AI will change work. The biggest loss won’t be economic

With recent dramatic advances in AI, people are making competing predictions about how it will transform work. Anthropic CEO Dario Amodei claims that AI will disrupt half of white-collar jobs in one to five years. Bill Gates predicts that we won’t need humans “for most things” in 10 years because AI will be able to do even the highly skilled work of doctors and teachers. But the Federal Reserve Bank of Dallas doubts that AI will lead to mass technological unemployment by then. In every previous tech revolution, permanent mass unemployment was avoided because technology enabled the creation of new kinds of jobs.

As a philosopher, I think the questions we should be asking are ethical, not prognostic. How AI will transform work is a decision that is up to us. To make such judgments responsibly, we need to understand the values at stake in prospective AI-driven transformations of work. What should we care about regarding work?

Most discussions of the value of work focus on its economic benefits. If we decide to adopt forms of AI that comprehensively replace human labor, we will need to ensure that people will still have a decent income, but I want to focus on the non-economic benefits of work. What is the role of work in a flourishing life, besides its economic benefits? What makes the activity of working itself meaningful?

When I consider what makes my own work meaningful, I think first about the difference I have made to my students. When I facilitate a lively class discussion, I help them build their skills in articulating their own point of view and learning from others’ perspectives. When I advise my students on research projects, I help them learn how to formulate a significant question, work out how to answer it, address criticism and develop their own voice in communicating their ideas. I help my students gain the confidence to tackle hard questions and take intellectual risks. More concretely, I help them get their dream jobs and teach them skills and perspectives that made them better doctors, lawyers or teachers. I also think about the impact of my research on others: how I have stimulated interest in neglected questions (about work in particular) and developed new approaches to old questions (such as the value of equality) that have transformed how people think about them. I also find meaning in my work because I enjoy autonomy in conducting my research and designing and teaching my courses. And I love to exercise the varied skills I need to do my job well. Every day I am learning new things, developing new ideas, and deepening my understanding of ethical problems as I interact with my students, colleagues and readers.

I bet that if you find your work meaningful and not just lucrative, you also think that your work matters to others, you have discretion over how you do your job, and you enjoy exercising varied skills at work and want to do it well. These three features—purpose, autonomy and mastery—lie at the heart of work that people find fulfilling.^[2]

A Puritan view of work

This ideal of meaningful work goes back to the Protestant work ethic, which was developed by 17th-century Puritan theologians. Today we tend to think of the work ethic as a grind. And it’s not wrong to think of the Puritans as the killjoys of their era. They didn’t indulge in idle pleasures, but they were very interested in something deeper than fun: they wanted us to find meaning in our work and get satisfaction from that. According to them, God commands everyone to promote everyone else’s well-being. We do this through disciplined work that provides goods and services to others. God has instituted a division of labor to enable us to advance everyone’s well-being most efficiently and has called each of us to work in a specialized occupation within it. We therefore have a duty to discover our calling.

These ideas led to the core principles of modern career counseling. In 1627, Puritan priest Robert Sanderson delivered a sermon in which he explained what a genuine calling is, and how any individual can find theirs. A genuine calling must contribute to others’ well-being and the public interest.^[3] To discover which calling is right for you, you need to find an occupation that fits your talents, education and training, so that you have developed the skills to be good at it. Your calling should also be something that you really want to do, in that you enjoy and take pride in exercising the skills needed to do the job well.

Because the individual is the best judge of their personal motivation, each worker is entitled to autonomy in choosing their occupation. Having autonomy over the conduct of work was also implicit in the work ethic ideal, because its model workers, who included most adult male English workers in the 17th century, were self-employed yeoman farmers or artisans. Sanderson further argued that every competent and diligent worker in a legitimate calling is entitled to respect and esteem for the job they do, because they are doing their part in a division of labor that best promotes human flourishing. Workers cooperate in the division of labor like the gears of a clock. Even CEOs need workers to clean the firm’s offices so that they can do their best work. So cleaners, too, should be respected and esteemed for their work. Other Puritan work ethic theologians argued that all workers are entitled to a living wage, safe working conditions and respectful (not tyrannical or demeaning) treatment from their superiors.

Workers today recognize all these features of the work ethic ideal as important components of a meaningful and fulfilling job. According to The American Job Quality Study, American workers want fair living wages, “a safe, respectful environment”, opportunities to develop and exercise their skills, “meaningful control” over how to do their jobs and a voice over their working conditions, productive technology, workload, the pace of work and working hours. Unfortunately, only 40 percent of U.S. workers have jobs that meet these criteria.^[4]

According to the Global Happiness Policy Report, these are the most important non-financial factors that positively affect people’s work satisfaction worldwide:

1. Respectful and supportive relationships with supervisors and co-workers.

2. Interesting work (task variety, which implies the need for varied skills).

3. “Independence”, or autonomy over how one performs one’s job.

4. Opportunities for advancement.

5. A match between the skills demanded by the job and the skills the worker has.

6. Usefulness or purposeful work, meaning that it matters to or helps others.^[5]

Workers regard the first two of these factors as more important to work satisfaction than their pay. If we separate the intrinsic content or nature of the job from the quality of social relationships in which it takes place (which can be changed without changing the job), then we get an ideal of meaningful work: it’s work that promotes others’ wellbeing by exercising skills that the worker takes pleasure and pride in deploying under their own direction. That’s the Puritan ideal. That workers worldwide find it so attractive suggests that these Puritans understood what is important about work.

Even jobs that involve intrinsically meaningful content can become miserable if their extrinsic circumstances are bad. The main non-financial circumstances that make a job miserable are: bad relationships with supervisors and coworkers (harassment, discrimination, contempt, a tyrannical or micromanaging boss); needlessly dangerous or stressful working conditions; scheduling problems and lack of work/life balance (which may be due to too many hours, unpredictable schedules or schedules that interfere with off-duty relationships and activities); and overwork (too many hours, or work at too fast a pace).^[6] These factors also undermine the ability of workers to do their work well, and thereby degrade the meaningfulness of their work.

With this ideal of meaningful work, and an understanding of the extrinsic factors that make workers hate their jobs, let’s consider possible uses of AI at work. I will start with two cautionary tales.

Avoiding moral injury

The first concerns using AI to reproduce for human services what the inventors of the factory system did for manufacturing work during the Industrial Revolution. They divided the labor required to manufacture any item into separate tasks and assigned each worker just one simple task that they had to repeat for the entire working day. Powered machines forced them to work at a relentless pace. Supervisors micromanaged everything they did. Later, in a system invented by Frederick Taylor, time-and-motion studies were undertaken to determine the minimum time needed for each task and thereby maximize workflow.^[7] The factory system dramatically increased the productivity of manufacturing labor. But it deskilled the workers, stripped them of autonomy, and turned their work into tedious drudgery.

AI enables the same techniques to be applied to human services. Break down the service into a finite number of discrete tasks. For example, for disabled people or nursing home patients, such tasks include washing, dressing, feeding and administering medicine. Different tasks would be involved with teaching schoolchildren, coaching athletes and other services.^[8] Whatever tasks may be needed, we could use stopwatches to time the workers on each task. Feed the data into AI, which schedules the tasks assigned to each worker. This minimizes the number of staff needed to provide services for each person, and thereby theoretically maximizes efficiency.

Unlike the parts in an assembly line, however, human beings are not interchangeable. They vary in size, capabilities, needs and desires. They change over time. They have feelings about the services they are getting, and about the workers serving them. Service workers need to be attuned to variations in all these factors. Deciding how to deal with the idiosyncratic challenges arising with each individual requires judgment, resourcefulness, flexibility and time. When algorithmically determined “efficient” staffing levels leave no time to workers for anything but the routine tasks assigned, service quality falters because workers must cut corners somewhere.

This hurts not just the people receiving services, but the service workers themselves. Like factory workers, they suffer stress, exhaustion and boredom from deskilled work which leaves no room for judgment and variation. Because they directly interact with the people receiving their services, such workers are also more likely to suffer moral injury. Moral injury occurs when workers are not allowed to fulfill what they view as their duties to the people they are serving. This degrades the purpose of their work.

This point generalizes to all kinds of complex services involving direct human relationships. Except in highly routine cases, direct human services cannot be reduced to a finite set of discrete, uniform tasks. Providers of complex direct human services must take people’s humanity and individuality seriously. Many aspects of service involve continuous attunement and responsiveness to the service recipient, in ways that can’t be fully articulated.^[9] In medicine, for example, physicians must come to terms not only with disease itself, but with what that disease means to the patient. Treatment can fail if there is too great a distance between the physician’s and the patient’s understanding of their problem. What is needed to bridge that gap is a creative act that can only be validated through interactions that build rapport and trust between caregivers and patients. Authentic emotional communication is critical here, as it is in other human services founded on ongoing social relationships such as teaching, coaching, therapy, legal representation and social work.

Some techno-optimists might think that, with the development of artificial general intelligence (AGI), this kind of creative work would be done better by bots, once the problem of aligning the AI to our values is solved. Then we could trust its advice, and it could deliver answers much more accurately and quickly than even the best professionals. But alignment to our values is not an engineering problem like using tools to ensure that every floor of a building is level. In that case, there is a fixed external standard against which to measure alignment. By contrast, the history of ethical thought and practice reveals perpetual change and development as people confront novel challenges, invent new ideals and reinterpret old ones, experiment with different ways of living, encounter diverse people and arrive at new self-understandings. This process never ends. It involves continuous attunement to each other’s concerns and responsiveness to each other’s claims. It’s an essentially collective process. What would alignment mean in such a perpetually dynamic situation, in which we continuously work out the terms of our relationships to one another? To outsource this ethical thinking to bots would be to outsource our humanity.

The second cautionary tale concerns the prospect of deploying AI on a mass scale to replace all or most human workers, at least in white-collar occupations. As robotics develops, human jobs that involve manual labor, from picking crops to skilled trades, might also be eliminated along with work that is already on its way to full automation, such as manufacturing and transportation. Let’s assume that all the technologically unemployed workers will get a decent income. Some people imagine that if we didn’t have to work, we would finally be truly free, because we’d have endless free time in which we could do whatever we pleased.^[10] Wouldn’t this be a utopia?

I have my doubts. Economist John Maynard Keynes loathed the work ethic, which he thought was merely an ideology that promoted endless drudgery for the sake of greed. In 1930, he wrote an essay in which he predicted that economic growth in one hundred years would enable rich countries to cut back everyone’s work to 15 hours per week. He hoped that in leisure, people would cultivate “the arts of life.” Yet, observing the idle rich of his day, Keynes worried that most had not made good lives for themselves. Most did nothing significant. I would add that they didn’t even raise their own children, having outsourced that work to governesses and boarding schools. Many had lives that were empty, boring and alienated from one another.

The same thing is true today of many citizens of oil-rich countries, who live off oil royalties. Many are frustrated and angry because they have nothing meaningful to do. In recognition of people’s desire for work, Kuwait guarantees a government job to any citizen who wants one. So many young adults have applied for a job that there is not enough work to go around. Many are assigned no tasks.^[11] They resent the lack of meaningful work and spend their time in basement offices playing computer games and watching videos.^[12] The Kuwaiti government is so concerned that it commissioned the World Bank to help it design reforms to afford better work opportunities for its citizens.^[13]

I don’t claim that everyone needs to work to have a meaningful life. Many lead flourishing lives at leisure by pursuing philanthropy, the arts, education, adventure, athletics and so forth. I merely claim that for many people, work is a major and irreplaceable component of their conception of a flourishing life. It’s like raising children: while many people lead fine lives without raising children, many would feel bereft if they were deprived of this opportunity.

AI and human flourishing

Recall the core features of intrinsically meaningful work. First, it serves some larger purpose than self-enrichment. It must promote the flourishing of other people, society at large or other goals that matter to others, such as scientific discovery, artistic creation and ecosystem flourishing. Second, in pursuing these goals, workers exercise various skills that they enjoy and are proud of. Third, workers have autonomy in deploying their skills and a say in shaping the division of labor within their organization, their relationships with co-workers and other work conditions. How can AI be designed to promote these features of meaningful work?

First, AI can expand the goals we are able to achieve or enable us to achieve them more efficiently. This is already happening in many occupations devoted to discovery and invention. Papyrologists are using AI and X-ray tomography to read the text of thousands of ancient burnt scrolls that would disintegrate into ashes if they were unrolled. Pharmacologists are using AI to identify already approved drugs that are likely candidates for treating rare diseases. Doctors are using AI to suggest possible diagnoses of rare diseases for patients who present with unusual symptoms. Mathematicians are using AI to test and suggest different strategies for proving theorems. Software engineers are using AI to offer conceptual advice on coding strategies for solving problems, to improve short blocks of code and explain how they work and to teach them how to solve unfamiliar problems.

In all these cases, AI is complementing or enhancing rather than replacing or degrading workers’ skills, judgment and autonomy. Deciphering the text is only one part of papyrology. Suggested drugs for rare diseases must still be clinically tested on patients. Doctors use their clinical skills in ruling out some AI-suggested diagnoses, counseling patients on treatment options and helping them understand their condition. Mathematicians and experienced software engineers are using their judgment to try out, reject or modify AI suggestions.

A common thread in many of these cases is that what is sought is not simply knowledge of particular facts. It is understanding those facts. Understanding involves drawing connections among facts in humanly intelligible ways that we judge to be significant. For example, mathematicians prize some proofs of a given theorem more than other equally valid proofs, because they do a better job in deepening our understanding, for example, in revealing surprising connections across disparate mathematical domains. Humans cannot be excised from this process, since our objective is that we understand the broader significance of some facts.

Second, AI can be used to remove much of the drudgery from jobs. Besides relieving workers of tedium, this can free them to focus more of their time on more sophisticated tasks. Maintenance workers are deploying AI to automatically complete documentation of their work and using AR smart glasses to display the relevant parts of technical manuals for equipment needing repair, so they can focus on doing the repair itself. Doctors are using AI to summarize patient visits, freeing them from their screens so they can pay more attention to their patients. AI is making quick work of non-profit organization grant applications by taking over the routine parts of drafting, while freeing workers to design and execute more ambitious projects.

Third, AI that upskills workers and enables them to be more self-directed can be used to replace or avoid AI that monitors and micromanages them. Human beings are agents: we have deep needs to act on our own judgment, use our skills and make a meaningful difference in the world. We feel more alive when we are empowered as self-directed agents. When we are empowered this way at work, we are more engaged, more motivated to do our best and don’t need the threat of surveillance and sanctions to be highly productive. People doing work that is usually designed as drudge work under authoritarian management are more productive when they work in teams empowered to redesign and direct their own work. They choose to divide the labor so that every team member works a variety of tasks.^[14]

In many cases, it can be hard to tell whether a particular implementation of AI is good or bad for workers overall. What is drudgery for the most highly skilled workers could be skilled work for others. Using AI to do that drudgery could displace the latter workers and force them to take even worse jobs. Yet in some cases it may be best if robots replaced workers who are currently assigned to certain kinds of tedious, exhausting and inherently dangerous work. How should we decide?

We can get some insight into answering this question by asking what workers would choose if the firm were organized as a workers’ cooperative, in which the workers are also the owners and manage the firm democratically. Workers’ cooperatives make different decisions from most capitalist firms in two important respects. First, they divide the labor more equally, often by way of job rotation. Workers thereby enjoy much greater task variety and acquire more skills than in capitalist firms. Due to democratic decision making, all workers also exercise some higher-level executive judgment. Second, in recessions and other adverse economic conditions, the workers choose to maintain staff levels at lower pay and hours for everyone rather than laying off some of their colleagues.

I suggest that, since people will generally not choose what will harm themselves, if a workers’ collective decided to adopt a particular kind of AI, that likely indicates that this choice would be good for the workers as far as the meaningfulness of their work is concerned. If they chose an AI that completely replaces some tasks currently done by humans, that would likely be acceptable because the workers would share the remaining work equitably among themselves. A similar redivision of labor between AI and workers would likely be comparably good for workers in capitalist firms.

This is not a perfect test. If some techno-optimists are correct, AGI will be so superior to humans that it will outcompete all of us. Workers’ cooperatives that do not fully automate will go bankrupt. Yet if we consider software engineering, an industry where AI is threatening to fully replace highly skilled workers, the difficulties with AI applications that disempower workers are evident. In some companies, software developers are required to maximize the use of AI to generate code. Instead of being authors of the code, the engineers are reduced to copy-editing AI-generated code—fixing bugs and security vulnerabilities and ensuring alignment. Since I’m not a senior software engineer, I am in no position to argue that AI agents can’t eventually replace the application security engineers currently in too short supply to fix all the bugs and security vulnerabilities in millions of lines of AI-generated code. As an ethicist, however, I have already explained why humans can never be taken out of the loop of ensuring that software aligns with our ever-evolving values.

Among the most important of those values is the continued availability of meaningful work for people to do. To be deprived of purpose and agency on a larger scale and significance than our mere choices of consumption and leisure activities is no utopia. For many, it’s a living hell. Today we know this from examining the bored lives of many leisured rich people across the ages. Today’s tech leaders would never choose such a life for themselves. Nor do vastly less rich and powerful people. The overwhelming majority of working-age multimillion-dollar lottery winners continue to work. Even prisoners view being deprived of unpaid laundry work as a punishment.^[15] Any ethical development of workplace AI needs to recognize the importance of meaningful work. There is no better way to ensure that than to empower affected workers themselves with a voice in its design.

And for those readers who are AI engineers, here is a purpose that would make your work highly meaningful: design workplace AI that enhances the purpose, skills and autonomy of other workers, relieves workers of drudgery and hyper-surveillance, gives them a voice over their schedules and promotes constructive workplace relationships.

^[1] I would like to thank Pam Hayes, Lisa Herzog and John Testa for helpful advice on this article.

[15] David Graeber, Bullshit Jobs: A Theory (New York: Simon & Schuster, 2018) 82.