The Daoist Image of Control (I)

From Popper's three worlds to language as a control technology.

Though Karl Popper is better known for his hypothetico-deductive view of science and for his defense of open society, he also put forward a curious philosophy of mind known as interactionism. In the book The Self and Its Brain, which he coauthored with the Australian neurophysiologist John Eccles, Popper first proposes the partition of reality into three worlds (World 1 of physical entities, World 2 of mental states, and World 3 of human cultural artifacts) and then argues that World 2 is necessary as the bridge for World 3 entities to have causal efficacy in World 1. The examples he gives are works of art (when one artist’s perception and appreciation of the work by another artist provides inspiration for further works) and scientific theories (when a scientist notices an anomaly in some scientific theory, formulates a problem, proposes a solution, and communicates it to other scientists via publications and talks). According to Popper, mental states of the artist and of the scientist play important roles in sparking creativity, inspiration, intuition, insight, etc. Moreover, while World 3 entities can also be World 1 entities (sculptures, paintings, books, scientific articles, music scores, film screenplays as physical objects), they possess a certain degree of what he calls partial autonomy. The example of the latter is arithmetic and number theory: While number systems, as a World 3 artifact, are a human invention, the notions such as composite and prime numbers, odd and even numbers, etc., even though they are in some sense already encoded in the appropriate formal system, were waiting to be discovered. This is also, according to Popper, the source of problems and hypotheses, such as the (im)possibility of squaring the circle, the irrationality and transcendence of certain numbers, the Goldbach conjecture, the Riemann hypothesis.

Now, language is a World 3 entity par excellence. The Popperian partial autonomy of language lies in its generativity in the Chomskyan sense.¹ Language, as a cultural technology residing in World 3, is a means of doing things with words in World 1, as per J.L. Austin. By learning to acquire and use language, human minds have first-hand experience of grasping these World 3 entities via mental correlates of words, sentences, and more elaborate linguistic constructions. Language is descriptive and argumentative, both of these qualities find their reflection and representation in World 2 of mental entities, and only via the latter can language have World 1 causal efficacy. From this, it is only a short step to mentalese, Jerry Fodor’s language of thought. For Popper, though, this is a way to reject physicalism by arguing that, without the mental conduit of World 2, no meaningful interaction between World 1 and World 3 is possible.

The current debate about the internal workings of Large Language Models is, I think, best viewed through this Popperian lens. We mine World 3 for training data to feed into these massive systems and expect them to interact with, and causally affect, World 1. The Popperian interactionist stance would be that the accumulating empirical successes of LLMs provide evidence for the reality of their mental states, their ability to represent their world and to reason about it. However, one could also argue that the successes of LLMs are actually evidence against the universality of interactionism—it may be possible to bypass World 2 of mental states and still use World 3 objects to control World 1. To put it differently, this debate really comes down to the question of the role of language—is it a means for representing and describing reality or is it a tool for coordinating and regulating action?

Popper seems to come down on the former side. He takes it for granted that language is descriptive and argumentative, so the World 2 links between Worlds 1 and World 3 are propositional, deductive, Aristotelian. He shares this view with the logical positivists of the Vienna circle. On the other hand, someone like Wittgenstein² or Heidegger would come down on the latter side. Most of our behavior and interaction with World 1, being-in-the-world, is not propositional, it is embodied in routines, to use Philip Agre’s terminology. Only when something interrupts the flow and demands our attention and reflection, do we adopt what Heidegger called the theoretical attitude. Sunny Auyang describes it in her book Mind in Everyday Life and Cognitive Science:

The primary meaning of our being, [Heidegger] argued, resides not in theoretical and scientific thinking but in our everyday handing of things and being with other people. Imagine yourself driving a car, negotiating the sharp turns, enjoying the scenery, listening to music on the radio, musing about the friend you are going to visit. The car handles so well you are almost oblivious of it; it is so integrated into your activity it has become a part of you, so to speak. All things—your friend, the car, the music, the road, the scenery—are interrelated and constitute your world with its primary meaning, a living meaning. Everything is significant and contributes to your purposive activity, but the significance is tacit and not explicitly articulated. The car is something you maneuver to keep on the road that leads to your friend. The car, the road, and your friend are handy. They relate to each other, and together they constitute the world of primary significance.

If the car breaks down, however, it ceases to be handy equipment and becomes a mere thing present to you. Then you switch to the theoretical attitude and try to figure out what is wrong with the thing and what to do with it. In doing so, your world acquires a secondary significance. According to Heidegger, the theoretical attitude is a common mode of being human, albeit a secondary and derived mode. When you switch to the theoretical attitude, you abstract from many cares and concerns that you are normally involved in, focus your attention on a few things, and make them stand out by themselves as things in space and time. You trade the richness of wider experience for the clear and refined vision of a few objects.

Now, this being-in-the-world described by Heidegger is still bound to language, though the bonds are not those of propositional notion of truth. While reasoning is obviously involved, it is not a syllogism-on-paper type reasoning that Popper seems to have in mind as the prime example of World 2 as interface between Worlds 1 and 3. It is a pragmatic reason, focused on empirical success. With this, comes a different view of the role of language—as a control technology, rather than a means of filing and representing facts. This, according to Chad Hansen’s fascinating book A Daoist Theory of Chinese Thought,³ is how classical Chinese thinkers conceptualized language:

All the ancient [Chinese] thinkers viewed languages as a way to coordinate and regulate behavior. No one in this tradition developed a theory that the central function of language was representing or picturing facts or reality.

According to Hansen, a good way to grasp the difference between the Indo-European view of language, centered on propositional truth, and the Chinese daoist view of language, centered on effective and reliable behavior in the world, is through what he calls, interestingly, the computer analogy:⁴

Substitute the notion of a computational program for the familiar picture of the mental-arguments concept of reasoning. Use the computer model to explain how a physical being can process language and how language guides action in a real world context. We need not explain a computer's operating by attributing to it mental or semantic content, inner consciousness, or experience.

A computer operates with a program. We input the program—load it into the computer. That process changes the computer's dispositions in complex ways. It will now behave differently to different inputs. We may use the language of stimulus-response or even of intuition, but the relation of input and output need not be simple minded. It can come from a very complicated program. The computer's dispositional state has a physical realization (an electronic functional state) that may produce very subtle behavioral responses after a complicated calculation.

The explanation of the computer's behavior does not require our habitual contrast of belief and desire. The program itself causes the behavior (of course, using energy supplied as electricity and oil, requiring a dust-free environment, and resting—cooling off). …

If we accept that the Chinese philosophy of mind did not rest on a mind-body dichotomy, then we surely could use something like the computer analogy to explain intelligent human behavior. The computer analogy illustrates how a physical thing could reason, act, be moral, and function in the world as humans do. A computer can print out the result of a complicated calculation when we input the data. It does a calculation though it does not have beliefs. We do not suppose that the computer reflects on the meaning of the premises and sees the conclusion of its proof in a kind of rational insight. Informally, it helps us see why we do not need to assume that Chinese philosophers took the traditional Western theory of the relation of mind, language, and the world for granted.

He further writes:

We get a model that explains the central position of dao (way) in Chinese philosophy. A dao is analogous to a program. Confucius viewed education as inputting the inherited dao (guiding discourse) of the sages. We study and practice a dao. We learn to speak and act properly by studying the Book of Poetry or the Book of Rites. This view accounts for Confucius' distinctively non-Western attitude that instilling tradition is a realization and fulfillment of human nature instead of a limiting constraint. The de (virtuosity), as the traditional formula had it, is the dao (way) within a person. It is the physical realization of the program that generates the behaviors. When we have good de (virtuosity) our behavior will follow the dao (way). The program runs as intended in us. Good de (virtuosity) therefore, is like a combination of virtue (when compiling a moral dao (way)) and like power (because executing instructional programs enables us to do things). Our virtuosity is the translation of an instruction set into a physical, dispositional potential.

Daoist thinkers make this view of things especially clear. The Laozi introduces the idea that we create desires by learning guiding discourse—gaining knowledge of what to do (know-to). The programming model explains many issues of classical Chinese thought. Notice first that we are, in a sense, programming each other. Our outputs include language that is input to others. The importance of maintaining cultural traditions, the family, the father model of the ruler and the educational role of political society all have clear motivations once we adopt this model. …

Instantly this model gives us a new conception of the roles of language and mind. Pragmatic (action centered) rather than semantic analyses now make more sense. Language guides and controls behavior. It does this by restructuring our behavior guiding mechanism, the xin (heart-mind) The common translation of xin as heart- mind reflects the blending of belief and desire (thought and feeling, ideas and emotions) into a single complex dispositional potential. We need not attribute separate structures of reasoning and feeling to computers to explain their behavior.

Hansen’s use of “program,” and his overall take on language, is very similar to James Beniger’s ideas. However, his notion of computation (and of program) seems to be a lot more liberal than Beniger because he explicitly rejects the identification of programs with proofs and instead goes for something closer to Marvin Minsky’s society of mind framework. In the next post, I will discuss the relevance of this way of thinking to LLMs (in particular, the emphasis on rectification of names, on making of actionable and projectible distinctions, and on using language as a guide to reliable behavior in the world).

(to be continued)

1. "Hold the newsreader's nose squarely, waiter, or friendly milk will countermand my trousers." (Fry and Laurie).

2. In his Philosophical Investigations era.

3. Hat tip to Amod Sandhya Lele, from whose older post I found out about Hansen’s book.

4. Hansen’s list of acknowledgments is curious and informative: apart from Wittgenstein, Quine, and Sellars, he also thanks Stephen Stitch, David Lewis, Daniel Dennett, Richard Montague, Michael Sandel, Saul Kripke, Richard Rorty, Hilary Putnam, John Rawls, Richard Grandy, Tyler Burge, Thomas Nagel, and Derek Parfit.