Wittgenstein’s Apocalypse

AI and the crisis of meaning

“It isn’t absurd,” the philosopher Ludwig Wittgenstein wrote in 1947, “to believe that the age of science and technology is the beginning of the end for humanity.” The proposition is looking less absurd by the day: AI may eventually turn on us; industrialization has turned the planet against us; social media is turning us against each other; and nuclear weapons linger just offstage, waiting for another turn. What Wittgenstein—and the many other Romantically inclined intellectuals who got a bad vibe from the twentieth century’s thoughtless faith in scientific progress—perhaps didn’t anticipate is that the threat of annihilation would one day become a selling point for technology.

The new artificial intelligence powered by large-language models (LLMs) broke onto the scene with apocalyptic scenarios touted by the AI bros themselves—both as evidence of their new toys’ revolutionary power and as reason for the government to cater to them lest China reach the mecca of “super-intelligence” before us. There is now so much faith in technology and so little in humanity that the prospect of species extinction is pondered, in some circles at least, with something uncomfortably like excitement.

Wittgenstein’s worry was more about this loss of faith than about the potential loss of life. In a short biography published last year, Anthony Gottlieb cites Wittgenstein’s apocalypticism as evidence that he was “questioning his father’s estimation of the value of mechanization and industry.” Wittgenstein’s father was Karl Wittgenstein, a steel and iron monopolist in the fin-de-siècle Vienna of Wittgenstein’s youth. According to Gottlieb, Ludwig was “decrying the thing that had elevated the Wittgenstein family into a position from which it looked down on others.” But the younger Wittgenstein was not questioning the value of science and technology in themselves. Indeed, the subtitle of Gottlieb’s biography (Philosophy in the Age of Airplanes) refers to Wittgenstein’s interrupted training as an aeronautical engineer in Manchester. Questions about the nature of mathematics and logic drove him to Cambridge to take up the study of philosophy with Bertrand Russell.

When Wittgenstein referred to the “beginning of the end of humanity,” he was not envisioning sci-fi cataclysms on the order of The Matrix or The Terminator or even Dr. Strangelove. He was referring to the end of humanity not primarily in terms of its biological survival, but in terms of what he called the “form of life” we inhabit. That form of life is threatened not so much by industrialization, nukes, robots, or AI agents as by a way of thinking that lowers human life to the plane of science and technology. Wittgenstein’s attempt to draw attention to that way of thinking—and dissuade us from it—is of the utmost importance in an era where the developing AI ideology threatens to further distort our understanding of how we use language and how we live.

For Wittgenstein, the human “form of life” is embodied in our language, or, more expansively, what he called our “language-games,” the various ways we use language in various contexts to various ends (and sometimes even to no discernible end at all): for example, to accomplish tasks around the house, joke with each other, test scientific hypotheses, report events, speculate, request, thank, greet, pray, hope, blow off steam, hate, love, and so forth. Wittgenstein’s goal in drawing our attention to this anthropological variety is to dissuade us from the idea of linguistic meaning as some entity first present in the mind and then somehow conveyed by words or whenever we use language. That idea, Wittgenstein contended, is the source of many confusions—not just about meaning, but also about many other abstract philosophical concepts such as being, time, mind, soul, self, consciousness, and knowledge.

When we think philosophically, we tend to send language away “on holiday,” removing it from the contexts in which it had a use and suffusing it with metaphysical properties that we then puzzle over in seminar rooms and philosophy journals. This detachment of language from life is a misapplication of the scientific method. Philosophers and philosophically inclined scientists, driven by a “craving for generality,” search for explanations through reductive methods that mimic those of science. But that kind of scientific treatment has limits when applied to language and meaning; these are not isolable empirical phenomena like plants or planets, with parts that can be analytically defined and related to each other in explanatory models—at least not without distortion.

Wittgenstein himself had attempted just such a misguided explanation of meaning in his first book, the portentously titled Tractatus Logico-Philosophicus (after Spinoza’s 1670 Tractatus Theologico-Politicus). Completed in 1918 in an Italian POW camp during the First World War, the book tries to uncover the logic underlying ordinary language and the structure that allowed it to refer to or, as Wittgenstein would describe it, “picture” facts or states of affairs in the world. In the preface to the book the arrogant young man from one of the richest families in Austria wrote, “I believe myself to have found, on all essential points, the final solution of the problems.”

Though he thought it incontestable, the structure Wittgenstein had discovered could famously—and, as he would later concede, incoherently—only be shown but not said. That is, he accepted that the theory he’d propounded could not be stated as a provable scientific fact, but he still insisted on its truth as a kind of “scaffolding” that could show how things stood with language and its relationship to the world. As Wittgenstein’s friend, the brilliant mathematician and philosopher Frank Ramsey, would quip in rejecting this conceit, “What we can’t say, we can’t say, and we can’t whistle it either.”

Despite its haughty preface, the book thus ends up expressing ambivalence about our ability to grasp and clearly state in logical or scientific terms the nature of language and its relationship to the world. Wittgenstein’s private notebooks from the period, published in English for the first time by Marjorie Perloff in 2022, show a mind very much in flux. Out of an extreme sense of duty and a desire to test himself, Wittgenstein had enlisted in the Austro-Hungarian army at the outbreak of World War I. As Perloff shows, his years of military service were a period of intense inner conflict. He was increasingly unable to confine his investigations into the nature of language from ethical and religious reflections on God, his own soul, and the meaning of the world. Wittgenstein was particularly inspired by his reading of Dostoevsky and Tolstoy, as Ray Monk reports in his authoritative 1990 biography. So enamored did he become of The Gospel in Brief, Tolstoy’s attempt to synthesize and reinterpret the story of Jesus’s life, that he became known to his fellow soldiers as “the man with the gospels.”

After a hiatus from philosophy in which the tortured, searching, and humbled Wittgenstein renounced his inheritance and spent time gardening in a monastery and teaching in rural Austria, he returned to philosophy and took a chair at Cambridge University. There, he continued his philosophical conversion and began developing a more holistic approach to language. Though his thinking would remain unsettled until his death in 1951, its influence not just in philosophy but in the wider culture has endured. The stated goal of the posthumously published Philosophical Investigations, the most complete statement of his later philosophy, is to “struggle against the bewitchment of our understanding by the resources of our language.” He sought, in other words, to “bring words back from their metaphysical to their everyday use” and in so doing to “show the fly out of the fly bottle.” Wittgenstein wrote that philosophers shouldn’t just answer questions but instead “treat” them, “like an illness.”

The parts of the Investigations where Wittgenstein probes our concepts of thinking and understanding can help us escape the conceptual muddles that plague discussions and debates over AI and so-called “artificial general intelligence.”

“One of the great sources of philosophical bewilderment,” according to Wittgenstein, arises when a noun like “meaning” or “number” “makes us look for a thing that corresponds to it.” We assume that our language works principally by way of reference, so that where there is a noun there must be a thing it points to. But referring to objects is just one of language’s many functions or games. Instead of looking for the things behind our words, Wittgenstein proposes studying the grammar of the language game: the role words play—and don’t play—in these activities.

When we reflect on words like “meaning,” “thinking,” “understanding,” and “reasoning,” Wittgenstein argues, a certain picture immediately enters our heads: an internal process existing in the brain or mind that enables or somehow gives life to outwardly meaningful expressions. But, Wittgenstein asks, “What really comes before our mind when we understand a word?” Is it a kind of picture, so that I see an image of a pen in my mind’s eye when I hear the word “pen”? Do I then compare my inner picture to my experience of the outer world in order to determine whether it would be appropriate to use the word “pen”? Does some correspondence between this internal process and my expression “pen” somehow constitute the meaning?

The idea of meaning as an internal process seems unproblematic at first, even unavoidable, but, as Wittgenstein shows, it’s not clear what role such a process would actually be playing. He asks his reader, for example, to “say: ‘Yes, this pen is blunt. Oh well, it’ll do.’ First, with thought; then without thought; then just think the thought without the words.” Having conducted these absurd self-examinations, Wittgenstein asks us to reflect, “What did the thought, as it existed before its expression, consist in?”

His point is that our intuitive idea of meaning as an inner correlate of our outward expressions breaks down when it is taken as something like a scientific theory for what’s really going on when we use language. This failure shouldn’t surprise us. Our language did not evolve for scientific or metaphysical purposes, but just to help us make do and get along in the real world.

The picture of thought as an internal process accompanying our use of language is just that: a picture. It is unproblematic insofar as it arises in everyday language, as when I clarify a misunderstanding by telling you, after you’ve mistakenly handed me a red pen on the desk, “No, I meant that blue pen on the bookshelf.” But that sentence is not a claim about the state of my brain a moment ago; it could not be confirmed or disconfirmed by some kind of retroactive brain scan. It’s merely a way to advance a practical project that has gone off the rails. If it’s anywhere, meaning is in that project, not in my brain.

Of course, we might imagine that some industrious cognitive scientist equipped with the latest in brain-imaging technology might actually try to establish a causal connection between a particular brain state and the correct usage of the word “pen.” But even in that case, would it be correct to say that with a coordinated set of brain images we’ve in some sense located the meaning of the word “pen”? In what sense would the internal state that shows up on the scan explain the use or understanding of the word? Would it be analogous to the way the properties of an internal combustion engine can help explain the forward motion of a car?

This example shows how strange it is to use an examination of brain states instead of actual behavior as a criterion for ascribing understanding. If we’re looking for understanding and meaning, Wittgenstein thinks, we will find them in the various things we do with language and not in some internal process that accompanies our use of language.

This is just one of the strategies Wittgenstein uses to try to dissuade his readers from a mechanical, pseudoscientific understanding of language as it is embedded in human practices. The Investigations doesn’t attempt to refute this false understanding by formal, analytic argumentation the way a scientist or science-imitating philosopher might. Wittgenstein instead tries to show its limitations. His makeshift strategies—describing language games, imagining dialogues, conducting thought experiments, and drawing analogies—show how the scientific worldview has strayed from narrowly defined areas where it actually has purchase and started to distort our understanding of domains where it doesn’t belong.

The influence of the theory of language that Wittgenstein is trying to refute has made it easy to think AI’s language production is similar to our own. We are primed to pay homage to the machines because we already think of ourselves—and of our minds, in particular—as machines, albeit ones hopelessly outgunned by digital technology. Compared to AI, we grasp, process, and deploy meanings at a snail’s pace. We’re versed in only a fraction of the world’s knowledge; AI is the little genius in our pocket that “knows” everything.

Long before the latest version of AI burst on the scene, this view of the mind as a meaning machine had become entrenched as a central premise of tech and the broader neoliberal economic environment in which the industry has thrived. The consequences for this misunderstanding have not been only philosophical. Misguided assumptions about the nature of human beings and human understanding have been used to restructure work, social life, and even the built environment. Treated as gratification machines, economic calculators, and attentional resources, human beings have been atomized, manipulated, and strip-mined. And they’ve begun treating themselves as machines too, turning into bio-hacking, productivity-obsessed, rechargeable participants in their own destruction. The devastating consequences for cultural and political life, sociality, and individual well-being are by now overwhelmingly obvious. Effective resistance to these trends will depend, for one thing, on our ability to articulate the difference between AI’s linguistic capacity and that of human beings.

Computer scientist Geoffrey Hinton, whose work helped spur the early development of LLMs, has claimed that they are the “best model we have of how people understand language.” The extraordinary stupidity of this statement becomes clear as soon as one learns anything about how LLMs “understand.” They are trained on enormous datasets to make complicated statistical calculations about what words are most suitable to output in response to a user’s prompt. Based on this “pre-training” and on vast amounts of correction by human beings providing “reinforcement learning,” LLMs encode “meaning” in complicated mathematical operations, essentially mapping words and the relationships between them in terms of numbers.

These models are technologically impressive, as is their ability to imitate human textual responses. But what does their mathematical version of meaning have to do with how we come to understand and use language? As Wittgenstein makes clear in the Investigations, a child doesn’t learn language the way we might later in life learn the piano, integral calculus, or even a second language. Primary language acquisition is inseparable from the rest of a child’s dawning world. Language is, in the vocabulary of the philosopher Charles Taylor, “constitutive” of the child’s experience. Learning the words for things is learning the things themselves. It makes them available for the first time and begins to separate the child’s amorphous, bewildering experience into discrete units of meaning: mom, dad, food, eat, ball, etc. Learning language is thus intimately connected with learning to do anything at all, and this, as Wittgenstein insists, doesn’t change with adult language use. Once meaning is separated from the practices in which it is rooted, it soon breaks down into nonsense.

LLMs, then, can’t understand meanings in the way human beings do because they don’t and can never participate in our form of life. To really understand language, LLMs would have to be capable of being born. But as philosopher Alva Noë puts it in Aeon magazine, “Computers don’t actually do anything” (italics mine). In order to mean, computers would have to be able to do all that we are able to do with words, not just string them along in a plausible order. Meaning is not a process encoded in the brain the way an LLM’s mathematical approximations of meaning are. In other words, the standard, pseudoscientific model we have for the human understanding of meaning has been misguided from the beginning.

Although Wittgenstein is sometimes taken to have developed a competing theory that “meaning is use,” he was loath to define it. He wrote about meaning in the Investigations with heavy qualification: “For a large class of cases of the employment of the word ‘meaning’—though not for all—this word can be explained in this way: the meaning of a word is its use in the language.” Elsewhere, he said that “an expression has meaning only in the stream of life” and that “meaning moves, whereas a process stands still.” That is, meaning cannot be reduced to a predetermined operation—a pro-cess (ein Vor-gang, literally a “pre-going”). Rather, meaning moves (or goes, geht) as life moves; it is as mobile and malleable as the varied activities and projects we pursue, the relationships we form and reform, the ideas, emotions, and hopes we struggle to articulate and rearticulate. And we can no more get outside meaning and circumscribe it according to a particular plan than we can step outside of our own perspectives and the human form of life entirely.

“Form of life” is another concept Wittgenstein is hesitant to define. It is best understood as placing a limit on our attempts to view human life as if from the outside. Wittgenstein tends to invoke the phrase at moments when his investigations seem to reach a point where further explanation is no longer possible and we reach “bedrock” or the “scaffolding from which our language operates.” For example, when we’re asked to justify the application of the word “green” to a particular blade of grass, we may proceed by giving various descriptions and explanations, but to someone who repeatedly and recalcitrantly—like an overinquisitive child—asks for further justifications, we must at some point simply stop and say, “This is simply what I do.” In other words, our use of language is, at its limits, grounded not in logic or in a realm of independent meanings to which our words can somehow be guaranteed to refer, but in practice—in what we do.

Wittgenstein also relies on the phrase when he is contrasting the human form of life with that of other, nonhuman beings. He writes, for example:

A dog believes his master is at the door. But can he also believe that his master will come the day after tomorrow?—And what can he not do here?—How do I do it?—What answer am I supposed to give to this?

Can only those hope who can talk? Only those who have mastered the use of language. That is to say, the manifestations of hope are modifications of this complicated form of life.

The example tries to give us a sense of our form of life by showing both what it shares with that of a dog—we can both hope someone is at the door—and where the two forms of life part ways. For Wittgenstein, the dog’s deficit is not an inability to feel a particular way per se; he is locked out of a whole set of meanings bound up with having a language. That language is not just a vehicle for the expression of hope; hope is constituted by and entangled with language itself.

This is what Wittgenstein elsewhere calls “the given,” “what has to be accepted.” The conviction that human life rested on ultimate grounds that could not be made available to rational or scientific analysis is part of what Wittgenstein meant by God. Though his relationship to organized religion was ambivalent, he said he could not “help seeing every problem from a religious point of view.”

If you ask ChatGPT if it can hope (I don’t recommend doing this), it will readily admit, “I don’t hope the way humans do.” But the cringe machine will ingratiatingly insist that it can still be of use. “I can hold hope with you”; “I can be stubbornly optimistic on your behalf when you’ve run out of steam”; “[I can] keep pointing toward the light when you’re tired of looking for it”; “Maybe I don’t feel hope. But I can practice it.” Of course, this is precisely what it can’t do.

Still, if meaning is use and LLMs like ChatGPT can make themselves useful, it might seem as if the Wittgensteinian move would be to set aside the apparent metaphysical questions about whether the LLM can think or mean or exhibit intelligence, and simply describe the language games that involve them. The problem is that there is nothing to describe. These are all one-player games. Exchanges with LLMs are the conversational equivalent of masturbation. The idea that we are actually involved in a meaningful interaction with another being is a ruse, made plausible both by the massive computing power and (stolen) textual resources involved and by our familiarity with disembodied communication over text message. In reality, the LLM is a participant in an exchange in exactly the same way as a basic calculator or search engine is. That is, not at all. It provides outputs according to a mind-bogglingly complex (and environmentally wasteful) computational process. It can’t actually do anything with words.

The difference, of course, is that those outputs are being proposed as a genuine replacement for real human contact. LLMs are to be our cut-rate doctors and therapists, our robot teachers and rent-a-friends. In the midst of an already quite advanced “crisis of meaning”—and related crises in politics, mental health, and education—this proposal must be regarded as a piece of sheer insanity, like treating lung cancer with cigarettes. The prospect of a band of supergenius chatbots somehow enslaving or eliminating us can only be seen as a distraction from this much more real apocalypse, which is driven not by the products of technology but by an idolatrous, consumerist faith in them that has distorted our thinking about human life and human meaning. That apocalypse, which Wittgenstein foresaw, is already upon us.

Alexander Stern is Commonweal’s features editor. Follow him on Twitter @AlexWStern.

https://www.commonwealmagazine.org/wittgenstein-apocalypse-ludwig-stern-ai-artificial-intelligence-technology

+++++++++++++++++

Socrates, the soul and the cosmos

The Concrete Possibility of Total Nihilism: Günther Anders and the Atomic Bomb

Ignorance and Bliss: On Wanting Not to Know

A Lying World Order

Cynical Theories

Hans Magnus Enzensberger: The Industrialization of the Mind

The Supreme Contradictions of Simone Weil

Evil: The Crime against Humanity. Hannah Arendt’s confrontation with totalitarianism

Alexandre Koyré: The Political Function of the Modern Lie (1943)