Michael Polanyi: Can the Mind Be Represented by a Machine?
By Paul Richard Blum
Full version now available as
Michael Polanyi: Can the Mind beRepresented by a Machine? Documents of the Discussion in 1949, in Polanyiana (Budapest) 19 (2010 [actually 2012]) 35-60.
On the 27th of October, 1949, the Department of Philosophy at the University of Manchester organized a symposium "Mind and Machine", as Michael Polanyi noted in his Personal Knowledge (1974, p. 261). This event is known, especially among scholars of Alan Turing, but it is scarcely documented. Wolfe Mays (2000) reported about the debate, which he personally had attended, and paraphrased a mimeographed document that is preserved at the Manchester University archive. He forwarded a copy to Andrew Hodges and B. Jack Copeland, who in then published it on their respective websites. The basis of this interpretation here is the copy preserved in the Regenstein Library of the University of Chicago, Special Collections, Polanyi Collection (abbreviated RPC, box 22, folder 19). The same collection holds the mimeographed statement that Polanyi prepared for this symposium: "Can the mind be represented by a machine?" This text has not been studied by Polanyi scholars.
First a summary of the debate as it appears from the minutes; an interpretation of Polanyi's statement will follow.
From the minutes of the discussion, it is obvious that the question concerning the human mind and computing machines was addressed -- and should be addressed -- from a variety of scientific angles. Dorothy Emmet, a scholar of Alfred North Whitehead, seems to have kept a low profile and yet she raised the typical philosophical question, whether consciousness is not what distinguishes human thought from machine operations. It is interesting that Alan Turing, in devising his later so called Turing Test, related this objection not to a philosopher but rather to the neurologist Jefferson (Turing 1950, p. 446). We have no evidence from the notes whether Jefferson had agreed with Emmet, but Turing's answer reveals that his approach to 'machine thought' focused on "sets of rules" ; he granted "conscious working" a status separate from routine operations that can be performed/emulated by a machine. Thus, looking for the philosopher's input first, we see immediately everyone concerned with the specter of materialism or of the "fourth continuity", i.e., the upcoming new shock (after Copernicus, Darwin, and Freud had evinced the continuity between humanity and the cosmos, animals, and mental illness) which consisted in creating a seamless transition between mind and machine (Mazlish 1967).
It seems Emmet was the only professional philosopher present, if we except Polanyi who in 1948 had moved from chemistry to "social studies". Mathematicians were present, namely Maxwell Herman Alexander "Max" Newman, Alan Turing, Maurice Bartlett, and Bernhard Neumann; furthermore the neurologists J. Z. Young and Jefferson, and then a person named Hewell (so far unidentified; note the question mark at the first appearance of his name) who also seems to be an expert in physiology of the brain.
Evidently, the discussion started with Newman responding to Polanyi's statement and his interpretation of Gödel's theorem. It appears that the mathematician conceded the assumption that the operations of machines "cannot do anything radically new" and that the assessment of the difference between mind and machine is a matter of experimental research, rather than a priori speculation. At this point, Turing intervenes by emphasizing that his idea of a "universal machine" entails some kind of self-referential operation, as one could translate the capability of "turning itself into any other machine". Consequently, the three questions summarized by Emmet after a break: machine-brain analogy, physiology, limitations of the machine, are all answered by the mathematician Turing with an enigmatic reference to "trial and error" in combination with memory. Turing seems to have no doubt that operating on 'past experience' and memory are not features that keep the machine and the brain apart. The next time a mathematician joins in, namely Newman, the problem of the design of the calculator comes to the forefront. He implicitly suggests using the design of a calculating machine, as the paradigm for investigating the mind. In doing so, he underlines the elementary problem raised by physiologist, Young, of internal versus external approach to operations. For "in the case of the mechanical brain we start with something which has been made by us", as Young said, so that the implied solution could be a methodical approach to what takes place when a calculating machine is being programmed. This would be a 'meta-programming' approach, in which the program and the design are not the same thing.
When the discussion circles around the problem of memory storage, Newman makes a further methodical suggestion that, broadly speaking, reflects the hypothetical-deductive method of science: "start like the atomists with a 'billiard ball' hypothesis”; that is to say, to hypothesize that the mind is a machine ("which is obviously wrong") in the hopes that experimental and theoretical research will falsify that, in due course. Not only the mathematician Neumann, with reference to consistency proofs, but also the neurologists agree. In spite of Polanyi's repeated objection that some achievements of the mind cannot be hypothesized with "crude models", the conversation goes on along the lines of models, analogies, and hypotheses and heads towards the notion of "incompatibles". If a system is observed from the outside, hypothetical paradigms help pointing out crucial changes that may or may not reveal alternatives or "choices" in the observed process. At this point, Turing stresses that "random operation can be made to become regular after a certain prevailing tendency has shown itself". The regularity is to be assumed first in the observation, that is, in the accommodation of the paradigm to the observed process, and then assumed to be indicative of the regularity of the process itself. So far it is a case of pure empirical research. Turing illustrates that with the operation in a machine, which on the input of incompatible data registers the contradiction and returns to the origin of the contradiction. Just recently, John von Neumann had suggested that the rounding-error of a calculator, which is caused by the limited number of digits, can be compensated by having three machines doing the same calculation and stopping in case of conflicting results. This is the moment when Turing explicitly sides with Polanyi in citing his insistence on the basic difference between "mechanically following rules" and consciously knowing rules. And yet there remained a dissent that needs to be clarified. Before that can be done, we need to take a look at the physiological/neurological view in the debate.
To summarize the mathematicians' contribution to the discussion: there is no doubt that machines are no minds, although Turing is seeking for the option to produce self referring machines, which opens the methodological discussion about design, method, and the meaning of rules.
Early in the debate Jefferson feels compelled to state that bodily functions may be interpreted in mechanical terms, "but not 'Mind'", to which Turing replies that even in engineering and operating a machine, there remains an element of playing and ignorance. If that was actually the response of Turing to Jefferson, as it appears in the minutes, then the mathematician is consciously rejecting a mechanistic approach to physiology. His later definition of purpose as "use of previous combinations plus trial and error" suggests, in this perspective, that biologists should not abolish hypothetical purposes in living organisms, but rather redefine purposefulness in terms of programming. In terms of programming, to "put a purpose into a machine" is not different from the operations of living organism. Bartlett seems to have no problem assuming that the brain can be functioning with statistical errors. So Young joins in: first, he interprets Turing's remarks as an invitation to neurologists to collaborate with mathematicians. Cybernetics would be the point in case. (Turing was involved in the birth of this new discipline: Wiener 1948/1961, p. 23.) Then Young mentions as a problem that the behavior of brain cells might be different from that of other cells in organisms. But more importantly, he points out that the "collaboration" could reveal a fundamental difference between physiology that investigates a self-sustaining object, whereas engineering a brain provides an object, the rules of which have been established and implanted by the researcher. He re-phrases the same difference by asking: "The physiologist can stimulate points and see what happens -- do the 'mechanicians' do the same?" The distinction between a mechanical and a physiological view on the mind could be that the brain as an organism is still good for surprises (it has to be studied empirically); whereas the mind as a machine would presuppose that its laws are known a priori (from its blueprint). Nothing unexpected should happen when poking a mind-machine. Histology and EEG are mentioned as recent techniques of physiological investigation and the specialists keep exchanging about its uncertainties until Young moves over to philosophical aspects.
In departing from empirical approaches, the physiologist seeks help from philosophy: first Young ponders the option that memory is not limited to a specific location in the brain, like memory cells, and then he endorses the term Gestalt of philosophical psychology and calls for logic to set up promising hypotheses. The promise lies in the pattern shift. Wholeness, paradigms, and logically supported hypotheses—that's what is needed in order to guide empirical research into the brain as the organ of thought. Polanyi very much appreciated Gestalt psychology giving it an anthropological and cognitive meaning: "Gestalt psychology has assumed that perception of a physiognomy takes place through the spontaneous equilibration of its particulars impressed on the retina or on the brain. However, I am looking at Gestalt, on the contrary, as the outcome of an active shaping of experience performed in the pursuit of knowledge." From this philosophical interest he can voice doubt whether "seeing stereoscopically" can play a role in the envisioned research program on the mind and, again, how would is possible to "derive from the model the conception of 'seeing in depth'"? The philosopher seems to be capitalizing on the inherent methodical limitations of mechanical physiology, which are both expressed in Jefferson's naïve dualism and in the antinomies of organisms that, as empirical objects, elude a singular method.
Eventually the physiologists joined the mathematicians in departing from the narrow confines of their disciplines. Cybernetics, in this perspective, opens an understanding of machine operation that matches the anatomical findings and lack thereof. Jefferson (1960, p. 43) lamented later that feelings miss "special abodes in the brain" and have been reduced to "fictitious … entities". Since he was quite informed about the history of sciences, and in particular the debate about the location of the soul (Jefferson 1949, cf. 1960, pp. 94-209), it is also legitimate to mention that the study of the physiology of the brain, in connection with attempts at anatomy and localizing the human mind and its functions, had had a steady career in 18th through 19th century medicine and philosophy (Hagner 2008). Mechanical optimism battled with philosophical skepticism. There was always at stake the question: is there any relation between functions of the human mind (sense perception, memory, morality, etc.) and the empirical data? Young's probing octopus brains and Jefferson's neurosurgery were just continuations of that century old debate. The aim remained the same, finding the interface between psychic and physical states. The news at this discussion was that decentralized memory storage and generally non-localized and non-mechanical forms of operation became thinkable. In that sense it marked the threshold to the computer age and cyber world.
As soon as unpredictability, randomness (Turing), non-quantifiability, and holistic approaches are admitted, Turing is able to redirect the perspective away from mind-as-a-machine towards the functioning of machines. Therefore, when he answers Jefferson's objection who doubted that human beings would be able to be perturbed by conflicting results of a thought process, Turing quips -- to the amusement of the people present -- that this is what mathematicians do. So we have the paradoxical situation that the physiologist gives less credit to the 'intelligence' of the mind than Turing gives to a well working machine. When someone in the audience asks: "are mathematicians human beings?" it becomes obvious that the Turing project is to analyze human thought by way of programming a computer. Mathematics is what humans do and what machines can do.
Therefore it now also becomes clear what is at stake between Turing and Polanyi. They agree that a mathematical interpretation of thought is not all there is. Yet, Turing tries to find in thinking as much mathematical procedure as possible, while Polanyi aims at capturing with philosophical precision that what remains. Jefferson's foggy "not the 'Mind'" and Young's apparently crestfallen swerving into philosophical methodology are to be remedied by Polanyi's insistence on formalization and specification. This is what he had to tell his colleagues in his own statement.
In his statement on the question whether the mind can be represented by a machine Polanyi pronounced five theses. First he interpreted the development of mathematics from Hilbert to Gödel as establishing a realm that cannot be formalized and hence is prior to computation that a machine can do. Second, he identifies this non-computing operation with reflection as the specific power of the mind. Third, the outcome of Gödel's and Tarski's discoveries do not disturb the understanding of human mind, they rather afford a philosophical tool to distinguish the primordial capability of reflection on rules which itself is not bound to those rules. His fourth point is the capability of belief that precedes empirical knowledge and is its foundation. Lastly Polanyi reasserts the denial of mechanical determinism. In Roger Penrose's classification (1994, pp. 12-16) Polanyi would probably be a "C- believer", according to whom "the problem of conscious awareness is indeed a scientific one, even if the appropriate science may not yet be at hand" (p. 16), that is, Polanyi seems to believe that some mental activities can be emulated by computers, but not all of them.
As is well known, Gödel had claimed that the system S contains propositions that cannot be proven and it contains undecidable problems (paraphrased from Gödel 1930, pp. 141-143). He had also added that his theorems "can be extended also to other formal systems" (p. 143); indeed, "[a]ny epistemological antinomy could be used for a similar proof of the existence of undecidable propositions" (Gödel 1931, p. 149 n 14). Now it is revealing how Polanyi phrased Gödel's "discoveries": To him this was originally about "arithmetic and advanced geometry". An unknown hand corrected his wording by saying that Gödel dealt with "number theory". This is factually correct, but it shows Polanyi's drive to extend the meaning of the theorem beyond number theory. A few lines down, when Polanyi concluded that there must be a "procedure for the discovery … which, by its very nature, is incapable of formalization", the same hand interjected that formalization is possible in meta-language. For Gödel 'formalization' was a term of art within mathematics, that is, to be "reduced to a few axioms and rules of inference", and his aim of 1930/1931 had been to show that it is not the case "that these axioms and rules of inference are sufficient to decide any mathematical question that can at all be formally expressed in these systems" (Gödel 1931, p. 145). This exchange makes it clear that Polanyi saw in Gödel's discoveries a point of departure from the need of formalization. As a scientist he was, of course, well acquainted with axiomatic systems. But as a philosopher, he was intrigued by the option of an infinite regress in formalization, a regress that is spurred by reflection, as he says, to the effect that reflection must necessarily stand outside of the mathematical/scientific procedure. When his reader appealed to meta-language that could reenter the process of formalization, i.e. axiomatization, he was kicking at an open door, for Polanyi had already integrated meta-language on his escape route out of the word of formalization. Tarski and Gödel are both witnesses to an "indefinitely extending programme of innovation, which can be achieved only by informal methods and not by a machine" (end of section 2).
When Polanyi returns to the subject in his Personal Knowledge he states (1974, p. 258) that "a formal system of symbols and operations can be said to function as a deductive system only by virtue of unformalized supplements, to which the operator of the system accedes: symbols must be identifiable and their meaning known, axioms must be understood to assert something, proofs must be acknowledged to demonstrate something, and this identifying, knowing, understanding, acknowledging, are unformalized operations on which the working of the formal system depends." Thus he groups a wide range of mental acts together as those that precede formalization (symbols, axioms, rules), and this is what in the 1949 debate he called "semantic function". "We call them the semantic functions of the formal system. These are performed by a person with the aid of the format system, when the person relies on its use." (Ibid.) He immediately adds that there is a "legitimate purpose of formalization", namely, an increasing reduction of "informal operations; but it is nonsensical to aim at the total elimination of our personal participation" (p. 259). So he remains faithful to his understanding that mechanization, and specifically a mathematical interpretation of thought, is appropriate as long as one acknowledges the existence of what he later would call the personal coefficient. Surprisingly he dismisses the Turing Test (Turing 1950) alleging that Turing had turned the question about thinking machines into "the experimental question, whether a computing machine could be constructed to deceive us to its own nature as successfully as a human being could deceive us in the same respect" (p. 263 n 1). If Polanyi did not reject Turing's project on the whole -- and he didn't -- he must have been alarmed by the playful implications of the mathematician's tongue-in-cheek approach to cognition. It is obvious that Turing's mental experiment successfully fooled a large audience into believing that thinking was a trickster game.
Much of Polanyi's energy was invested in unmasking imposters and simplifiers. In the opening chapter of Personal Knowledge, he takes to task the myth of objectivity that makes believe there were no personal investment in discovering objective facts of nature (cf. Blum 2010b). In a paper of 1950 on "Scientific Beliefs" (that will become part of Personal Knowledge¸ chapter 12), he attacked standard positivism: "A genuine scientific theory must operate like a calculating machine, which, once the keys representing the dividend and the divisor have been depressed, determines the result automatically" (p. 27). It is this broader cultural context that interested Polanyi when he joined the debate about mind and machine. The misunderstanding of the working of a machine is an expression of the mistaken anthropology. Positivists and mechanicists believe to "construct a machine which will produce universally valid results. But universal validity is a conception which does not apply outside the commitment situation." (p. 35). Polanyi dedicated an offprint of that paper "to A. M. Turing with best thanks". In a later class on "Unspecifiable Elements of Knowledge" (his famous book was out since 1958) he boldly uses machines as a paradigm. When the design of a machine had been invoked to solve the problem of the mind-machine-riddle, then it was even more fitting to choose as the "leading example a class of comprehensive entities of which we can specify both the particulars and their coherence." The surprising result is that the philosophy of machines is not much developed, and therefore it has been overlooked that machines "embody rules that are not laws of nature"; even more, the failures of machines ("bursting of boilers") are part of their essence, namely, as "imperfect embodiment of its ideal". In this lecture Polanyi fought submitting to the machine as the ideal or paradigm and recuperated it for the anthropological inquiry of knowledge.
Therefore, it is important to notice Polanyi informing his readership of Personal Knowledge that Turing's contribution to the Symposium "Mind and Machine" was "foreshadowed" by his paper on "Systems of Logic Based on Ordinals". That paper "deserves to be read and understood far more than it has been." (Turing 1939, p. 71, introduction.) It addressed Gödel's incompleteness theorems, but towards the end, the author leaves technical mathematical language behind and reflects upon mathematical reasoning; and it was most likely this § 11 that caught Polanyi's attention. "Mathematical reasoning may be regarded rather schematically as the exercise of a combination of tool faculties, which we may call intuition and ingenuity." (Turing 1939, p. 214) In a footnote he clarifies that he is "leaving out of account that most important faculty which distinguishes topics of interest from others; in fact, we are regarding the function of the mathematician as simply to determine the truth or falsity of propositions." Again an example of Turing's sense of irony: the most important thing is left out. But that leaves us with understanding that the reach of his number theory goes exactly as far as truth and falsity of propositions go. This is no reductionism. In describing the function of intuition and ingenuity, he emphasizes the role of intuitive judgment and the need for "suitable arrangements of propositions" and takes it for granted that "these two faculties differ of course from occasion to occasion, and from mathematician to mathematician." Again, assuming that Turing is leaving technical mathematical language behind, his description can only be understood as the establishment of the competence that Polanyi would have understood to be 'personal'. It seems this is the passage Polanyi had in mind when he invoked Gödel for having proven that there is a non-formalized capability of the human mind. "In pre-Gödel times it was thought by some that it would probably be possible to carry this program to such a point that all the intuitive judgments of mathematics could be replaced by a finite number of these rules. The necessity for intuition would then be entirely eliminated." (Turing 1939, p. 215) Reductionism in the sense of eliminating the personal component has been overcome by Gödel and, consequently, Turing. To eliminate the personal component is a methodical aim for the sake of mathematical theory. Therefore during states: "We are always able to obtain from the rules of a formal logic a method of enumerating the propositions proved by its means. We then imagine that all proofs take the form of a search through with this enumeration for the theorem for which a proof is desired. In this way ingenuity is replaced by patience." He calls that "heuristic". (Ibid.)
Turing's paper on "Systems of logic based on ordinals" may have been extraordinary within his own production, but Polanyi found it more important than his famous 1950 paper. It seems the mathematician was "steaming straight ahead with the analysis of the mind, by studying a question complementary to 'On Computable Numbers'," as Andrew Hodges put it. "The Turing machine construction had showed how to make all formal proofs 'mechanical'; and in the present paper such mechanical operations were to be taken as trivial, instead putting under the microscope the non-mechanical steps which remained." (Hodges 1999, pp. 19-21) Therefore if the analysis of human thought is at the focus of attention, the distinction upon which Polanyi and Turing agreed, namely, that between rules and knowing rules turns out to be constitutive for any theory of thinking and computing.
As a young man, Turing had mused about the "Nature of Spirit" and described the same relationship as follows: "As regards the question of why we have bodies at all; why we do not or cannot live free as spirits and communicate as such, we probably could do so but there would be nothing whatever to do. The body provides a something for the spirit to look after and use." (Hodges 1983, p. 64) This is patently the traditional language of body-soul-dualism, and it will take some education to translate that into problems of logic and mathematics. But looking back the structural identity is clear. There is a relationship of independence and manifestation that cannot be 'reduced' or 'eliminated'. Obviously pure non-formalized thought would be as 'boring' as an absolutely free spirit. On the other hand, science is after the laws of matter. Polanyi expressed that in the context of his recapitulation of the Manchester debate by assuming that a mechanical approach implies to determine that a particular object is seen to be a machine, a perspective that in and of itself leaves already simple mechanistic views behind. "A machine is an interpretation of an observed mind ... and not of an observing mind ..." (Polanyi 1952, p. 315) "For a machine is a machine only for someone who relies on it … for some purpose, that he believes to be attainable by what he considers to be the proper functioning of the machine: it is an instrument of a person who relies on it." (Polanyi 1974, p. 262)