# Find Rituals When It Allows
Published: 2026-04-17
Canonical: https://journal.xiaotianfanx.com/journal/find-rituals-when-it-allows

On "Work and meaning in the age of AI" by Daniel Susskind, and "The Simple Macroeconomics of AI" by Daron Acemoglu, and Anthropic's research on labor market impacts.



	Before we delve into the normative space that Susskind's "meaning problem" opens up, I'd like to take a step back—or rather, a step down—into the empirical territory first. It seems to me that the keyword in both Susskind's and Acemoglu's analyses, and indeed through much of our ongoing discussion, is discrepancy. Discrepancy between what AI could theoretically do and what it is actually doing; discrepancy between demographic groups; discrepancy between the hype and the reality; discrepancy between the "meaning of meaning" across individuals, cultures, and historical moments; and eventually, discrepancy between our normative ambitions and the positive ground on which they must stand. 







The Discrepancy Between Theory and Reality



	The first and perhaps most obvious discrepancy is between what AI systems could theoretically accomplish—in terms of task automation, delegation, and "replacement"—and the actual status quo of AI adoption in real lives. The Hulten's theorem application—GDP share of impacted tasks multiplied by average task-level cost savings—yields a TFP gain of roughly 0.71% over ten years at the upper bound, dropping to about 0.55% when accounting for the distinction between easy-to-learn and hard-to-learn tasks. These numbers are, to put it politely, extremely far below the "typical" market projections that have dominated the popular discourse—whether a 7% GDP increase or $17–25 trillion in global economic value. 



	I'd say the easy-hard task distinction does offer some explanation for the discrepancy. Easy tasks—those with a clear mapping between action and outcome, with observable success metrics—are exactly where current LLMs shine, while the hard tasks lack clear outcome measures and involve complex, high-dimensional interactions between action and context. In those domains, AI can at best converge to average human performance because it learns from human behavior rather than from objective outcome data. That alone should temper the revolutionary rhetoric.



	Anthropic's own research on labor market impacts this March provides a rather illustrative visualization of this gap. Their radial chart plots "theoretical AI coverage" against "observed AI coverage" by occupational category, where the theoretical coverage looms impressively over the observed coverage. At first glance, the gap is staggering as LLMs are theoretically capable of covering 94% of tasks in computer and math, 90% in Offic & Admin. Yet observed coverage sits at only 33% for computer and math, for instance.







	But I'd be cautious with the visualization itself—and it ties back to the broader epistemological point about how data is presented and consumed at all. The radial chart itself is proportional to the square of the plotted value, which makes the red (observed) area look far smaller relative to the blue (theoretical) than a linear bar chart would suggest. Whether this was Anthropic's intention or merely a design choice matters—as the visual rhetoric of the chart amplifies the impression of untapped potential (presumably aiming at promoting a promising future, in theory). Moreover, the blue area itself is also theoretical—it is their estimate of what current AI could do, not what it is doing as AI is not performing 20% of creative work now—and I would argue, with some confidence, never will in the sense that "performing" is typically intended by such metrics. The entire framing risks conflating demonstrated capability on isolated benchmarks with actual integration into the messy, contextual, socially embedded reality of human work. To put it less charitably: the entire thing is, to a significant extent, hype visualized. 



	On a broader scale, if we reconsider the infiltration rate of AI among the global population, an assessment from February 2026 suggests that around 84% of the world population—roughly 6.8 billion people—has never used AI. Among the rest, about 1.3 billion are free chatbot users, and fewer than 25 million pay for AI services—comprising less than 0.3% of the global population. Those who use coding scaffolds like Copilot or Claude Code number perhaps 2 to 5 million, or about 0.04%. These numbers are decently small, and I guess they should evidently give pause to anyone tempted to speak of AI as if it were already a universal presence.







	However—and I think this is where the discrepancy gets interesting, rather than merely deflating—if we compare AI's infiltration timeline to, say, global smartphone coverage (which reached about 67% by 2025 after more than a decade of mass-market availability), AI's adoption rate could actually be seen as remarkably fast given its recency. Similarly, a Gallup report released just days ago finds that half of U.S. employees now use AI at work—a rather staggering figure for a technology that only became widely accessible in late 2022. Both things can be true simultaneously—AI adoption is soaring at an unprecedented pace, and there remains a substantial gap between the theoretical and the actual.



	The reasonable takeaway, from my perspective, is that we should resist both extremes (as a usual stance I would take), and the truth—for those of us who spend a decent amount of time catching up with the latest-latest, like I myself—is that someone who regularly follows every new model release, every benchmark, every capability announcement can easily fall into an extremely hyped information cocoon. 







The Discrepancy Between and Among Demographic Groups



	Acemoglu's finding that AI exposure is more equally distributed across demographic groups than pre-AI automation, unlike robots and manufacturing software, which disproportionately displaced middle-skill production and clerical workers—hitting specific demographics (particularly less-educated men in manufacturing regions) with devastating precision—sounds, at first pass. But the diffuseness of the exposure, unfortunately, doesn't automatically translate into equity of outcome. It is also shown that overall between-group inequality may increase slightly, the capital-labor income gap will widen, and real wages for underrepresented and already disadvantaged groups are actually predicted to decline. As modeled, even when AI improves low-skill workers' productivity in specific tasks, general equilibrium adjustments can result in those workers being displaced from the improved tasks, pushed into competition for other tasks where their relative disadvantage is greater. 



	Moreover, if we take a broader scope than Acemoglu's U.S.-focused analysis—considering global populations, the developing world, communities with limited digital infrastructure, populations with no access to computational resources—the discrepancy probably would deepen significantly. The 84% of the global population that has never used AI is not uniformly distributed but disproportionately constituted by those already disadvantaged along existing axes of inequality: geography, income, education, language, infrastructure, etc. The AI economy, such as it exists, is being built by and largely for a very narrow slice of the global population, while the infrastructure that sustains AI is owned and controlled by a few, and the benefits, at least in the near term, accrue accordingly.



	Beyond demographics, we probably should also note the discrepancy between what Acemoglu considers the most beneficial applications of AI to be and where the industry is currently investing, as this is where the empirical side of the story and our broader normative project would converge. The gap between what AI could do for human work and what it is doing may be a failure of will—or more precisely, a will that is oriented toward profit maximization rather than toward what Acemoglu calls "good jobs" and what Susskind frames as the broader "meaning problem." This naturally leads us to the normative question: should we navigate the industry toward something else—and can we? That, eventually, is the question about meaning.1

 





The Discrepancy Between Work and Meaning



	Now we finally arrive at Susskind—and the discrepancy here is, if anything, more layered than the empirical ones. As Susskind notes, the relationship between work and meaning is deeply heterogeneous—across individuals, groups, regions, cultures, and historical moments. That heterogeneity is the relationship—or rather, it is the reason we cannot speak of a single relationship at all between work and meaning. Some people derive immense meaning from their work; many do not. Some work is meaningful in some reasonably "objective" sense; much of it is not (Graeber's "bullshit jobs"). The balance differs dramatically between a subsistence farmer in rural South Asia and a student in Abu Dhabi (now relocated back to Shanghai) typing this very entry—and neither of us gets to generalize our experience.



	As discussed in class, the Freudian understanding of a mature person—one who has the capability to love and to work, and to find meaning in each or both—captures something that resonates outside psychoanalytic frameworks, echoing Aristotle's notion of ergon—that there is a distinctive function of human beings, and flourishing consists in performing it well—although Aristotle himself would have been horrified to identify that function with the manual labor he relegated to the artisan class.



	But, as Susskind relates, whether work is necessarily linked to meaning by itself is still very much open and heterogeneous across different thinkers, let alone different people. Even among scholars, the "meaning of meaning" varies—some define it subjectively (what the worker feels), some objectively (whether the work actually contributes to some good), some relationally (work as a site of social connection and identity), some teleologically (work as the expression of a calling or vocation). The fact that these definitions don't converge is indeed a substantial normative problem that makes normative planning extremely difficult on a practical level.





 

On Arendt, Construction, and Liberation

 

	Hannah Arendt's observation—that we live in a "society of laborers" that "does no longer know of those other higher and more meaningful activities for the sake of which this freedom would deserve to be won"—is both an empirical claim and a challenge in my understanding. The empirical-positive claim is that our social order is built so thoroughly around work that we have lost the capacity to imagine meaning coming from elsewhere. The challenge, at least at this moment, is whether this is a contingent fact of our historical moment or a deeper truth about human nature.

 

	I won't argue fully for the adoption or refusal of Arendt's point—but I think it works both for and against the argument regarding meaning in a post-AI era, and that's precisely what makes it instructive. On the one hand, supposing AI does disrupt work and takes away most of it (even only most of the "routine" and "easy" work, in Acemoglu's terms), then, given the current status quo of society, it would surely dislocate the meaning-making process on a macro level. People who derive their identity, daily structure, social relationships, and sense of purpose from work could be stripped of all of those simultaneously—and a UBI check probably would not fill the gap (which gives JGS its edge). On the other hand, precisely because this centrality of work is a social and historical construction—not a feature of human nature as such (I suppose the ancient Greeks would have found our worship of work, if indeed true, bizarre)—it is completely possible, even likely, that in a post-AI era we would eventually detach what is currently conceived as "work" from meaning-making. The construction can be reconstructed anyway in a general de-and-re-construction circle. This shift could eventually liberate the very concept of working—or at least widen it beyond the narrow definition of "paid employment in a labor market" that currently dominates.

 



 

The Privilege

 

	Of course, it's always worth noticing—as I always notice it about myself along this thread—that it is exceedingly easy for us to discuss the relationship between work and meaning from a privileged standpoint. When I say "work," I presumably refer to the kinds of work that are meaningful to me, potentially even at the level of vocation—the calling. Writing, thinking, composing, creating, researching—the things I would do regardless of whether anyone paid me.2



	However, as Susskind also says, there are forms of work that carry what economists have modeled as "pure disutility"—work that is chosen hardly because it provides meaning in any obviously imaginable way and seems very much improbable for us to approach willingly with a typical framework of meaning and flourishing, but is chosen because the alternative (starvation, homelessness, debt) is worse. For those in such positions, the relationship between work and meaning may not merely be absent but actively negative: work is what stands between them and any possibility of meaning-making. 



	How are those works treated as we discuss them in a post-AI era, if ever? In our seminar, the market-based pay system, where pay is inversely related to job desirability, is discussed. The logic is elegant: jobs that nobody wants to do are compensated more highly precisely because their unpleasantness constitutes a cost that must be offset, while jobs that people would be willing to do for free (or even pay to do) can command lower wages. I'd say it is, in a sense, a formalization of the "compensating differentials" that Adam Smith observed centuries ago, but elevated to a principle of distributive design rather than a market accident.



	Regarding being privileged myself, I would personally agree with and would like to participate in such a scheme—precisely because my understanding of work is, admittedly, quite broad, considering everything I do in life—besides meeting basic survival needs as a human being—to be a kind of "working." In fact, I am often criticized by my friends for failing to distinguish between work and life, or for failing at "work-life balance," as I do perceive that "balance" as nonexistent. As long as I am still living, it is a form of working, per se, in my experience, a mode of engagement and thus a mode of work in the deepest sense. Obviously, this is quite blameworthy as elitism or a privileged understanding—and rightly so. It may only be applicable to a few people on this planet, if not only to me. (Though I suspect I'm not entirely alone in this.)







Maslow and the Skyscraper



	Another related model we discussed in class—and one I had debated with friends a few weeks prior—is Maslow's hierarchy of needs (or more accurately, his motivational theory, later visualized as a pyramid). The standard image with the widest base is physiological survival, ascending through safety, belonging, esteem, and finally self-actualization—at the narrow tip. The conventional reading is that "higher" needs (meaning, self-realization) can only be pursued once "lower" needs (food, shelter, security) are satisfied.



	What I had as an argument—and this is speculative, for sure—is to envision a transformation of that strict pyramid towards something more like a skyscraper architecture, or somewhere in between: where the higher levels become wider and wider rather than remaining a narrow tiptoe accessible only to a few. In other words, if AI (or any technological development) genuinely alleviates the burden of survival-maintaining labor—automating the base of the pyramid—then more people could, in principle, access the upper levels. The pyramid inverts, or at least widens, at the top. Obviously, this is very much another elite perspective, but it also links back, somewhat hauntingly, to the statistics we discussed earlier—about how much of the global population is actually using AI, let alone benefiting from it. If the base of the pyramid is still unsecured for 84% of the world's population in terms of AI access, then my skyscraper vision is, at best, aspirational and at worst, a projection of my own narrow context onto a species that overwhelmingly does not share it.







Eventually, Values—and Beyond



	That said, eventually we'll have to go back to our own limitations and values. We have to realize our values are only our beliefs, right? And when it comes to systematic design or policymaking, building upon the normative beliefs we have and relying on our own perspective could be extremely biased. I guess that's a part of the reason we do need empirical evidence to always correct our notions—as I hope this entry has at least attempted to demonstrate, at its minimum, by grounding the normative discussion in light of some positive data before letting any philosophical ambitions take over.



	Nonetheless, as a semi-pure conjecture, let me entertain the scenario with this radical but cliched thought experiment: what if most jobs are indeed automated by AI or whatever the next technology is? To some extent, I do acknowledge that meaning-making is linked to work in our current social configuration. However, I would say—perhaps boldly and naively—that humankind in general will always be fine regarding our nature as an adaptive species. Following the thread of Arendt's observation, we will probably widen our definition of work, or find some substitute that fills the gap of our time, or (more likely) both simultaneously and messily. Essentially, from my perspective, it is impossible to do "nothing" in the sense of literal nothingness. As long as we are still living, there is something we must be doing—even without all the survival-related work that we currently and reluctantly do to fulfill our basic needs. The doing may not look like "work" in the labor-market sense. But it will be activity, and activity—directed, chosen, or even merely endured—is the raw material from which meaning has always been made.



	What will those things be? I'm not sure, of course. But I'm certain that we will do something. That process of doing—and finding something to do—may itself substitute for the "work" that some of us now consider a necessary component for making meaning. Envision someone who decides to do nothing: physically doing nothing, sitting there for an entire day. Although they are not actively doing anything physically—or even intellectually, in any productive sense—they are, in a way, "zooming out," the very decision to stay in that state is, in fact, a way of acting and doing in reality. It is a choice, even if it is the choice of stillness. As long as that kind of choosing remains available to us, I don't perceive any fundamental risk of losing meaning at all. The risk is real and practical and distributional and enormous—but it is not, I think, existential. Of course, it is a question highly related to whether we have free will and genuine choices. Although my personal stance may actually incline towards skepticism on that front—towards something closer to hard incompatibilism tentatively—the feeling of having such a sense is significant enough. And I emphasize "feeling" very deliberately, even if it may not survive the most rigorous metaphysical scrutiny—at the end of the day, it is the phenomenological ground on which we actually live. And that ground, however shakable, is not nothing.







The Kojèvian Postscript



	In fact, another thread or debate we may draw upon—one I've discussed with friends in the context of the broader "crisis of modernity"—is the dialogue between Leo Strauss and Alexandre Kojève. The Kojèvian scenario is, in its most extreme form, a vision of the "end of history" in which all substantive problems have been solved, all needs have been met, and what remains is essentially ritual—the formal repetition of activities that have lost their original purpose but persist as cultural forms. It is arguably the ultimate post-work society—a world in which the question "what do we do now?" has no answer except "whatever we've been doing, but emptied of its historical necessity."



	Even in that scenario—a scenario that many would find bleak if not hopeless—I would still argue it doesn't mean that all meanings have collapsed for the possibility of human flourishing. It is indeed envisionable that we will derive meaning from rituals—maybe not in the formal, religious, or institutional sense, but in the sense of practice: repeated, disciplined, chosen activities that structure our time and attention.3 This is already present in everyday contemporary life with things that are considered meaningless in a broader context but are deeply meaningful to individuals precisely as rituals.4



	Personally speaking, calling back my lessons learnt from the tennis court, from the stage, and from the Buddhist traditions I've written about here, it is that meaning is not a property of the activity but of the relation between the person and the activity. An activity that looks like nothing from the outside can be everything from the inside, and an activity that the world calls important can feel perfectly hollow. If that relational quality survives the automation of labor—and I believe it will, because it is a feature of consciousness, not of economics—then meaning will survive too. Differently and less legibly, perhaps. But it will survive.



	Undoubtedly, to my admission, projecting into a radical case—presumably in the far future, if ever—won't really help guide our current decision-making and policy-making. In that sense, I only hope this will relieve some certain existential dread, if any. But it is very much acknowledged that, especially as the second half of Susskind's project is on, the decisions to make currently are much trickier than the what-ifs.





1 And this is where I pause to go full Marxist, although the forces-of-production / relations-of-production analysis I explored in a previous entry feels increasingly relevant—as the productivity leap has happened (or is happening), the relations of production have not caught up. ↩

2 And yes, being able to maintain this stance presupposes a baseline of material security that is not universally available—to say the very least. ↩

3 As some may suggest, the meaning-making, the working, or even the learning, could be fundamentally rooted in a process of struggling and attempting. ↩

4 Although one could further argue back that it only works when there is this comparative status between the meaningful and meaningless, making the active choice of meaningless meaningful. In that sense, I really don't have an extra defense prepared for now. And we will have to see—though I doubt I will witness it in my lifetime. ↩