InCognito

In 1952, Hayek published The Sensory Order: An Inquiry into the Foundations of Theoretical Psychology. The aim of his work was to describe processes by which the neurological system classifies sensory inputs. He asserts that the mechanism's he describes support human capacity for higher order classification and classificatory structures.

Perhaps Hayek did not fully understand at the time of his writing that he was coparticipating in the computing revolution. The authors that he drew from - e.g., Ross Ashby, Norbert Weiner, and Walter Pitts and Warren McCulloch - were leading theorists in cybernetics. Pitts and McCulloch, in particular, provided a theoretical map of neurons that allowed for neurons to be treated as deterministic machines - i.e., computers. Paul Lewis notes this tension that Jack Birner and Phil Mirowski view Hayek's work as suffering from "reductionism [Birner]" that seeks to "reduce thought to mechanism [Mirowski]" (Lewis 2016, 143-144). Lewis disagrees, but in order to do so, he must appeal to emergent phenomena as exerting causal force of their own. He argues that Hayek championed the same sort of view:

Hayek endorses a hierarchical worldview, according to which individuals possessing their own distinctive emergent properties (at one level of reality) constitute the building blocks out of which higher-level groups are formed, with those higher-level systems also possessing their own distinctive emergent properties. And, significantly, Hayek's account is one that implies that the properties of those higher-level entities include the capacity to react back on, and to shape, the attributes of those individuals who comprise the elements from which it is formed, casting further doubt on the claim that Hayek portrays people in atomistic terms. [emphasis mine].

This claim from Lewis lies at the heart of the tensions faced by Hayek within the context of early research in computer science.

While it seems that Hayek valued his contribution on its own terms - in a letter to John Nef he claimed that it was "the most important thing I have yet done" (Caldwell 1994, 239) - he would later emphasize his intention to apply the methodology that he used in The Sensory Order to his work on social theory.

In the decades following its publication, Hayek faced both tension and admiration for his work. I am not the first to recognize that The Sensory Order was recognized by none other than Frank Rosenblatt (1958) in his work, "The Perceptron: A Probabilistic Model for Information Storage and Organization in the Brain." In the history of machine learning, "The Perceptron" is a major turning point as it was a formal computational model of a neuron that integrated feedback that automatically improved the prediction of the neuron. (In a future post I will elaborate the Perceptron in detail.)

Hayek also attracted detractors. Phil Mirowski shows that Oskar Morgenstern regularly pilloried Hayek:

[12 January 1945]: Yesterday a discussion at the Weyl’s, with also Lutz, Neumanns. Theme: Hayek’s ‘Road to Serfdom’. It was just verbal and I found the book unbearable, blurred and full of contradictions. But it was very neat.

[19 April 1945] Hayek is here for ten days. I gave a party. He is clearly unchanged; It is a pity he is in the hands of [ublen Gesellschaft]. I heard him in a colloquium after a dinner. I can’t stand it any longer. Neither him, nor his opponents; that’s no science. I could predict all the answers.

[22 May 1946] Need to write another time more, especially about ‘Schmonzologie’ (Johnny [von Neumann] about Hayek’s presentation)...

[23 May 1946] in the evening Hayek delivered the Stafford Little Lecture about competition. Weak, literary, and scholastic. Although there were 200 people in the Frick Auditorium, it was no success ... H is enormously self-centered; what is done by others doesn’t interest him. He never started reading the ‘Games’, but he is ‘against it’. What he said about epistemological questions was very primitive.

[26 May 1946] Once, Utopia was an economic proposition. Nowadays these speculations are only heard in the natural sciences. What can be done there in 50 years, can only be dreamt of by economists. The prospect of wealth is very good, even within reach. It makes political problems the thorn, or the enslavement, as Hayek thinks of it, etc. (Mirowski 2007, 360-61)

Other researchers in cybernetics, including Morgenstern's coauthor, John von Neumann, specified their propositions using statements of symbolic logic. Hayek viewed the price system as a telecommunication system best described in terms of participation by human economic actors, not lightning fast calculators that were often presupposed by such systems. The methodology of complexity was, for Hayek, an approach that naturally recognized the need for liberty. Complexity without clearly stated symbolic logic was, at least for some, suspect.

Information theory as it developed from the 1940s onward, was concerned about discrete packets of information with clearly identifiable features. Early work by Claude Shannon (1948) sought to map the possible space transmission, identifying the capacity for information transmission measured in bits per second, the possible number and variety of characters encoded in a transmission, variation in transmission length for a character, use of Bayesian inference to predict a character, employment of redundancy to enable error correction, and so on. Clear framing and clever construction enabled error correction by units receiving transmissions were at the heart of the computing revolution.

Similarly, early researchers in computing described precisely the operations of a machine. Machines are conveniently deterministic so that if one knows the state of a machine at time $t$ and has knowledge of all relevant inputs, then the state of the machine at all periods $p > t$ can be computed. Sure, a program might run infinitely into the future, but the halting problem did not mean that many features of systems could not be described, often with great efficiency, using recursion.

While much is left to be said on the matter, here it will suffice to recognize that Hayek had outed himself in 1952 when he attempted to use Gödel's proof to infer the limits of self-understanding. He attempted to support his conclusion that we must be left to navigate a methodological dualism that could do no better than to link human consciousness to the mechanisms that enable learning at the level of the nervous system. The construction of models with hierarchies of abstraction was for Hayek, supported by iterations of the same mechanisms operating at the level of individual neurons and grouping of neurons, and larger scale grouping. But the exact chain could not be elaborated continually to the point of the emergence of consciousness. This imposition of a limit to understanding was a step too far for those who hoped to more clearly link consciousness to a clear chain of reasoning capable of surmounting this methodological dualism. One need not stake a claim with either camp to recognize the significance of this tension.