New Article: My Last Five Years of Work
Advances in AI raise the question of human flourishing after employment. Can we be happy without work? History and psychology shed light on the answer.
This article by Avital Balwit was published on Palladium Magazine on May 17, 2024.
I am 25. These next five years might be the last few years that I work. I am not ill, nor am I becoming a stay-at-home mom, nor have I been so financially fortunate to be on the brink of voluntary retirement. I stand at the edge of a technological development that seems likely, should it arrive, to end employment as I know it.
I work at a frontier AI company. With every iteration of our model, I am confronted with something more capable and general than before. At this stage, it can competently generate cogent content on a wide range of topics. It can summarize and analyze texts passably well. As someone who at one point made money as a freelance writer and prided myself on my ability to write large amounts of content quickly, a skill which–like cutting blocks of ice from a frozen pond– is arguably obsolete, I find it hard not to notice these advances. Freelance writing was always an oversubscribed skillset, and the introduction of language models has further intensified competition.
The general reaction to language models among knowledge workers is one of denial. They grasp at the ever diminishing number of places where such models still struggle, rather than noticing the ever-growing range of tasks where they have reached or passed human level. Many will point out that AI systems are not yet writing award-winning books, let alone patenting inventions. But most of us also don’t do these things.
The economically and politically relevant comparison on most tasks is not whether the language model is better than the best human, it is whether they are better than the human who would otherwise do that task. This makes the objection that AI systems are not yet coding long sequences or doing more than fairly basic math on their own a more relevant one. But these systems will continue to improve at all cognitive tasks. The shared goal of the field of artificial intelligence is to create a system that can do anything. I expect us to soon reach it. If I’m right, how should we think about the coming obsolescence of work?
It is worth noting up front that even today, work is far from the only way to participate in society. Nevertheless, it has proven to be the best way to transfer wealth and resources; it provides personal goods like social connection, status, and meaning; and it offers social goods like political stability.
Given this, should we meet the possibility of its loss with sadness, fear, joy, or hope? The overall economic effects of Artificial General Intelligence (AGI) are difficult to forecast, and here I will focus on the question of how people will feel without work—whether they will, or can, be happy. There are obviously other vital questions, like how people will be able to meet their material needs. Many have examined this question, with no final answer yet adopted as official policy for this contingency by any government. I am instead going to do something that may feel like cheating. I will go ahead and assume that people can meet their financial needs through universal basic income or other transfers and will solely concentrate on the question of whether people can and will be happy—or at least as happy as they are now—without work.
The Obsolescence of Knowledge Work
I expect AI to get much better than it is today. Research on AI systems has shown that they predictably improve given better algorithms, more and better quality data, and more computational power. Labs are in the process of further scaling up their clusters—the groupings of computers that the algorithms run on. Machine learning is a young field, with an enormous amount of “low hanging fruit” in terms of discoveries, meaning that researchers continuously find improvements to the algorithms of these AI systems. While an enormous amount of data has already been fed through them, there is still more to be found and it can also be generated by the systems themselves. So, given the “scaling laws”, we can reasonably foresee that these systems will keep getting better—at least until these inputs run out.
And at what rate will they get better? Language models are not, for the most part, continuously improving. They get better in discontinuous jumps. A rough analogy to the current LLM process is that making a new model is like baking a cake. You figure out your data and algorithms—like mixing the batter—and then you pretrain the model, that is, run it on a large number of computers for several months—like putting it in the oven—and then at the end you do some “post training”—like frosting and decorating the cake. Post training can adjust the model in certain ways, often to make it more harmless or honest, or to make it particularly good at some specific skill or use case—but most of what matters for the model’s capabilities, at least right now, is the underlying “cake,” and this can’t be easily adjusted without starting over and baking something new. So when it comes to the rate of progress, when models seem to plateau, you should actually assume that that just means that the next model is in the oven but hasn’t come out yet.
Many expect AI to eventually be able to do every economically useful task. I agree. Given the current trajectory of the technology, I expect AI to first excel at any kind of online work. Essentially anything that a remote worker can do, AI will do better. Copywriting, tax preparation, customer service, and many other tasks are or will soon be heavily automated. I can see the beginnings in areas like software development and contract law. Generally, tasks that involve reading, analyzing, and synthesizing information, and then generating content based on it, seem ripe for replacement by language models.
Obsolescence is unlikely to come for all types of work at the same pace, and even once we have “human-level AI,” the effects will look very different before and after the widespread deployment of robotics. The pace of improvements in robotics lags significantly behind cognitive automation. It is improving as well—but more slowly. Anyone who makes a living through delicate and varied movements guided by situation specific know-how can expect to work for much longer than five more years. Thus, electricians, gardeners, plumbers, jewelry makers, hair stylists, as well as those who repair ironwork or make stained glass might find their handiwork contributing to our society for many more years to come. Regulated industries like medicine or the civil service will have human involvement for longer, but even there, I expect an increasingly small number of human workers who are increasingly supplemented with AI systems working alongside them.
Finally, I expect there to be jobs where humans are preferred to AIs even if the AIs can do the job equally well, or perhaps even if they can do it better. This will apply to jobs where something is gained from the very fact that a human is doing it—likely because it involves the consumer feeling like they have a relationship with the human worker as a human. Jobs that might fall into this category include counselors, doulas, caretakers for the elderly, babysitters, preschool teachers, priests and religious leaders, even sex workers—much has been made of AI girlfriends, but I still expect that a large percentage of buyers of in-person sexual services will have a strong preference for humans. Some have called these jobs “nostalgic jobs.” It is possible, given deflationary pressures, that real wages in these remaining occupations remain enough to keep most people going, at roughly current labor force participation rates.
The Psychology of Employment
As automation rolls out across these industries, how should we expect people to feel? The common assumption about automation, even setting aside the financial effects, is that people will be incredibly unhappy without work. More evidence than not seems to point to unemployment having numerous and diverse negative physical and mental health ramifications, although the size of these effects varies:
“The relationship between unemployment and health has been extensively explored. Alternative identification strategies, data sets, and labor market conditions have produced a vast array of results. These range from very large effects on mortality after job displacement (50–100% increases) as in Sullivan and von Wachter (2009), Eliason and Storrie (2009a) or Browning and Heinesen (2012) to relatively small ones (10–15% increases) as in Rege et al. (2009). The variance of the estimated effects on other—less severe—health and mental health outcomes is even larger, both within and across outcomes. Remarkably, several excellent studies find almost negligible (Kuhn et al. 2009; Black et al. 2015) and even zero effects (Salm 2009; Browning et al. 2006; Roulet A: The Effect of Unemployment on Health: Evidence from Denmark, unpublished).”
One challenge of examining the effects of unemployment on health and wellbeing is the causality. Unhealthy or mentally ill people are more likely to lose or quit their jobs and are more likely to stay unemployed for longer. This means that results showing that the unemployed are mentally and physically sicker shouldn’t be taken to mean that unemployment necessarily makes them sicker.
One study that tried to tackle this looked at the effects of unemployment caused by the collapse of the Spanish construction industry on mental and physical health. This particular study was attempting to disentangle the causality because people who lose their job during a nationwide collapse of an industry will avoid this selection effect: these individuals are no more likely to have mental or physical issues than other members of the population. By looking at large-scale survey responses before and after the crisis, they found that unemployment did appear to increase the likelihood of reporting poorer health, by about 15% in their sample, and an increase in the chance of reporting having a mental disorder, by about one third.
It does seem that, overall, unemployment makes people sadder, sicker, and more anxious. But it isn’t clear if this is an inherent fact of unemployment, or a contingent one. It is difficult to isolate the pure psychological effects of being unemployed, because at present these are confounded with the financial effects—if you lose your job, you have less money—which produce stress that would not exist in the context of, say, universal basic income. It is also confounded with the “shame” aspect of being fired or laid off—of not working when you really feel you should be working—as opposed to the context where essentially all workers have been displaced. Intuitively, it seems there should be more negative psychological effects from losing a job in a way that feels like a personal failing, or that sets one apart from one’s peers, versus losing a job in a “blameless” way, or at the same time and in the same manner as one's peers. At least with this aspect, there are ways to isolate it.
One study that gets around the “shame” confounder of unemployment is “A Forced Vacation? The Stress of Being Temporarily Laid Off During a Pandemic” by Scott Schieman, Quan Mai, and Ryu Won Kang. This study looked at Canadian workers who were temporarily laid off several months into the COVID-19 pandemic. They first assumed that such a disruption would increase psychological distress, but instead found that the self-reported wellbeing was more in line with the "forced vacation hypothesis," suggesting that temporarily laid-off workers might initially experience lower distress due to the unique circumstances of the pandemic.
Using survey data of unemployed and employed workers, and interviews with a subset of laid-off workers, they found that individuals who were temporarily laid off in April 2020 reported lower levels of distress compared to their peers who remained employed. By May 2020, the distress gap observed in April had vanished, indicating that being temporarily laid off was not associated with higher distress during these months. The interviews revealed that many workers viewed being left without work as a "forced vacation," appreciating the break from work-related stress and valuing the time for self-care and family. The widespread nature of layoffs normalized the experience, reducing personal blame and fostering a sense of shared experience. Financial strain was mitigated by government support, personal savings, and reduced spending, which buffered against potential distress.
Stress ticked up later in the pandemic for these workers, which could indicate that long stretches of unemployment are psychologically different than short ones, but could also indicate that financial stress increased as people ate through their savings, a factor that would not apply under genuine universal basic income. The study suggests that the context and available support systems can significantly alter the psychological outcomes of unemployment—which seems promising for AGI-induced unemployment.
Plant closures, like the pandemic, might provide a “shame-free” way of entering unemployment. The study “Effects of Layoffs and Plant Closings on Depression Among Older Workers” by Jennie E. Brand, Becca R. Levy, and William T. Gallo looked at several hundred men and women who experienced layoffs as compared to plant closures, using longitudinal data from the Health and Retirement Study. They found that reported depression increased after both forms of displacement for men and women. But they found that depression was higher for men after being laid off than for a plant closure, and the reverse for women. The male side of this was hypothesized to be caused by the fact that being laid off felt more like a personal failure that singled the men out, whereas a plant closure was not related to their own worth and was experienced equally by their community. They didn’t have a great explanation for the women’s results, which leaves something to be desired.
From the studies on plant closures and pandemic layoffs, it seems that shame plays a role in making people unhappy after unemployment, which implies that they might be happier in full automation-induced unemployment, since it would be near-universal and not signify any personal failing.
A final piece that reveals a societal-psychological aspect to how much work is deemed necessary is that the amount has changed over time! The number of hours that people have worked has declined over the past 150 years. Work hours tend to decline as a country gets richer. It seems odd to assume that the current accepted amount of work of roughly 40 hours a week is the optimal amount. The 8-hour work day, weekends, time off—hard-fought and won by the labor movement!—seem to have been triumphs for human health and well-being. Why should we assume that stopping here is right? Why should we assume that less work was better in the past, but less work now would be worse?
Removing the shame that accompanies unemployment by removing the sense that one ought to be working seems one way to make people happier during unemployment. Another is what they do with their free time. Regardless of how one enters unemployment, one still confronts empty and often unstructured time. Is this, in and of itself, bad for people?
One paper, titled “Having Too Little or Too Much Time Is Linked to Lower Subjective Well-Being” by Marissa A. Sharif, Cassie Mogilner, and Hal E. Hershfield tried to explore whether it was possible to have “too much” leisure time. They hypothesized that discretionary time had an inverted U-shaped relationship with subjective well-being—meaning at low levels of discretionary time, increasing it increased well-being, while at high levels, increasing it decreased well-being. Using survey data (and a somewhat less reliable seeming “imagination activity") it found that participants reported that no amount of “social” or “productive” discretionary time would be too much, though “solo” and “unproductive” discretionary time could be.
The paper concluded that it is possible to have too little discretionary time, but also possible to have too much, and that moderate amounts of discretionary time seemed best for subjective well-being. More time could be better, or at least not meaningfully worse, provided it was spent on “social” or “productive” leisure activities. This suggests that how people fare psychologically with their post-AGI unemployment will depend heavily on how they use their time, not how much of it there is—there is a path to positive well-being, if people spend time exercising, playing with their kids, spending time with friends, and so on.
The End of the Protestant Work Ethic
Can we envision people free from the psychological burdens of shame and duty, using their time well, and being happy while not working? Yes, we certainly can, because many of us already, at least occasionally, are: people on weekends, on vacation, during summers when they are students. Perhaps these groups reflect the “optimal amount of free time” argument above—that some time off can be good, while too much off can be bad. Or they reflect the shame piece—people in these situations do not feel ashamed to not be working. Or perhaps some combination.
We also tend to view retirement positively. If people are in fact happier after retirement, it might even suggest that not working under certain conditions is actually beneficial for well-being. A study published in Frontiers in Public Health looked at men in urban China and found that they were happier after retirement. Similarly, a study found English men reported better mental health and better subjective physical health after retirement.
Other studies find contradictory results, that retirement is actually associated with worse outcomes: the paper "The Effects of Retirement on Physical and Mental Health Outcomes" by Dhaval Dave, Inas Rashad, and Jasmina Spasojevic used data from the Health and Retirement Study covering 1992-2005 and found that retirement significantly increases difficulties in mobility and daily activities (5-16% increase), illness conditions (5-6% increase), and negatively impacts mental health (6-9% decline). These effects are mediated through lifestyle changes, such as reduced physical activity and social interactions. Mitigating factors include being married, maintaining physical activity, and part-time work post-retirement.
In contrast, “Effect of retirement on major chronic conditions and fatigue: French GAZEL occupational cohort study,” used a sample of about fifteen thousand French individuals with repeated measurements from seven years before to seven years after retirement. The study found that “retirement did not change the risk of major chronic diseases but was associated with a substantial reduction in mental and physical fatigue and depressive symptoms, particularly among people with chronic diseases.”
Overall, there is a fair amount of support for a vaguely U-shaped trajectory of happiness based on age, meaning that older people tend to self-report as happier, particularly between 60 and 75, though gender and income also affect the exact shape. This means that older people—those who, in developed countries, tend to be retired—self-report as happier, on average, than working people. Automation-induced unemployment could feel like retiring depending on how total it is. If essentially no one is working, and no one feels like they should be working, it might be more akin to retirement, in that it would lack the shameful element of feeling set apart from one’s peers.
Women provide another view on whether formal work is good for happiness. Women are, for the most part, relatively recent entrants to the formal labor market. In the U.S., 18% of women were in the formal labor force in 1890. In 2016, 57% were. Has labor force participation made them happier? By some accounts: no. A paper that looked at subjective well-being for U.S. women from the General Social Survey between the 1970s and 2000s—a time when labor force participation was climbing—found both relative and absolute declines in female happiness.
Of course, saying that women were not made happier by joining the workforce does not guarantee that leaving it would not make them unhappy. But I think women’s work and AI is a relatively optimistic story. Women have been able to automate unpleasant tasks via technological advances, while the more meaningful aspects of their work seem less likely to be automated away.
When not participating in the formal labor market, women overwhelmingly fill their time with childcare and housework. The time needed to do housework has declined over time due to tools like washing machines, dryers, and dishwashers. These tools might serve as early analogous examples of the future effects of AI: reducing unwanted and burdensome work to free up time for other tasks deemed more necessary or enjoyable. Roombas are a comic and clear continuation of this trend. It seems likely that more advanced AI systems will also soon enter the home to fold clothes, cook meals, and the like.
That said, it seems less likely that AIs will so thoroughly automate childcare and child-rearing because this “work” is so much more about the relationship between the parties involved. Like therapy, childcare and teaching seems likely to be one of the forms of work where a preference for a human worker will persist the longest.
Another view on unemployment comes from the aristocrats of the past. In the early modern era, landed gentry and similar were essentially unemployed. Perhaps they did some minor administration of their tenants, some dabbled in politics or were dragged into military projects, but compared to most formal workers they seem to have worked relatively few hours. They filled the remainder of their time with intricate social rituals like balls and parties, hobbies like hunting, studying literature, and philosophy, producing and consuming art, writing letters, and spending time with friends and family. We don’t have much real well-being survey data from this group, but, hedonically, they seem to have been fine. Perhaps they suffered from some ennui, but if we were informed that the great mass of humanity was going to enter their position, I don’t think people would be particularly worried.
If we do manage to obtain a world where people have their material needs met but also have no need to work, aristocrats could be a relevant comparison. Doubly so because they were part of a social world where their peers were similarly unemployed. I sometimes wonder if there is some implicit classism in people’s worries about unemployment: the rich will know how to use their time well, but the poor will need to be kept busy. I suppose we’ll soon find out.
We also have to consider what effects unemployment will have when it encompasses entire civilizations. The Culture, the people of Iain Banks’ eponymously named science-fiction series, face exactly this conundrum. They are a completely post-scarcity society. Money is viewed as crude and irrelevant for allocating resources. Living space, raw materials, and energy are produced in abundance for its citizens: “the capacity of its means of production ubiquitously and comprehensively exceeded every reasonable… demand its not unimaginative citizens could make.” Yet, the Culture has at least one need that this abundance cannot satisfy:
“The only desire the Culture could not satisfy from within itself was one common to both the descendants of its original human stock and the machines they had (at however great a remove) brought into being: the urge not to feel useless. The Culture's sole justification for the relatively unworried, hedonistic life its population enjoyed was its good works; the secular evangelism of the Contact Section, not simply finding, cataloguing, investigating and analysing other, less advanced civilisations but - where the circumstances appeared to Contact to justify so doing - actually interfering (overtly or covertly) in the historical processes of those other cultures.”
This society is fully able to meet the material needs of its people—but it still must reckon with their “spiritual” needs, of feeling like they have a reason for being. In this case, one that they satisfy via studying and interfering with other galactic species. One does have to wonder what happens when this work is done—all the species found and uplifted. And for us, will we too turn to the stars? It isn’t crazy. I suppose the question is really: will we turn to the stars even when the machines we make are able to do that better than we can, just as they can do everything else better too?
Do you do anything that you are notably worse at than other people just for the sheer value of your doing it, either the joy or the meaning? I do. I dance ballet although I know that being a prima ballerina now that I am in my mid-twenties is long behind me—but moving my body like that brings me joy. We can think of this as the hedonic reason for doing an activity others can do better.
Although a trained therapist might be able to counsel my friends or family through their troubles better, I still do it, because there is value in me being the one to do so. We can think of this as the relational reason for doing something others can do better. I write because sometimes I enjoy it, and sometimes I think it betters me. I know others do so better, but I don’t care—at least not all the time. The reasons for this are part hedonic and part virtue or morality.
A renowned AI researcher once told me that he is practicing for post-AGI by taking up activities that he is not particularly good at: jiu-jitsu surfing, and so on, and savoring the doing even without excellence. This is how we can prepare for our future where we will have to do things from joy rather than need, where we will no longer be the best at them, but will still have to choose how to fill our days.
We will also not need to choose how to fill our time alone: in the context where we are all out of work—and where this is one of our main worries—it means we built relatively-aligned artificial general intelligence. For the same reasons I expect us to reach AGI, I expect it to progress beyond this point, to where we have “superhuman” systems. For the same reason these systems will be helpful with anything, we should expect that these systems will be able to help with the problems that they create. If we believe there are solutions to unhappiness or a feeling of a loss of purpose, and that these solutions can be found with intelligence, then we should expect these systems to be able to help us find them. This may sound self-serving or wishful, and will doubtless leave many unsatisfied. But, I believe that if we really think these systems will be able to replace us, there is no reason to believe they will not also be able to help us in our search for meaning.
Avital Balwit lives in San Francisco and works as Chief of Staff to the CEO at Anthropic. This piece was written entirely in her personal capacity and does not reflect the views of Anthropic. You can find more of her writing here or follow her at @AvitalBalwit.