How do structure and meaning relate, and how do we explain the existence of such structures? By story or by reason?

Welcome back to Part III of our series on the philosophical foundations of AI. (In case you missed them: here are Part I and Part II)

This post explores questions of logic and structure (and their relation to meaning) and examines how myth and reason were separated by Plato and Aristotle (man as rational animal). From this time on, Western thinking assumed the purposes of poetry (i.e., the humanities) and science (i.e., natural philosophy) were orthogonal to one another. One deals in fiction, the other, truth. This distinction, maintained until today, led some twentieth century commentators to complain of a growing gap between the Two Cultures, both of whom were largely ignorant of key ideas in the other culture. I would argue that now, more than ever, the future of AI requires people well-versed in both areas of human knowledge.

To understand the split, it’s helpful to go back to the mid-1600s, when René Descartes used the power of reason to advance the methods of science. Typically the “rationalist” approach of Descartes is contrasted with “empiricist” approaches, which value the power of our senses to obtain knowledge.

We’ll see how this philosophical debate is echoed in current machine learning where prominent AI researchers, such as Gary Marcus and Yoshua Bengio, argue about the relative importance of learning from data vs. imposing structure (causal or otherwise) in advancing the field. I claim that Kant’s ideas could be useful here in reaching a compromise.

Finally, we close with a critique of the atomistic view of the subject advanced by Descartes. This critique will eventually lead to alternative paradigms in cognitive science/AI, such as embodied cognition, which borrow and extend many insights from European philosophers such as Martin Heidegger and Maurice Merleau-Ponty.

Language, Logic, Pattern, and Structure

Even before the time of Plato, the ancient Greek philosopher Parmenides claimed it was logos, (here translated as reason) that provided the ultimate principle behind reality. What participates in logos is, by its nature, intelligible. It is no accident the Greek word Kosmos, which we use now to denote the universe, was originally used to describe things in their proper arrangement.

We find a similar idea in Chinese philosophy in the 道教 (Daoism), and in particular, the book known as the 道德經 (Dao De Jing). Daoism sees the universe as a grand, connected whole. As you may know, 道 means something like “path” or “way,” and, most interesting for us, the phrase 有道理 means roughly “makes sense” or “logical,” in a non-technical sense. In fact, the character 理 is used to describe the patterns and textures found in jade and wood. So Western culture is not alone in associating logic with reason, reality, and structure. Let’s explore these connections in more detail.

Reading and Understanding

What does not make sense, we cannot understand. It is unintelligible. Note that this very word — unintelligible — derives from from the Latin leger, meaning to read (for comparison, we have theSpanish leer andFrench lire). In English we say something that we cannot read is illegible. We take this to mean that it contains no patterns (理) from which we can derive meaning. In other words, there is no structure by means of which we can locate part within whole. To not make sense of some thing means we are confronted with some part or piece that cannot be connected to a whole from which the piece derives its identity and therefore meaning. This is the idea behind Structuralism that Derrida and others would reject.

Here’s an example. If I look at an unfamiliar script I cannot understand (say, Arabic), the written marks may indeed point to something, but they must either point to something which I cannot identify, or perhaps not point at all.Practically speaking,there is no difference for meas a person ignorant of Arabic; they might as well be random splotches of ink dried on a page.

Try as I might, I cannot orient this pointing from part to whole.It is only through knowledge of Arabic that I can get access to the “correct” relation of part to whole. The reader of Arabic sees the marks and knows what they refer to. We might use the analogy of one big lookup table inside our skulls. Maybe “knowing” Arabic consists in building up, piece by piece, a giant look up table linking patterns of signs with objects and sounds. (Sidenote: in the 1980s, Doug Lenat’s Cyc project tried to do “ontological engineering” by amassing a giant collection of millions of “facts about the world” in the hope that once some critical mass of information was collected, a machine might be able think like a human. Needless to say, enthusiasm for the the project eventually died out. Oh yeah, and neural networks got popular.)

If you’re not familiar with the Chinese characters written earlier, I imagine you experienced this problem first-hand just a few moments ago. We will return to the relation between meaning and structure when we examine Wittgenstein’s account of meaning. If you’ve heard of John Searle’s Chinese Room thought experiment, you might also be starting to see the connection with symbolic AI here. Be patient, we’ll get there eventually…

Western culture is perhaps unique, though, in its singular focus on reason as means to grasp reality or logos. Other Asian religious traditions, such as Buddhism, deliberately use the body to train the mind in the practice of mindfulness. Yoga also uses the body to help train the mind. At this point, I should also mention that a key principle in Daoism is called 無為, roughly translated as “inaction” or “non-interference.” This will contrast with Western notions of activity of the mind as the means through which truth is known. We will see that a major task of later Enlightenment thinkers was to find universal procedures that would generate air-tight conclusions in any and all situations.

Mythos vs. Logos

Let’s return to Greek thought to further explore the Western split between reason and story, mind and body. The English word poetry stems from the Greek poesis, meaning to create. Thus poets, at least in Plato’s time, told stories about the creation of the world, the gods, and the origin of humanity. Poetry and reality were intertwined. Different poets told different stories about creation: you simply believed whichever story moved you most. We might view stories as axioms: statements whose truth is assumed in order to make sense of other observations.

But Plato (as the mouthpiece of Socrates) and his student Aristotle would later sever whatever link had existed between logos and mythos, reason and story, philosophy and poetry. Indeed, natural philosophy would be the precursor to Western Science (Latin for knowledge) until at least the late 19th century. For Plato (and Socrates) Philosophy was thought to be the method through which a special, indubitable kind of knowledge might be obtained. Philosophy dealt in truth, while poetry dealt in fiction. Philosophy, taken literally, means love of (philo) wisdom (sophia). A philosopher, then, is someone who loves knowledge and uses reason to obtain it.

It would turn out, however, that the Socratic method of dialogue created a kind of gap between the myths of the poets and the wisdom of the philosophers. Later philosophers such as Jürgen Habermas and Chaïm Perelman would try to rescue the foundations for knowledge by working within this gap. Perhaps reasonableness, and not pure reason is the answer. Perfection can sometimes be the enemy of the good enough. (Sidenote: Alan Turing will make a similar point in foreshadowing the move from the logical foundations of computing, based on deductive logic, to machine learning. The secret, Turing realized, was that by accepting some possibility for error, we would gain much in flexibility.)

In any case, Plato claimed it was the Forms which were really real. Plato held that what we experienced was but a mere shadow to the reality of these forms (the Ideas). The artistic objects created by the poets were mere copies of copies. That is, they were copies of shadows of the Forms. We cannot encounter a perfect circle in experience, but we recognize the geometric object known as a circle because it “participates” in the form of the circle (Idea of the Circle).

Truth is what we get when we use reason to ascend to the realm of forms. In this realm live the objects of pure mathematics, devoid of any semantic content. We reach truth when we stop attending to mere appearances and instead exercise the faculty of reason to contemplate the forms themselves. Plato famously imagined his Republic devoid of dramatic poets. The philosopher kings could not be bothered by those only interested in stimulating our lower desires for humor and drama.

Descartes’ Method and Legacy of Dualism

Now that we’ve briefly toured the more ancient foundations of Western thought as exemplified by Plato, let’s now turn to more modern notions. This is the point at which mind and body become indelibly separated in Western academic thought. (random sidenote: the influence of Greek/Roman culture on European learning is still immense 2000+ years later. For example, in Germany students hoping to attend university attend what’s called Gymnasium and often study Latin, the original “language of science”).

Descartes was obviously a genius. I mean, he succeeded in connecting geometry and algebra in a way that revolutionized mathematics. But I nevertheless have the feeling that in his wake, academic knowledge accumulates not by growing the stores of knowledge, but by reducing, so to speak, the space of truth.

After Descartes, Western science becomes obsessed with method. Science becomes what we get by applying the method, thus resulting in a paradoxically dogmatic adherence to methods. Tunnel blindness would be an apt descriptor for the later Western obsession with “universal” methods for generating error-free knowledge.

In fact, Descartes’ 1628 book Rules for the Direction of the Mind essentially teaches algorithmic thinking by analysis. Analysis literally means to break into pieces. In effect, Descartes uses the computer science strategy of divide and conquer and recursion to tackle any problem and break it into smaller ones until eventually a simplest “base case” is reached that is solvable. But in doing so, Descartes mistakenly divided mind from body, leaving science to deal with the “easy” problem describing a world devoid of consciousness. But I digress.

Here’s an example taken from his 1637 autobiography Discourse on the Method of Rightly Conducting One’s Reason and of Seeking Truth in the Sciences. Notice a couple things: the priority of intellectual understanding over the senses, the process of breaking ideas into atomic pieces, and the method of enumeration, which we called “definition by extension” in Part I of this series.

The first was never to accept anything as true if I did not have evident knowledge of its truth: that is, carefully to avoid precipitate conclusions and preconceptions, and to include nothing more in my judgments than what presented itself to my mind so clearly and so distinctly that I had no occasion to doubt it.

The second, to divide each of the difficulties I examined into as many parts as possible and as may be required in order to resolve them better.

The third, to direct my thoughts in an orderly manner, by beginning with the simplest and most easily known objects in order to ascend little by little, step by step, to knowledge of the most complex, and by supposing some order even among objects that have no natural order of precedence.

And the last, throughout to make enumerations so complete, and reviews so comprehensive, that I could be sure of leaving nothing out.

With this little taste of Descartes’ approach to philosophical inquiry, our goal is to build up a feel for the Cartesian view of the world that leads us to the important foundational AI concepts of knowledge and representation. When we finally reach various post-positivist/postmodernist critiques of science, we will then have some familiarity with the issues.

Meditating alone by candlelight, Descartes famously concluded that his act of thinking presupposed the existence of a thinking substance, a self. Cogito ergo sum. The certainty with which he came to this conclusion further led him to distinguish — much to the chagrin of future scientists and philosophers — the two fundamental substances of mind and body.

During these meditations, Descartes concluded we could know things through the mind differently than things materially extended in time and space. In time, Newton’s physics would grow out of Descartes’ notion of res extensa (“extended things”) or “bodies” extended in space with properties of length, breadth, and depth, in contrast to res cogitans, or (“thinking things”) without body. Material objects may or may not exist, and we may be mistaken as to their physical properties, but Descartes claims that his existence as a thinking thing is certain beyond any doubt.

Rationalists vs. Empiricists

Descartes’ epistemological distinction further led to centuries of “rationalist” versus “empiricist” debates about the proper foundations and justifications of scientific knowledge. Empiricists, such as Bacon and Locke, wanted to build the foundations of knowledge upon what could be known through the senses and scientific instruments of observation. Induction was the paradigmatic intellectual mechanism for generating scientific knowledge. Only later did Hume point out the circularity of relying on induction to justify induction as a method of producing secure knowledge (see my previous post here for an introduction). Indeed, for Hume, all knowledge derived from either matters of fact intuited by the senses, or relations of ideas existing in the mind.

On the other hand, rationalists, such as Leibniz and Kant (to some extent), argued that the mind is what provided for the unity and coherence of experience and was thus ontologically primary to sense experience. What we know about the world is fundamentally constrained and filtered by the properties of mind. For idealists such as Hegel, knowledge of our minds was, in essence, knowledge of the world.

For rationalists, deduction was key to generating scientific and mathematical knowledge. The empiricists’ assumption that there could be a “given,” a kind of neutral, observational sense-data unfiltered and untainted by the workings of the mind was mistaken. The belief in the “myth of the given” managed to persist until the project of Logical Empiricism finally imploded in the late 20th century. Thomas Kuhn’s book Structures of Scientific Revolutions may been the death blow.

But since Descartes’ split between mind and body — also glossed as one between solipsistic subject (mind) and object (body) — Western philosophers and scientists since him have never quite managed to figure out how the principles by which mind and matter interact.

Rather than accept defeat, modern Western science accepted one horn of this dilemma by doubling down on the belief that mind might be “explained” solely by reference to body. Thus the reductive materialist believes we can “naturalize” the mind by reducing it to various physical and material phenomena, such as the firing patterns of neurons. Ultimately, the materialist does science with the implicit belief that one day our knowledge of physics will be able to (at least in principle) explain the macro-properties and behavior of objects like persons and economies.

Where Descartes Went Wrong: Self and Other

Let’s try to link this back up to our discussion of names and essences earlier in Part I. In my view, the fatal flaw of Descartes’ cogito is that he never stopped to think how it was that he was thinking. What were the preconditions for the possibility of this thought? What was the “substrate” of his thoughts that led him to conclude that mind and body were two separate kinds of things? His thoughts of course were realized in some language. A human language.

But how did he learn such a language for the expression of his thoughts? Was he born speaking a language? Clearly not. Another human must have taught him! Descartes cogito thus permits him to conclude that not only must he exist as a self, but a they or other must also exist such that the self can learn what a self is. This is a view similar to one held by the German Idealist philosopher J.G. Fichte whose writings on self-consciousness have influenced much thinking in Continental philosophy.

The point is this: the other is a precondition for the possibility of a self. Identity can only be understood by contrast with difference. Thesis and antithesis are resolved via synthesis, i.e., the dialectic (see my post here to learn more). What was initially understood as different was found to be the same when viewed with an expanded faculty of reason.

Self/other, identity/difference, thesis/antithesis are ultimate distinctions that drive the metaphysical process behind Being and becoming.

This insight serves as bedrock for much of continental philosophy (deriving ultimately from Hegel and other German philosophers) and in social and cultural psychology. For instance, George Herbert Mead’s notion of an “I” and a “Me” is based on Hegel’s notion of self and other. It is no accident dialogue (the method of Socrates)and dialectic share a similar linguistic form. As just a couple more recent examples, psychologist Michael Tomasello has studied how human identity forms reflexively through social interactions with others in social groups. And social and cognitive scientist Dan Sperber has claimed it was our need to create reasons and justifications for our actions in social settings that led to evolutionary pressures on our ability to reason.

The basic idea is that our identities as persons are both cause and effect of the identities ascribed to us by others. Against Descartes, the atomistic individual person does not and could not exist. We are not particles. We are social, self-interpreting animals. To speak a language is to participate in a shared form of social life, according to certain behavioral conventions that serve various cultural purposes. Put differently, to be a person is to be a certain kind of social being who is enculturated in a certain way and whose self-identity itself is itself the reflexive interplay between self and other, subject and object. We cannot define either of these terms independently of its counterparts.

We covered a lot, so let’s review the main points. Early on in Western thought, logic and the use of reason were separated from myth and poetry, echoing a distinction between reality and appearance. Yet, positing certain creation myths for the world seems analogous to positing certain axioms in mathematics.

Later in the 17th century, Descartes introduced an obsession with an analytical, atomistic (“divide & conquer”) method aimed at generating an absolute Archimedean foundation for knowledge. This resulted in cleaving off of mind from body, leaving scientific descriptions of the world without an explanation of subjective consciousness. Descartes also failed to realize how his thought must be couched in language, and so his cogito ergo sum should have led him to conclude the existence, not only of a self, but of an other as well.

About the author

Travis Greene is a Ph.D. candidate at the Institute of Service Science in Taipei, Taiwan, where he studies the philosophical, ethical, and judicial implications of modern data science, machine learning algorithms, and recommender systems.

If you want to follow more of his writing, visit his Medium page Datasophy:
https://medium.com/@datasophy

He previously appeared on the ACIT Science Podcast, discussing how human rights should shape our online lives.