Deep Utopia: Life and Meaning in a Solved World

MONDAY

Hot springs postponed

Tessius: Hey, look at this poster. Nick Bostrom is giving a lecture series, here in the Enron Auditorium, on The Problem of Utopia.

Firafix: Bostrom—is he still alive? He must be as old as the hills.

Tessius: It’s all those green vegetable elixirs he used to make.

Firafix: They worked?

Tessius: Not at all, but they became very popular for a while. That’s how he made his money, you see—the recipe book. Then he could afford the anti-aging therapies as they came along.

Tessius: It just started ten minutes ago. Shall we go in?

Firafix: Sure, why not.

Kelvin: We can take the baths after dinner instead. They’re open late.

Bostrom: —leukemia. It’s really important to find a cure or at least some way to alleviate her suffering. On a larger scale, we have extreme poverty, deprivation, malnutrition, soul-crushing physical and mental disorders, family-destroying traffic accidents, alcoholism, oppression, the killing and maiming of civilians in war zones… There are, at present, more than enough problems to provide meaningful challenges for even the most resourceful and enterprising among us.

Tessius [whispers]: The old doom-monger is in great form. I’m feeling worse already!

Firafix: Shhh.

Bostrom: Perhaps these humanitarian challenges feel meaningful only to those who care enough about others to have a sincere desire to help. But even pebble-hearted egoists are well catered for in today’s world, with a rich buffet of negative circumstances which they are motivated to ameliorate or prevent from getting worse. One person might be struggling with excess body weight, another with getting a job, a third with social isolation, a fourth with a difficult relationship. Rarely do we hear people complain, "The only problem I have is that I have no problems—life, you know, is just too perfect, and it really bugs me!".

In short, we appear in no imminent danger of running out of woe. As far as the eye can see, there is an abundance of actual and potential sorrow to keep the worry-mill a-churning, and to provide altruist and egoist alike with bountiful opportunities for worthwhile striving.

Nevertheless, in these lectures I want to talk about the problem of utopia: the problem we will face after we have solved all the other problems.

This may not seem like the most pressing priority in our current situation… There are, we must concede, other causes and tasks with most legitimate claims on our attention. Still, I don’t think it would be unbefitting for our civilization to at least cast a glance at what lies ahead if things were to go well: to consider, that is, where we eventually end up if we continue along the present path and completely succeed in what we are in the process of trying to accomplish…

The telos of technology, we might say, is to allow us to accomplish more with less effort. If we extrapolate this internal directionality to its logical terminus, we arrive at a condition in which we can accomplish everything with no effort. Over the millennia, our species has meandered a fair distance toward this destination already. Soon the bullet train of machine superintelligence (have we not already heard the conductor’s whistle?) could whisk us the rest of the way.

And what would become of us then?

What would give our lives meaning and purpose in a solved world?

What would we do all day?

These questions have timeless intellectual interest. The concept of deep utopia can serve as a kind of philosophical particle accelerator, in which extreme conditions are created that allow us to study the elementary constituents of our values. But the questions may also come to have immense practical importance, as the telos of technology is actually reached or closely enough approached—very possibly within the lifetime of many of you here in the audience, in my estimation.

Tessius: Shall we sit down?

Kelvin: There are some seats over there.

Firafix: Yes, I want to hear this. I’ll stand here.

Bostrom: In any case, the problem of utopia is in the water. Can we not sense it—a certain half-embarrassed latent unease? A doubt lurking in the depths beneath us? A faint shadow sweeping across our conception of what it’s all for?

Argumentum ad opulentium

And sometimes this concern breaches the surface of awareness, and we see a fin approaching… For example, Bill Gates wrote:

"It is true that as artificial intelligence gets more powerful, we need to ensure that it serves humanity and not the other way around. But this is an engineering problem … I am more interested in what you might call the purpose problem. . . . if we solved big problems like hunger and disease, and the world kept getting more peaceful: What purpose would humans have then? What challenges would we be inspired to solve?"1

And Elon Musk, in an interview with CNBC:

"How do we find meaning in life if the AI can do your job better than you can? I mean if I think about it too hard, it can frankly be dispiriting and demotivating. Because—I’ve put a lot of blood, sweat, and tears into building the companies, and then I’m like ‘should I be doing this?’. Because if I’m sacrificing time with friends and family that I would prefer, but then ultimately the AI can do all these things. Does that make sense? I don’t know. To some extent, I have to have deliberate suspension of disbelief in order to remain motivated."2

Perhaps there is a sense in which worrying about purpose is a luxury problem? If so, we might expect utopian prosperity to increase its prevalence. But in any case, as we shall see, the issue runs far deeper than anything to do with a mere surfeit of money and material possessions.

Some of my friends like for there to be a model of impact—a story of why, out of all the things that one could be working on, the thing one proposes to do would be the most impactful and beneficial. They seek the highest expected utility.

Were I to attempt such a story for our present proceedings, it might go as follows. Our civilization looks to be approaching a critical juncture, given the impending development of superintelligence. This means that at some point, somebody, or all of us, might be confronted with choices about what kind of future we want—where the options include very different trajectories, some of which would take us to radically unfamiliar places. These choices could be highly consequential. Yet perhaps some of the choices must be made under time pressure, because the world refuses to wait, or because we ourselves are going crazier by the week, or because delaying would mean getting preempted by more decisive actors, or because we don’t want to stop moving for fear that we might then never start moving again.3 Or perhaps there is no discrete time when these choices get made, and instead they are and will be made incrementally over time, but in such a way that earlier partial choices limit the range of later feasible outcomes. Either way, there could be value in getting pointed in a positive direction sooner rather than later. And if there is an actual distinct period of pivotal deliberation, it would be useful to have some suitable preparatory material for that—you know, to equip the deliberators with some relevant concepts and ideas, and help them get into a good frame of mind.

Make of that impact story what you will. Another possible explanation for why I’m doing these lectures is that I agreed to it a long time ago, in a moment of weakness.

Let me say what this lecture series is not. It is not an attempt to make a case for something. It is instead an exploration. When exploring a topic as deep and difficult as the one before us, one wants to bring into view multiple considerations, pursue various lines of thought, place one’s hands on competing evaluative conceptions—allowing the tug of each thought and each inclination to be experienced as keenly and as sympathetically as possible. One does not want to prematurely dismiss a natural perspective, even one that is ultimately to be turned away from. For the value of one’s opinions, in a matter like this, is a function of how generously one has allowed the alternatives to play with one’s soul.

Walls of sausages

Let us consider first the simplest kind of utopia: that of sheer material abundance.

This utopian conception is exemplified by the myth of Cockaigne, or The Land of Plenty. It was an important part of the medieval imaginary, and found frequent expression in popular art and writing as well as in the oral tradition:

"No work is done the whole day long,

By anyone old, young, weak, or strong.

There no one suffers shortages;

The walls are made of sausages."4

Cockaigne is essentially a medieval peasant’s daydream. In the land of Cockaigne, there is no backbreaking labor under scorching sun or nipping norther. No stale bread, no deprivation. Instead, we are told, cooked fish jump out of the water to land at one’s feet; and roasted pigs walk around with knives in their backs, ready for carving; and cheeses rain from the sky. Rivers of wine flow through the land. It is perpetually spring, the weather is beautiful and mild. You make money while you sleep. And sexual taboos have been loosened—we find descriptions of nuns turned upside down with their bottoms showing. Disease and aging are no more. There is continual feasting, with a great deal of dancing and music-playing, and lots of time for resting and relaxing too.5

Similar fantasies are found in many other traditional societies. For example, in classical antiquity, Hesiod wrote of the happy inhabitants of an imagined earlier Golden Age:

"And they lived like gods, not a care in their hearts,

Nothing to do with hard work or grief,

And miserable old age didn’t exist for them.

From fingers to toes they never grew old,

And the good times rolled. And when they died

It was like sleep just ravelled them up.

They had everything good. The land bore them fruit

All of its own, and plenty of it too. Cheerful folk…"6

In many respects, we are now living in the Golden Age, or in Cockaigne—or in Avalon, The Happy Hunting Ground, The Land of the Ancestors, The Island of the Blessed, Peach Blossom Spring, Big Rock Candy Mountain. We here of course excludes those hundreds of millions of humans who still live in abject poverty, along with the great majority of farmed and wild animals. But if we use the term we to refer to the people in this room (we the happy few), then it seems fair to say that with our overstocked fridges and around-the-clock delivery services we have in fact achieved a pretty good approximation of roasted pigs wandering the streets and cooked fishes jumping to our feet. We have also achieved everlasting spring—at least inside our air-conditioned buildings and transportation vehicles. The fountain of youth remains to be located, but disease has been considerably reduced and lifespans extended. Furthermore, I have it on good authority that if somebody is intent on looking at female bottoms, including those of apparent nuns, an online search shall not disappoint.

We do, however, still put in a significant amount of work. Our jobs are generally less grueling than those of medieval peasants; but it is nevertheless a bit surprising that we continue to work as many hours as we do.

Keynes’s prediction

This utopian vision of Cockaigne anticipates the conception of progress that we find in modern economics. The latter couches the ideal in a more abstract lexicon of productivity, income, and consumption, rather than sausage-walls. But the core idea of felicity through abundance remains the same.

So this may be a good place to start our exploration—by reviewing the land and its constraints through the binoculars of economics and evolution; and, tomorrow, we will also look at some ultimate technological limitations. But I want to say that, provided you stay for the entire lecture series, you will find that the tenor of our inquiry will gradually shift. We will descend from the external perspectives and cold abstractions of the dismal sciences, down into the valleys where we will get a more humanistic and internal view of the issues of deep redundancy. And it will shift again as we then begin drilling down into the philosophical mantle, in an effort to reach the core—the core of our values, the heart of the problem of utopia.

So hang in there!

I could perhaps say more about what exactly I mean by the problem of utopia and how I plan to approach it. But I think it’s better we just jump right in, and we can sort out any definitional or argumentative-structural issues as they arise.

The renowned economist John Maynard Keynes considered the goal of material abundance in his widely influential essay, Economic Possibilities for Our Grandchildren.7 Published in 1930, the essay argues that humankind is on its way to solving its economic problem. Keynes predicted that by 2030, accumulated savings and technical progress would increase productivity relative to his own time between fourfold and eightfold.8 Such a dramatic rise in productivity would make it possible to satisfy human needs with far less effort; and, as a consequence, the average working week would decrease to 15 hours. This prospect worried Keynes. He feared that the surfeit of leisure would cause a kind of collective nervous breakdown, as people would go stir-crazy not knowing what to do with all their spare time.9

As we approach 2030, the first part of Keynes’s prediction is on track to vindication. Productivity has increased by more than fivefold since 1930, and GDP per capita by more than sevenfold.10 We thus make much more per hour of work than our great-grandparents did.

The second part of Keynes’s prediction, on the other hand, would appear to be about to miss its mark, if trends are extrapolated. While it is true that working hours have declined substantially over the past ninety-plus years, we are nowhere near the 15-hour work week that Keynes expected. Since 1930, the typical work week has been reduced by about a quarter, to roughly 36 hours.11 The proportion of our lives spent working has seen a somewhat sharper drop: we join the workforce later, live longer after retirement, and take more leave.12 And our work is on average less strenuous. For the most part, however, we have used our increased productivity for consumption rather than leisure. Greed has triumphed over Sloth.

But perhaps Keynes only got the timing wrong?13 A revived Keynes—we can picture him emerging from a cryonics dewar, his hat and mustache covered in frost—might argue that we only need to wait for productivity to rise a bit more to see his prophesied 15-hour work week become a reality. If the historic trend were to continue, we’d see another 4- to 8-fold productivity increase in the next 100 years, and a 16- to 64-fold increase by the year 2230. In such a world, would people still choose to spend a large fraction of their waking life working?

Consider two possible reasons for working:

To earn income

Because working is an intrinsically valued activity

We will return to (2) in later lectures, so let’s set that aside for now. But if productivity grew another 8-fold, or even another 64-fold, would we then see Keynes’s vision of the leisure society come true?

Maybe, or maybe not. There are reasons to be skeptical. In particular, new consumption goods may be invented that cost a lot, or we may undertake very costly social projects. We may also find ourselves compelled to spend more on arbitrarily expensive status symbols to maintain or enhance our relative standing in a zero-sum rat race.

These sources of motivation could continue to operate even at very high income levels. Let us examine each in turn.

New needs and niceties

First, there may be new consumption goods. It is conceivable that there could be an unending series of ever more exquisite—and ever more expensive— market goods that enhance leisure; so that no matter how high your hourly salary, it is worth allocating a third or more of your waking hours to working, for the sake of being able to enjoy the remainder at a higher level of consumption. This was the line taken by Richard Posner, the eminent American legal scholar; we’ll come back to him later.

This view, however, is highly implausible in today’s world, where money has steeply diminishing marginal utility, and where many of the best things in life are indeed free or very cheap. Boosting your annual income from $1,000 to $2,000 is a big deal. Raising it from $1,000,000 to $1,001,000—or even, I should think, to $2,000,000—is barely noticeable.

But: this could change. Technological progress might create new ways of converting money into either quality or quantity of life, ways that don’t have the same steeply diminishing returns that we experience today.

For example, suppose there were a series of progressively more expensive medical treatments that each added some interval of healthy life-expectancy, or that made somebody smarter or more physically attractive. For one million dollars, you can live five extra years in perfect health; triple that, and you can add a further five healthy years. Spend a bit more, and make yourself immune to cancer, or get an intelligence enhancement for yourself or one of your children, or improve your looks from a seven to a ten. Under these conditions—which could plausibly be brought about by technological advances—there could remain strong incentives to continue to work long hours, even at very high levels of income.14

So the future rich may have far more appealing ways to spend their earnings than by filling up their houses, docks, garages, wrists and necks with increasing amounts of today’s rather pathetic luxury goods. We must therefore not unquestioningly assume that money won’t matter beyond some given level. The biomedical enhancements I just mentioned are one example of a kind of good that could continue to provide value at high levels of expenditure. And if we imagine—as I tend to do—a future that is mostly populated by digital minds, then the convertibility of wealth into well-being becomes even clearer. Digital minds, be they AIs or uploads, need computation. More computation means longer life, faster thinking, and potentially deeper and more expansive conscious experiences. More computation also means more copies, digital children, and offshoots of all kinds, should such be desired.

The returns curve of infrastructure spending for a digital mind depends on what it is that one is aiming to achieve. Beyond a certain speed of computation, the marginal cost of accelerating a mind’s implementation further may rise sharply or hit a hard limit. On the other hand, some algorithms parallelize well, and if they instantiate something that is valued, the returns to computation could be close to linear. Certainly, if you’re happy simply making copies of yourself, you need not see diminishing returns even at very high levels of expenditure.

Social projects

Second, if we look beyond the sphere of selfish indulgences, we see many additional opportunities to convert huge amounts of resources into valuable outcomes before hitting discouragingly diminishing returns. For example, you might want to build a veterinary system for animals that are sick or injured in the wild. [Applause.] People who care about such ambitious projects could have reason to continue working long hours even as their productivity and hourly salary soar to stratospheric heights, because they could keep scaling up their impact. Until there is a clinic on every hill and in every dale, in each bush and each briar, there is an underserved population.

In fact, the altruistic reason for working additional hours may theoretically get stronger the higher a person’s wages. More additional wild animal hospital rooms could be funded with an extra hour of work if your hourly rate is a thousand dollars than if you’re making minimum wage.

I say may theoretically get stronger, because as the level of wealth in society increases, it is possible that the lowest-hanging or juiciest fruits in the altruistic opportunity tree get depleted. However, the tree is big and it keeps growing new fruit: so as long as you can keep making money, you can most likely keep doing good. This is readily seen if we consider altruistic reasons not only for removing negatives from the world but also for adding positives, such as by bringing new happy people into existence. You could always create more, and the number scales linearly with resources.

By the way, are there any questions so far? Feel free to interrupt at any time if something isn’t clear. —Yes, you there in the aisle, with the buttons?

Student: Are you saying that, like, we should have as many kids as possible? Wouldn’t that be selfish?

Bostrom: No, I’m not at this point expressing any moral view. I’m discussing some possible motivations that could conceivably drive some people to continue to work long hours for money even if they could cover all their ordinary needs by working just one or two hours a week. One such possible motive is altruism: make more so you can give more to those in need. Okay, but then what happens if society gets sufficiently rich and utopian that there are no more people in need? I was pointing out that even in that case, some people might be motivated to continue to earn so that they could create more people. No matter how affluent everyone is—indeed especially if everyone is very wealthy—you could, in principle, create additional happiness by bringing additional happy people into existence. There certainly are folk who think that would be a good thing, such as total utilitarians, and who could thus remain motivated. There are others, of course, who have no desire to maximize any such measure of aggregate utility. This not being a course on population ethics, we don’t need to concern ourselves here with what arguments or justifications there might be for these different views. Though I may note, for the record, that I’m not a total utilitarian, or indeed any kind of utilitarian, although I’m often mistaken for one, perhaps because some of my work has analyzed the implications of such aggregative consequentialist assumptions. (My actual views are complicated and uncertain and pluralistic-leaning, and not yet properly developed.) Does that help?

Another student: But what about global warming?

Tessius [whispers]: Some are especially easy to automate.

Bostrom: Well, I think we must make some postulations in order to focus our investigations on the central question that we will be exploring in this lecture series. This means that we will be bracketing a bunch of practical matters entirely, in a bid to get to the philosophical crux. More specifically, we are conducting a thought experiment in which we assume that technological as well as political difficulties are somehow overcome, so that we can focus on the problem of what I call deep utopia. I was planning to talk about the technological boundary conditions tomorrow, so hopefully things will become a bit clearer then.

So, as I was saying, you could always create more people, especially of the digital sort.15 The number of digital minds you could create is proportional to the amount of computational resources you could deploy, which we can assume is proportional to the amount of money you have to spend.

Of course, this type of scalable altruistic motivation is reserved for the moral elite. If you don’t care about bringing more joyful beings into existence, and you don’t have enough universal concern for the welfare and suffering of other sentient beings that already exist, and you don’t have some other open-endedly ambitious unegoistic project that you can feel passionate about, then you may not drink from this fountain and you’d have to seek other ways to quench your thirst for purpose. —Let’s take one more question.

Yet another student: What do you mean by digital minds?

Bostrom: A mind implemented on a computer. Could for example be an upload of a human or animal mind, or an AI of a design and sophistication that makes it a moral patient, i.e. one whose welfare or interests matter for their own sake. I think there’d be a prima facie case for this in the case of a conscious digital mind, though I don’t think consciousness is necessary for moral status. For the purposes of the present discussion, probably nothing essential hinges on this point.

Okay, let’s press on. We have a lot of stuff to get through.

The desire for more

I mentioned a third reason why we might continue to work hard even at very high income levels: namely, that our appetites may be relative in a way that makes them collectively insatiable.

Suppose that we desire that we have more than others. We might desire this either because we value relative standing as a final good; or, alternatively, because we hope to derive advantages from our elevated standing—such as the perks attendant on having high social status, or the security one might hope to attain by being better resourced than one’s adversaries. Such relative desires could then provide an inexhaustible source of motivation. Even if our income rises to astronomical levels, even if we have swimming pools full of cash, we still need more: for only thus can we maintain our relative standing in scenarios where the income of our rivals grows commensurately.16

Notice, by the way, that insofar as we crave position—whether for its own sake or as a means to other goods—we could all stand to benefit from coordinating to reduce our efforts. We could create public holidays, legislate an 8-hour work day, or a 4-hour work day. We could impose steeply progressive taxes on labor income. In principle, such measures could preserve the rankings of everybody involved and achieve the same relative outcomes at a reduced price of sweat and toil.17

But failing such coordination, we may continue to work hard, in order to keep up with all the other people who continue to work hard; and we’re stuck in a billionaire’s rat race. You just cannot afford to slack off, lest your net worth remain stuck in the ten digits while your neighbor’s ascends into the eleven…

Imagine standing on the deck of your megayacht, SV Sufficiens. You are gliding across the ocean, making good headway with your date, who is suitably impressed. You inch closer in preparation for a kiss, and… next moment you’re bobbing ignominiously up and down in the wake of your colleague’s gigayacht, NS Excelsior, as it roars past you. There he is, at the aft of his far grander vessel, grinning patronizingly down at you and waving his stupid sea captain’s hat! The moment is quite ruined.

It is also possible to have a desire for improvement per se: to desire that tomorrow we have more than we have today. This might sound like a strange thing to want. But it reflects an important property of the human affective system—the fact that our hedonic response mechanism acclimates to gains. We begin taking our new acquisitions for granted, and the initial thrill wears off. Imagine how elated you would be now if this kind of habituation didn’t happen: if the joy you felt when you got your first toy truck remained undiminished to this day, and all subsequent joys—your first pair of skis, your first bicycle, your first kiss, your first promotion—kept stacking on top of each other. You’d be over the moon!

Well, our limbic system (that old curmudgeon) puts paid to that. The hedonic treadmill continuously retreats under our feet, making us keep running while preventing us from ever getting to any fundamentally cheerier place.

But how does this provide an incentive to work in a world of radical economic abundance? We may crave improvement, either for its own sake or as a means of getting a jolt of reward, but it still seems like this craving depends on there being other desires to define what counts as an improvement. I mean, if you didn’t want the toy truck in the first place, then obtaining it wouldn’t be an improvement and wouldn’t bring you joy. So we need some type of underlying good that you can keep accumulating and still benefit from getting more of. If there is such a good—maybe the biomedical enhancements or the altruistic initiatives I spoke of earlier—then the desire for things to be improving can serve as an amplifying factor, giving us even stronger reason to continue working beyond that given by our desire for the base-level good itself.

So much for the desire for improvement per se. But let’s return to the desire to have more than other people—more money, or more exclusive status symbols. This desire, it seems, can stand on its own without presupposing that there is some other more basic desire that defines an unbounded betterness metric. (Strictly speaking, if what we are after in wanting to have more than others is social status, then the construction might require the existence of additional desires in the sense that we especially want to have more than other people of something that they also covet: but the item in question is fundamentally arbitrary and need not be desired by anybody for its own sake apart from the role it happens to play as a focus of such social contestation—it could be an NFT or civet coffee or something else that hardly anybody would want unless others wanted it too.)

The desire for relative standing is therefore a promising source of motivation that could spur work and exertion even in a context where man’s economic problem has been solved. Provided only that other people’s incomes keep rising roughly in tandem with our own, vanity could prevent us from slacking off no matter how rich we get.

The desire for relative standing has another feature that suits it to be a motivator in the age of abundance. Ranking is, to a significant extent, ordinal. That is to say, what matters is who has more than whom, not necessarily how much more. So if your rival’s yacht is 10 meters long, the important thing might be that your yacht be at least 11 meters long. Similarly, if his is 100 meters, it is paramount that yours be longer—but it doesn’t have to be 10% longer for you to maintain the advantage; 101 meters is enough. This is convenient because it means that—to the extent that we covet this kind of ordinal social rank—the objective gains we make don’t have to be proportional to our cumulative previous gains in order to remain significant. Small incremental gains can continue to be very attractive, so long as they have the potential to shift our rank in the relevant comparison group.18

Perfect or imperfect automation

Might we not work just because we enjoy working? Well, I won’t count an activity as work if we do it simply because we enjoy it. But what if we enjoy it because it is useful? Well, then there needs to be some other reason aside from it being enjoyable, such as one of those we discussed. To repeat, the three types of consumption desire which could plausibly continue to motivate people to work even at very high levels of productivity and income were: to acquire novel goods and services that provide some noncomparative personal benefit; to accomplish ambitious social projects; and to acquire positional goods that help one gain status.

Theoretically, these could stave off the arrival of the leisure society indefinitely—ensuring, for better or worse, that man’s economic problem never gets fully solved, and that the sweat of our brows continues to trickle.

Aye, but there’s a catch! All the preceding discussion of whether people will continue to work rests on one assumption: that there would still be work for people to do.

More precisely, our discussion has presupposed that the income one could earn by selling one’s labor remains significant compared to the income one derives from other sources, such as capital holdings and social transfers.

Recall the billionaire with the megayacht: no matter how badly he envies the decabillionaire’s gigayacht, he would not continue selling his labor if the most he could make is the minimum wage or some other amount that is trifling compared to what he makes from his investments (or compared to what he can afford to spend for the rest of his life by slowly drawing down his savings).

Here we come to a juncture where we need to consider that the labor market impacts of advanced AI may be different and more transformative than those that result from even very large increases in productivity brought about by capital accumulation and technical advancements of the sort that Keynes envisaged in his essay.

Historically, labor has been, on net, a complement to capital. At the aggregate level, this has held true since the beginning of tool use and through all subsequent epochs of technological change and economic growth.

You all know what complements and substitutes are in economics, right? We say that X is a complement to Y if having more of Y makes extra units of X more valuable. A left shoe is a complement to a right shoe. If, instead, having more of X makes Y less valuable, we say that X and Y are substitutes. A lighter is a substitute for a box of matches.

Okay, it turns out that labor and capital have been complements. Each has enhanced the value of the other. Of course if we zoom in, we can see that some particular kinds of labor have become less valuable as a result of technical innovation, while other kinds have become more valuable. But the overall effect, so far, has been that labor has become more valuable than it used to be. This is the reason why wages are now higher than they were a hundred years ago or at any other time in human history.

So long as human labor remains a net complement to capital, growth in capital stocks should tend to drive up the price of labor. The increasing wages could then motivate people to continue to work just as hard as they currently do even if they become very rich, provided they have the kind of insatiable desires that I just described. In reality, permanently higher wages would probably cause people to work a bit less, as they would choose to use some of their productivity gains to increase leisure and some to increase consumption. But in any case, the degree to which labor is a complement to capital is a function of technology. With sufficiently advanced automation technology, capital becomes a substitute for labor.

Consider the extreme case: imagine that you could buy an intelligent robot that can do everything that a human worker can do. And suppose that it is cheaper to buy or rent this robot than to hire a human. Robots would then compete with human workers and put downward pressure on wages. If the robots become cheap enough, humans would be squeezed out of the labor market altogether. The zero-hour workweek would have arrived.19

If we consider a less extreme scenario, the picture gets more complex. Suppose that robots can do almost everything that humans can do, but that there are a few tasks that only humans can do or that humans can do better. (This might include various new jobs that arise in opulent high-tech economies.) To determine the outcome for human wages in this scenario, we need to consider several effects.

First, as before, there is downward pressure on wages due to competition from robots.

Second, the economy in this full-bore automation scenario would most likely expand explosively, causing average income to shoot up. This would increase demand for labor, since higher-earning consumers would spend more on goods and services, including ones which we assumed only humans could produce. This increase in demand would create upward pressure on human wages.

Third, the increased average wealth in this scenario would likely reduce labor supply, since wealthier people would choose to work less at any given wage level. Such reduced labor supply would create upward pressure on wages.

Thus, there are at least these three basic effects: one that tends to depress wages, and two that tend to raise wages. Which of these effects dominates is not determinable a priori.

Therefore, whereas the effects of perfect automation technology are clear—full human unemployment and zero human labor income—the consequences of imperfect automation technology for human employment and human wages are theoretically ambiguous. For example, it is possible in this model that if robots could do every job except design and oversee robots, the wages paid to human robot-designers and robot-overseers could exceed the total wages paid to workers today; and, theoretically, the total number of hours worked could also rise.

We would have to make a whole bunch of particular and rather speculative empirical assumptions if we wanted to derive more specific implications from our model. At that point, we might as well start to disaggregate the impact of automation and look not at the total level of employment but at how individual sectors of the labor market would be impacted. No doubt, some occupations would do better and some worse in such scenarios of partial automation. But as none of that is particularly germane to our topic, we will leave it to our friends in the economics department to work out the details.

It is interesting, though, to glance at what happens to human wages and working hours if we start with an imperfect automation scenario and gradually transform it into one of increasingly perfect automation. If we consider a scenario in which automation technology is very nearly perfect—machines that can do virtually everything that humans can do, better and more cheaply, with only a few minor exceptions—then I would expect that humans would work only a little. People might work on average a couple of hours a week, doing the very few things that machines can’t. As for labor income, however, we cannot even conclude that there is such an asymptotic convergence to the case of perfect automation. For it is possible that hourly wages could rise so steeply that even if people work only two hours a week, they might still make more money than they currently do in a forty-hour work week. (I think it is also theoretically possible, though empirically unlikely, for the factor share of labor to increase in such scenarios.)

Now you might wonder: What are the limits to automation? How close to perfectly will robots substitute for human labor? This is a key factor that will determine whether we end up in a Keynesian leisure society, or an even more extreme scenario in which humans are entirely out of work and in which we consequently will confront the full force of the purpose problem.

We’ll get to that question. But before we do, I’d like to take a little detour to talk about how humans could make money even if the substitution were perfect and there were no jobs for humans. I mean, it’s reasonable to wonder about income and not just purpose in an AI-driven full-automation future.

A simple three-factor model

Consider a very simple three-factor model in which economic output is produced by combining labor, capital, and what is commonly referred to as land. Land here means any non-labor inputs that we cannot produce more of, so not just planetary surface area but other basic natural resources as well. We will consider an extreme scenario in which the share of income that goes to labor is zero: one in which, consequently, the combined factor shares of capital and land is one hundred percent.

Let’s first consider what happens if we assume that there is no change in population, no technological progress, and no increase in land, but there is an unexpected shock, namely the sudden invention of cheap robots that are perfect substitutes for all human labor. We’ll also assume a fully competitive economy with no monopoly rents, and we’ll assume fully reliable property rights (and that the robots remain under human control).

We start with an economy of full human employment. Then the perfect robots are invented. This causes massive amounts of capital flow into the robotics sector, and the number of robots increases rapidly. It is cheaper to build or rent a robot than to hire a human. Initially, there is a shortage of robots, so they don’t immediately replace all human workers. But as their numbers increase, and their cost goes down, robots replace human workers everywhere.

Nevertheless, the average income of humans is high and rising. This is because humans own everything, and the economy is growing rapidly as a result of the successful automation of human labor. Capital and land become exceedingly productive.

Capital keeps accumulating; so eventually land is the only scarce input. If you want to visualize this condition, you could imagine that every nook and cranny has been filled with intelligent robots. The robots produce a flow of goods and services for human consumption, and they also build robots and maintain and repair the existing robot fleet. As land becomes scarce, the production of new robots slows, as there is nowhere to put them or no raw materials with which to build them—or, more realistically, nothing for them to do that cannot be equally well done by the already existing robots. Non-physical capital goods might continue to accumulate, goods such as films, novels, and mathematical theorems.20

There are no jobs and humans don’t work, but in aggregate they earn income from land rents and intellectual property. Average income is extremely high. The model doesn’t say anything about its distribution.

Even though economic work is no longer possible for humans, there may continue to be wealth flows between individuals. Impatient individuals sell land and other assets to fuel consumption spurts; while more long-term-oriented individuals save a larger fraction of their investment income in order to grow their wealth and eventually enjoy a larger total amount of consumption. Another way to climb the wealth ranking in this steady state of the economy may be by stealing people’s or countries’ property, or by lobbying governments to redistribute wealth. Gifts and inheritances may also move some wealth around. And beyond these sources of economic mobility, there is always the craps table and the roulette.

This may all seem a bit wild?

But notice that if we replace robot with farmer, what we have is not a bad description of most of human history.

At equilibrium, both farmers and robots earn subsistence-level income. In the case of farmers, this means enough bread to raise two reproducing children per couple. In the case of robots, it means the revenue generated by each robot equals the cost of its manufacture and operation.

In this analogy, the landowning aristocrats of the past correspond to the rich future human population, which, just like their historical counterparts, extracts rents from their landholdings.21

What allows the average income of the future humans in this model to rise above subsistence is the stipulation that the human population is capped. If the number of humans (like the number of robots) were permitted to grow freely, then average human income would fall to subsistence level (like the robot’s income falls to their subsistence level) once the size of the human population attains its evolutionary equilibrium.

We would then have a situation in which there is a vast number of robots, a vast number of humans, very high world GDP, and mere subsistence-level average incomes. This would be essentially just a scaled-up version of the bleak picture of the world that Thomas Malthus presented.

This simple three-factor model makes a number of assumptions which can of course be questioned.

The assumptions that there is no technological progress and no increase in land are, I think, less rickety than might initially appear. I expect that the rate of economically relevant technological progress will eventually asymptote to zero (once most useful inventions have already been made). Land growth (from space colonization) will asymptote to a polynomial rate, since the volume of the sphere reachable from Earth by a given time is bounded by the speed of light. In the very long run, land growth will asymptote to zero, since the expansion of space means that sufficiently remote galaxies are forever unreachable from our starting point. But even during the long period in which a polynomial rate of land growth could be sustained, a decline of average income to subsistence can easily occur, since a population is able to grow at an exponential rate.22

The assumption that humans will remain in perfect control of the robots is definitely open to doubt, though it is not one that I intend to discuss in these lectures. If that assumption is relaxed, the result would either be the same as above except with a somewhat smaller human population and a somewhat larger robot population at equilibrium; or, in the case of a more complete failure of control, the human population could disappear altogether and there would be even more robots.

By the way, I should say that when I speak here of the robot population or the number of robots, what I mean to refer to is the factor share of the automation sector in the economy. Rather than a population composed of some specific number of independent robots, it could all just be one integrated AI system that controls an expanding infrastructure of production nodes and actuators.

Another assumption in the simple three-factor model is that property rights are fully preserved and that there is, for example, no redistribution program or welfare system. And we haven’t yet considered economic inequality within the human population. Let’s poke some more…

(It might seem as if we are going on a bit of a tangent here, but if one is pondering possible futures that involve notions of sustainable abundance, it is useful to be aware of these considerations and constraints. It also helps us to explicate our past human condition, thereby providing a backdrop against which utopian aspirations will stand out in sharper relief. And it begins to illustrate the many and various ways in which the quest for a better world, and for utopia, is often paradoxical.)

Paradoxes of a Malthusian world

We often think of economic inequality as bad. In a Malthusian context, however, it appears to have a silver lining.

Given unrestricted population growth, inequality is the only way that at least some fraction of the population can enjoy consistently above-subsistence-level incomes. If one holds that it is intrinsically important that there exist at least a few people who enjoy the finer things in life, then such an unequal arrangement might be deemed better than one in which there exists a slightly larger number of people but where everybody has a muzak and potatoes life (to borrow a phrase from Derek Parfit).23 Historically, there have also been instrumental benefits to having some rich folk around who could patronize the arts and sciences and create pockets of privilege, sufficiently isolated from the immediate struggle for survival, so that new things could be invested in and tried out.

You might think that, in the Malthusian equilibrium, average income would obviously be higher if there is inequality—since if there is no inequality, then everybody earns subsistence wages, whereas if there is inequality, then at least some people have above-subsistence incomes. But things are not quite so straightforward.

Consider that, where there is inequality, the classes that enjoy above-subsistence income—for example, the landowning elites—reproduce at above replacement levels. Some of their children must therefore leave the class they were born into and fall to a lower stratum. This trickle-down of population, from the higher classes to the lower, implies that average income among the lower classes is below subsistence level in the steady state; since otherwise, the total size of the population would increase. So in this model, the peasant class has below subsistence income, yet its numbers remain constant as it gets continuously replenished from above by the drip of excess progeny that is falling from the bottom stratum of the landowning elite.

(We could analogize the situation to that of a lump of ice floating in water. If we have a thin flat sheet of ice—perfect equality—all the ice crystals will be near the surface of the water: at the level of bare subsistence. If instead we have a tall and pointed shape of ice, an iceberg, then some parts could stick up high above the surface, enjoying economic plenty; but this necessarily depresses other parts of the ice, to income levels below subsistence.)

Inequality could however raise average income in the Malthusian equilibrium if we assume that the relationship between income and fitness is not linear. This is easiest to see if we consider an extreme example: a king and a queen who have an income 100,000 times larger than that of a peasant couple—yet the regal pair would not have 100,000 times more surviving children. So inequality probably would increase average income in the Malthusian steady state.

On the other hand, inequality might reduce average well-being, since a person’s well-being is not proportional to her income but rather perhaps to the logarithm of her income or some other such functional form of rapidly diminishing returns. If the king and the queen gained some new tributaries and increased their income tenfold, their expected well-being would presumably increase by much less than 10x.

In reality, the Malthusian condition was only ever roughly approximated. It was frequently disrupted by exogenous shocks. Every now and then, a plague, a famine, a massacre would cull the herd, thereby increasing the land and capital available to each of the survivors. For a time, even the majority could then enjoy significantly above-subsistence incomes.24 This improved comfort led to lower childhood mortality, causing the population to grow back to the point where land was again scarce enough to suppress the average farmer’s income back to mere subsistence—or slightly below, given the existence of economic inequality.

What is it like to live in a Malthusian condition? The simple assumptions we’ve made so far do not allow us to derive any general statement about this.

For example, you could have a model of fluctuating fortune within a life, where an individual dies if at any point their fortune dips below a certain threshold. In such a model, an individual may need to have a high average level of fortune in order to be able to survive long enough to successfully reproduce. Most times in life would thus be times of relative plenty.

In this model, inventions that smooth out fortune within a life—such as granaries that make it possible to save the surpluses when times are good and use them in times of need—lead to lower average well-being (while increasing the size of the population). This could be one of the factors that made the lives of early farmers worse than the lives of their hunter-gatherer forebears, despite the advance in technology that agriculture represented. Those grain depots smoothed out consumption, enabling farmers to survive long enough to reproduce even when their average income over their lifetime was hovering just above subsistence. Without the ability to store food, average conditions would need to be