through extensive leverage (reputedly 4x). But-and there’s usually a but-the market peaked, and began to drop. And as has happened time and time again (it wasn’t just LTCM) the iron logic of leverage took over. Wrong way market moves beget margin calls on leveraged positions. Margin calls mean cash. Mark-to-market paper gains aren’t cash. So the imperiled fund was forced to sell positions to raise money-which only exacerbated the price declines.

Worse, the financial distress of a big position holder can’t be kept a secret. Evidently many others were apparently more situationally aware than Situational Awareness, and started to unload positions in stocks in which it was heavily invested. This front running exacerbated the price declines, and hence severely leveraged SA’s distress.

It happened so fast!, Aschenbrenner is no doubt thinking.

Sure did, (ex) Golden Child. Leverage comes at you fast.

And so do sharks. Jane Street, Millennium, and Citadel circled. In a single night, the last of these negotiated a deal to acquire all of SA’s publicly traded holdings at a discount estimated to be 10 percent.

The markets breathed a collective sigh of relief when news of Citadel’s purchase eliminated fears of a dumping of SA’s shares on the market: tech stocks rallied on the news.

Though no doubt his motives were purely self-interested, and are not as public spirited as J. P. Morgan’s intervention in the Panic of 1907, Citadel’s Ken Griffin did the markets a big favor. It is not from the benevolence of the butcher, the brewer, or the baker–or the hedge fund mogul–etc. (NB: I am not a reflexive Ken Griffin fan, as this old post demonstrates).

But as the saying goes, no good deed goes unpunished. Many-mostly Leopold fanboyz apparently-have argued that Griffin plotted SA’s demise (a reprise of CZ and SBF, maybe) by releasing a note predicted a Fed rate increase some time before the last Fed meeting. When the Fed did meet, and held rates steady-not raise them, mind you-the markets dropped, delivering the coup de grace to SA’s strategy that was already bleeding profusely. So per the fanboyz: See! Citadel caused this and then swooped in to take advantage!

Er, no. SA was already teetering. Citadel’s prediction-wrong, in the event-did not cause the final, fatal sell off. The Fed did.

For his own part, in a letter to investors Aschenbrenner rounded up the usual suspect-short sellers! It’s kind of pathetic. I have more respect for the natural gas fund operator who gave a blubbering, cringing apology when his widow maker strategy imploded.

If you are looking for other culprits here-beyond Aschenbrenner-look at SA’s prime brokers. Why did they supply so much leverage to back a poorly diversified strategy focused on a sector where sentiment had started to turn prior to the initial declines?

And if you want to do an It’s a Wonderful Life homage and wonder what would have happened if Ken Griffin/Citadel hadn’t been born, look at the disastrous unwind of another highly leveraged hedge fund-Archegos. Another example of an undiversified, highly leveraged fund caught in a doom loop when the markets turned against it.

In that instance, its prime brokers got together and agreed to assume the fund’s positions, and undertake a controlled liquidation. So far, so good. But one of them-Morgan Stanley-played “defect” in the prisoners’ dilemma facing the PBs, and sold $5 billion of Archegos’ holdings. Then the others jumped in and a full-scale fire sale occurred, leaving all the PBs badly singed. With Morgan Stanley (a la Margin Call) coming out best by getting out first (and not telling its counterparties what it knew about Archegos’ impending demise).

One firm-Citadel in this instance-didn’t face a prisoner’s dilemma. It-with the support of SA’s prime brokers-took the whole portfolio. Fire sale averted-to the markets’ relief (as noted above).

In sum, the Situational Awareness story is news, but not news. It is a story that has been repeated many times in many places. Leverage makes you look like a genius when the markets move your way. But with probability bounded away from zero, it will kill you when the market turns.