AI Revolution as Crisis of Meaning, and More

AI continues to be deeply panickable because it disrupts human meaning-making.

An Introduction of Sorts

In late 2022 I wrote the germ of the below, excerpted and abridged here from Panic Now? Tools for Humanizing with the kind permission of University of Tennessee Press. I finished tinkering with it for publication in early 2024.

If it is often, as an academic, gratifying to prove generally correct about the shape of things to come, it is in this case mostly distressing. Thus far and to a large extent, the AI revolution has proceeded in ways that seemed to me—beginning from my own late 2020 exploration of genAI capacities through a startup plugged into an early version of OpenAI’s large language model—deeply panickable.

At stake is a profound crisis of meaning. I fear we have not yet begun seriously to grapple with this crisis, by and large, and that its worst material sequelae unfold ever more rapidly.

This is not to say there is nothing of value in the AI revolution. And certainly, there is a great deal (an enormous amount, more than can be reckoned with; this is a point made in the chapter excerpted here) of significance to the existence, ongoing development, and the many, many intellection-products of generative automated intellection systems.

There are, indeed, open questions about AI regarding which a person may be not unduly optimistic.

For instance:

How might municipalities finetune open-weight (or even open-source) LLMs and host these locally to invite citizens into richer democratic participation? What breathtaking new discoveries in the physical world will LLMs enable (if there is any way to prevent such discoveries from becoming rent-seeking intellectual properties held by the worst oligarchs in the world)? Might AI in medicine be deployed as extraordinary supplement to rather than replacement of human connection in the most intimate and vulnerable domains of care?

The larger philosophical questions remain exciting, too, if often frightening. Are machines currently or on a near-future track to become moral persons? What do we owe them, if so? And, of course: Paperclip kill everyone?

The basic picture for human meaning-making is, however, right now pretty grim.

An AI revolution has barely begun, and already a cultural category of “AI psychosis” has emerged. A disturbingly large number of genAI chatbot users are finding themselves increasingly divorced from consensual human reality, and not just those suffering from “AI psychosis.” Social media is crawling with AI slop, vast swathes of it shared by humans who seem not to have any idea that it is such.

We are in the relatively early moments of an extended crisis of meaning and a corresponding crisis of political economy.

In societies where vast swathes of symbol production are accomplished at what the late, great anthropologist and political economist David Graeber called “bullshit jobs,” work with no natural audience for its intricate symbolic dances, generative AI presents an enormous threat to labor. But, then, this has always been the case for new modes of automation. (And the scholarly literature both warning against the threat and dismissing such warnings is voluminous.)

What is newest of all, and profoundly concerning, and yet also very difficult for many people to really grasp, is the crisis of meaning presented by the AI revolution. It is this to which most of the following excerpt is dedicated. (For more, including a substantial note apparatus, I hope a reader may find Panic Now? itself useful.)

Cover image of Panic Now? Tools for Humanizing, by Ira J. Allen

---

The AI Revolution

The years 2021 through 2023 saw AI development so rapid as to constitute the initial moments of a revolution. In this span, it became possible for virtually anyone to make a computer program produce a theoretically infinite stream of more or less human symbols (words, sentences, computer code, images, even video and “spoken” voices) at a relatively low cost. On its face, that alone is not catastrophic. To the contrary, AI advocates promised (when not threatening imminent destruction by a machinic superintelligence) to solve all manner of intractable human problems: climate change, inequality, even the drudgery of email. The devil was in the details. Or, really, in one detail: scale.

[. . .]

The AI revolution meant a radical increase in symbol-production. It was a revolution of scale. All of a sudden for many people, incomprehensibly much more symbol-production had become possible, at inconceivably faster rates and higher efficiency, with little real oversight by anybody. Humans aren’t very good at disengaging from symbols presented for our attention. Even when we ignore them, they get under our skin (that’s the whole point of internet advertising banners). We’re meaning-makers, whether we like it or not.

A great many of the new symbols being produced by machines—sentences and documents and pictures and movies—will be received by humans. (They are already.) We’ll make sense of them when we encounter them. We can’t help it. Plenty will displace other symbols, maybe displace symbols that are in some sense better. The already overwhelming welter of symbols, of information clamoring for attention from all sides, will grow exponentially worse. How will we respond?

Meanwhile, all that machine production of symbols will displace many humans previously at work producing symbols (as in advertising, media, and other white-collar jobs). The AI revolution is one dimension of polycrisis because it will change, by dint of scale, human relationships to symbol-production and -reception.

From a technological perspective, this revolution was driven especially by Large Language Models (LLMs) that drew on vast datasets and deep learning methods. The core advance was in predictive generation of “tokens,” next pieces of text or other symbols and marks that were statistically likely (but not too likely) to follow on from a given prompt. AI, in these years, produced discourse that for the first time felt authentically human to a wide range of audiences.

As I write in spring 2023, AI or automated intellection—and let us think of it as automated rather than artificial, that is, self-organizing rather than wrought or artificed, and as intellection rather than intelligence, i.e., a discursive product rather than the internal process of a conscious self—is on the verge of being universally recognized as panickable. By the time everyone agrees it is time to panic, perhaps even before you hold this book in your hands, it’s always a bit late. Still, better late than never.

In coming months and years, AI will radically reshape human meaning-making habits and processes, will shift the meaning of meaning itself. The exact unfolding of that will look uncertain for perhaps a few years yet, but panic is the right affect for getting at the magnitude—and risks—of the shift.

The symbolic architecture of human societies is changing faster than individual human minds, or even collective norms (much less policy and regulation), can adequately contain. By contrast with previous waves of automation (and yet not wholly unlike them), what is being automated today is creation of something once supposed not only exclusively, but quintessentially, human: intellection products. Intellection products, from poems and pictures to reports and technical specifications to the code that underwrites all of digitally organized life, have typically been understood as results of human cognition. Indeed, for many people intellection and its symbol-products long seemed the very essence (for better or worse) of human cognition. No longer.

The previous paragraph relied on passive voice, but of course somebody is doing the automating. It isn’t simply happening, with no actors. Two primary groups are driving the AI revolution, one very publicly and the other more quietly.

Giant tech companies with overwhelmingly huge datasets, gleaned through ubiquitous surveillance of digital life, own the proprietary LLMs that have most captured contemporary imagination. Led especially by Microsoft-backed OpenAI’s Sam Altman, tech honchos hyped AI’s ability to create intellection products through public performances of fear blended with excitement. In that general big tech ambit, Microsoft (and others) swiftly rolled out versions of software that incorporated increasingly advanced AI “assistants.” Alongside big tech, capitalist firms of every possible description pursued CaCaCo’s [the carbon-capitalism-colonialism assemblage that is “our world”] basic logic as they began navigating AI: reduce production costs relative to the market value of their products. As it became possible to automate more slices of their work-product, they automated more slices of their work-product.

[. . .]

There is every reason to suppose that the AI revolution will impose extraordinary new costs on societies ill-equipped to contain these costs, which will thus appear as crises. And yet, the ways in which early panic about the AI revolution was publicly negotiated offer an excellent lesson in what not to do with our panic. If we want to panic wisely, we’d do well to tarry a while first with how we may have panicked unwisely thus far and, equally, have unwisely dismissed panic. Much of the AI-panic discourse has centered on education, and a brief history is instructive.

In autumn of 2022, massively well-funded tech startup OpenAI released a new interface for the then-current iteration of its natural language processing engine, GPT-3. The interface was titled ChatGPT (perhaps hoping the ubiquity of chatbots, primarily in customer service applications, would render this new wonder market-soluble). For several months, academics and the general public alike became weirdly fascinated with one specific application of this interface: plagiarism and academic cheating more generally.

Because ChatGPT could output English-language text that met common metrics for clarity and style, users newly aware of AI worried that students would put it to work in their places, outputting essays for K-12 and college classes alike.

Subsequent months would show that this was an entirely realistic concern. But, it failed to get at what’s most significant about the AI revolution. The concern-reaction about student writing, as such reactions will be, was quickly dubbed a “moral panic.” Critical media literacy scholars Nolan Higdon and Allison Butler, alongside others in cultural outlets ranging from Inside Higher Ed to The New Yorker to social media and academic articles, warned against overhyped AI concerns: “we are not panicking, and we do not think any educator should,” because “ChatGPT is simply the latest tool in the century’s long saga of academic dishonesty.” AI deflationism came into vogue. It was just like calculators!

For its part, after several weeks of demonstrating ChatGPT’s power, OpenAI screwed the spigot down to a trickle. The interface, which after all costs real money to run novel commands through—primarily spent burning fossil carbon to generate electricity that powers and cools silicon-based microprocessing units, which are themselves produced at a staggering energy and greenhouse gas cost—began returning increasingly stock replies to all sorts of queries. Before long, an enterprising young Princeton student announced that he had built an engine that could detect GPT-3 writing with great accuracy. The trick, 22-year old Edward Tian explained, was to look for both “perplexity” and “burstiness.” Meanwhile, OpenAI quickly began promising its own tools to detect AI-generated text, though the company cautioned that any such tool would by nature be “not fully reliable.” For a large chunk of popular discourse, generative AI’s impacts had been reduced to cheating and, at least imaginarily, thereby resolved.

No need to panic.

But the meaning of AI for society could not be so readily resolved as that. In other fora, coders, professors, office workers, marketers, and people in dozens of other professions quietly agreed that, whether through ChatGPT or other engines, they would be automating wide swathes of onerous tasks. A third group driving the revolution was everyday symbol users, if to a lesser extent. From business plans to environmental compliance reviews, GitHub documentation to videogame coding, departmental audits to letters of recommendation, journalism to, well, the professional version of college essays (articles and books), knowledge workers had seen the future and it was now. This vision was quickly validated by economists.

Indeed, while a good portion of the chatterati was focused on whether or not to panic about one application of the AI revolution—academic dishonesty—the revolution was proceeding faster than could be adequately described in CaCaCo’s legacy media.

[. . .]

The results were astonishing. Suddenly, anyone with a fifty bucks in an OpenAI account and the willingness to overcome a couple small technical hurdles could create a website or incorporate and begin marketing a business or self-publish a book on Amazon, all without coding or really even understanding how any of the digital machinery they were using operated. A person could tell a machine to do these things without themselves ever having thought through the book or business or website. The age of domain-general AI, a theoretically infinite complexity of symbol-production from a single prompt, had arrived.

Over the course of two-and-change months in spring 2023, the scope of what humans could do with machines—and what machines could do with machines—had broadened exponentially. Even setting aside the hype bubbling characteristic of CaCaCo’s tech industry, any serious observer could see an AI revolution afoot. As you know better than I, this was hardly the end of it.

Troublingly, at first relatively few people asked what the near-term future portended by this explosive growth in AI capacity meant. At one extreme of minimizing, educators and pundits outside tech focused largely on student cheating. At the opposite extreme, tech conversations about the meaning of AI oriented toward the grandest of endpoints: artificial general intelligence, or AGI, and a threat of total human extinction.

[. . .]

No less promising and catastrophic than AGI and AI’s many explicit costs are implicit costs of the AI revolution for human meaning-making writ large. This is where it’s helpful to have automated intellection, rather than artificial intelligence, in mind.

What does it mean to automate “intelligence”? We don’t even agree on what intelligence is in the first place. There’s little question that intellection—the reception and production of symbols that mean—is a core human activity. We humans do not merely have intellects, as a noun; we intellect, as a verb. What we think that thinking itself means has a lot to do with our phenomenologies, our understandings of what it is like to be somebody. At the very least, intellection involves the reception and production of marks that mean, symbols at work both within and between individual human subjects. In an age where much formerly human intellection can be and is being produced by machines, it is precisely such understandings that stand in question.

Generative AI models all automate, or make machinic and self-organizing, something we long believed was particularly human. Their and our intellection products alike are bits of symbolic flotsam and jetsam that, once sufficiently accumulated, we call culture. AI isn’t exactly the “artificing” of such intellection products, since as models have scaled up their creators have come less and less to understand how the iterative learning of natural language processing engines actually happens. And at the same time, these engines don’t have relatively clear and singular goals: producing, say, an artificial flower that mimics a real one.

The intellection products generative AIs make, shy of AGI, are concatenations of lots and lots of smaller, recursively generated predictions. We don’t have much evidence that they constitute (yet) a machinic culture or are intelligence (per se). And yet, as all those intellection products aggregate and are received by humans, they will become human culture. We just can’t help making sense of symbols when we see them.

Of course, there’s nothing intrinsically wrong with human culture’s being shaped by nonhuman forces. Quite the contrary. The twentieth century and early twenty-first made clear to many scholars that much long supposed uniquely human (and then at least primate) wasn’t. Humans hardly hold a corner on intellection or cognition. Crows, ravens, and other corvidae, for instance, turned out to craft tools and actively reconcile after conflicts, and have paralleled great apes in performing social and physical cognition tasks.

Still, the creation of intellection products—abstract marks on the world available for meaningful decoding—has continued to strike many humans as at least a characteristically human activity. This isn’t to say we all produce and receive symbols in the same way, or even that it’s such a great a thing to do. It’s just that, whatever exactly it means (and arguing about what it means is part of what it means) intellection has continued to be a central feature of what human beings think being human entails, even long after it’s been apparent that we aren’t the only people in the world who do it.

Culture, in the broadest sense, is an accumulation of meaning-making moments in collective forms of behavior. As such, the who and how of culture-making matter very much. At stake is the meaning, if any there be, of our own lives. And, by the same token, the regularized forms of action that will and will not be available to us.

What and who are we regular humans, in a world where our usual intellection activities happen for us (on behalf of some of us), are automated (for the profit of a majority shareholder class)? What are we for when the production of symbols doesn’t need terribly many of us, but we are unable ever to stop receiving them?

Answers to these questions remain substantially unclear.

In this connection, it’s useful to think of the AI revolution as structured by at least three subcrises: alignment, disposition, and replacement. Each of these opens onto a future filled with not merely threats and opportunities, but a twinned certainty that the world will look very different and uncertainty about what that difference will be or mean. Each, moreover, represents fairly grave threats to the present political economic order of the world, threats that CaCaCo—which needs automation to reduce labor costs as energetic costs rise dramatically, but produces “surplus humanity” in the process—is profoundly ill-equipped to resolve or even contain. An AI revolution is panickable along any of the three axes of subcrisis.

Generative AI (or any other AI) that is badly aligned to human users poses a wide range of risks—up to and including existential risks. The replacement of human labor across multiple domains, if this occurs with any rapidity at all, is virtually guaranteed to be wildly socially destabilizing. And the dangers of bad algorithmic disposition are increasingly well-understood, in the production and dissemination of “fake news,” say, or the proliferation of “astroturfing” bots across all forms of social media. Rather than detail all three, I want to drill down a little on just one of these: disposition.

[. . .]

Disposition is where automated intellection, both generative and otherwise, comes home to rest in and as individual human cognition. Since I was trained as a rhetorician, I’m especially interested in the persuasive force of symbols, how intellection products orient and organize not just our current ways of being, but our capacities for becoming new versions of our selves over time.

How AI disposes humans is, centrally, a problem of our capacity or lack thereof for meaning-making.

One thing rhetoricians know is that nobody escapes being influenced, disposed in one way or another by symbols. And not just by the content of symbols, but by symbol systems writ large. Humans are symbolic animals. We each become who we are in discourse, remake ourselves and are remade by others many times over in the course of a life lived through symbolic exchange. Our very selfhood is, in many ways, a rhetorical production, a lifelong process of influencing and being organized by others in symbols. This is what it’s like to be a human somebody (and probably a crow or dog or octopus somebody, too).

On the crass backside of this fact, there’s a reason advertising and marketing are not only the infrastructural backbone of the internet as we know it, but also one of the highest-grossing industries in human history.

But influencing one another with discourse, i.e., the products of intellection, is not merely a matter of coming up with arguments to persuade your parents to get a dog, or of being persuaded by your friends or an advertisement to try a new restaurant. It’s just as much the background noise of “becoming somebody” in general as it is particular persuasive content, good or bad. For instance, the performative influence of gender norms, national ideologies, and iteratively reinforced habits of mind (rhetoricians refer to these habits and their accompanying beliefs, all together, as doxa) are all just as important as the louder flashpoints of this or that battle in ever-ongoing wars of culture.

Being a symbolic animal in general is part of what it means to be human. And being a symbolic animal means being disposed by and disposing other symbolic animals, shaping one another and being inclined to act and make meaning ourselves by arrangements of symbols, by culture.

Generative AI is on the verge of becoming a new dispositif, a new mode of disposition, for human experiences of culture—and that’s worth panicking about.

A dispositif, as political theorist Davide Panagia explains, is a kind of person-organizing assemblage of bits and pieces of culture. These assemblages reflect plenty of individual human intentions, but they do more than that. They are media that dispose us, in one way or another, simply by being the sorts of media that they are. This is about more than just content, the things we see or hear. It’s about the way media organize us at a still-deeper level. Panagia’s aim is to “develop an account of media that looks to their dispositional powers.” If we can think of media as a dispositif, he suggests, we may be able to negotiate the ways in which we are being politically organized by different ways of ordering and receiving, making meaning of, the world.

The point is that we are disposed to be what we are by the kinds of media we live and make meaning through. When Panagia says media, he means not specific sources (like FOX or CNN) or platforms (like TikTok or even ChatGPT), but rather the plural of medium. Media are bigger technological assemblages that include and contain all the sources and platforms we think of as “media” in a more everyday sense. Ultimately, any given medium is a form of life. For nearly the whole of human history, symbol-production and -reception has shaped who and how we can come to be, and so what it is like to be somebody—and, in that, has shaped what meaning itself means.

Writing, in its various forms (from hieroglyphics to videogame design, shorthand to C++), is one sort of medium. Humans make marks that mean. Generative AI is, in its fullest development, not just an expansion of writing—humans may (or may not) set the intention, but the marks that mean are all made machinically. It is a whole new medium. It’s pretty hard to know how to make sense of that.

Comprised of both material infrastructure and digital intellection products, the medium of AI disposes us toward not merely wanting or thinking, but actively becoming this, that, or the other. AI was a revolutionary medium, disposing us in truly novel ways, even before recent leaps in generative AI. But now it’s obvious to everyone.

We find ourselves increasingly organized or shaped by the automated intellection products of this medium. That can cash out at the level of the content of thoughts (your racist uncle believes George Soros invented covid-19) and the content of actions (he’s more likely to post meandering screeds or vaguely menacing memes on Facebook than he was a few years ago). More importantly, though, it cashes out at the level of forms of life and identity generally. Your uncle is becoming someone else entirely; the meaning of his life falls ever further away from your ability to make sense of it; he wants and feels and moves around in the world, but you no longer know who it is doing this wanting and feeling and moving around, or why.

What’s important is that nobody’s dominating anyone here, not exactly. Your uncle wasn’t captured, forced to become someone else. But he also didn’t sign up for it. Again, not exactly. He has become differently disposed.

His news feed shifted in accordance with algorithmic uptakes of his credit card usage, where he lingered in the grocery store, how long he hovered over a link before clicking, and millions more small intimacies of life captured as data. For the most part, no one involved in disposing your uncle differently (hadn’t he always been a little racist, anyhow?) wrenched him out of himself, plopped him down somewhere else. It’s unlikely anyone even particularly wanted anything from him, besides his eyeballs on some advertisements. But his life came to mean something different all the same.

And you think the same has not been true for you?

All that was happening even before the phase of AI revolution ushered in by generative automated intellection. Your racist uncle got worse back in the days when disposing a person some which way still required a fair bit of human labor power in the mix. And now, when it scarcely any longer does?

Dark predictions abound, foretelling how generative AI will shape us, at the level of the content of our thoughts and actions. Many are surely correct.

We should be at least as worried about how the medium of generative AI will dispose us, arrange our ways of being and meaning and acting. What does a human life mean within the web of an AI revolution, where intellection products make a humanly received inhuman culture? What can it mean to “belong” to cultures made with machinic ultra-efficiency, on behalf of CaCaCo’s biggest beneficiaries?

To what will the aggregate culture of automated intellection products dispose us? Panagia observes that, “like the practices of dispositio in classical rhetoric, the dispositif’s role is not that of transmission of meaning but of arranging moving parts.” Toward what wholes and whose ends will the AI revolution move us, we its parts and means who perhaps mean less than ever even as our lives are captured ever finely more by data harvesting apparatuses?

From the algorithmic disposition of attention to the possibility of totalizing disposition by machine-generated symbols to the construction of human lives in environments shaped machinically to serve (i.e., “in alignment with”) the preferences of CaCaCo’s majority shareholder class, there is good reason to wonder what most of us are even for in an era of supercharged AI.

Nor is this an idle question. Given that international regulation of fuzzy, emerging threats on behalf of a general humanity has been defunct since at least the 1989-1992 interval, what’s to stop the small portion of humanity that holds virtually all the cards from deciding that the rest of us, being unnecessary, must fall by the wayside if their “longtermist” vision of humanity is to survive the climate crisis, sixth mass extinction, and novel chemical crisis? How might we find ourselves organized for the worse, in our own minds and in our ways of being together or seeing one another, along the way?

Might generative AI, controlled by CaCaCo’s majority shareholder class, even dispose much of humanity to be more readily disposed of as the polycrisis unfolds?

Less than a Conclusion

I wish my concluding question from that section of Panic Now? no longer seemed relevant. I wish AI had at least proven thus far a beneficent new form of dispositif. Alas, matters stand very much to the contrary.

Each day offers new suggestions that a global sociopath class would like nothing more than to dispose of the rest of us, the better to consume uninterruptedly as the planet burns. Or, at least, would like AI to dispose us toward greater obedience as they finish up with their bunkers and what’s left of democracy alike.

For reasons detailed in the larger book, I am not sanguine about the possibilities for regulating AI on behalf of democracy or a broadly human capacity for meaning-making. These are the sorts of solutions always being proposed by well-meaning commentators, and that miss over and over again the larger and darker forest.

The AI revolution is underway. It will continue and accelerate and transform all our lives, even in the face of a likely crash or short-term deflation of its presently enabling market bubble. The meanings we make now of the AI dispositif itself will determine, to a very large extent, our own future capacities for meaning-making.

The challenge, for scholars as for everyone else, is at once to resist and to salvage, resist and salvage.

What variants of automated intellection may be salvaged from the very maw of the machine? Where does AI, in what forms, serve the interests not of a majority shareholder class but of ordinary people? How? Why? What forms of resistance to what forms of “progress” in the large-scale deployment of automated intellection may prove most productive, most protective of human capacities for meaning-making and for living meaningful lives? How? Why?

The market’s answer thus far has simply been “more AI.” Our unfolding crisis of meaning, a slow-motion catastrophe that is of a piece with the rest of the polycrisis, suggests that this is the wrong answer.

The AI revolution remains as panickable today as when I began writing Panic Now?. I am heartened that ever more of us are finding ways to panic wisely, inventing trajectories at odds with an AI-induced crisis of meaning (and more).

Let us do so faster.

---

From Ira J. Allen’s Panic Now?: Tools for Humanizing. Copyright © 2024 by The University of Tennessee Press. Reprinted by permission. See https://utpress.org/title/panic-now/ for more.