What We Owe The Future

This book is about longtermism: the idea that positively influencing the longterm future is a key moral priority of our time. Longtermism is about taking seriously just how big the future could be and how high the stakes are in shaping it. If humanity survives to even a fraction of its potential life span, then, strange as it may seem, we are the ancients: we live at the very beginning of history, in the most distant past. What we do now will affect untold numbers of future people. We need to act wisely.

I now believe the world’s long-run fate depends in part on the choices we make in our lifetimes. The future could be wonderful: we could create a flourishing and long-lasting society, where everyone’s lives are better than the very best lives today. Or the future could be terrible, falling to authoritarians who use surveillance and AI to lock in their ideology for all time, or even to AI systems that seek to gain power rather than promote a thriving society. Or there could be no future at all: we could kill ourselves off with biological weapons or wage an all-out nuclear war that causes civilisation to collapse and never recover.

Future people count. There could be a lot of them. We can make their lives go better.

This is the case for longtermism in a nutshell. The premises are simple, and I don’t think they’re particularly controversial. Yet taking them seriously amounts to a moral revolution—one with far-reaching implications for how activists, researchers, policy makers, and indeed all of us should think and act.

It’s not just technology that has improved people’s lives; moral change has done so, too. In 1700, women were unable to attend university, and the feminist movement did not exist.If that well-off Brit was gay, he could not love openly; sodomy was punishable by death. In the late 1700s, three in four people globally were the victims of some form of forced labour; now less than 1 percent are. In 1700, no one lived in a democracy. Now over half the world does.

And though the future could be wonderful, it could also be terrible. To see this, look at some of the negative trends of the past and imagine a future where they are the dominant forces guiding the world. Consider that slavery had all but disappeared from France and England by the end of the twelfth century, but in the colonial era those same countries became slave traders on a massive scale. Or consider that the mid-twentieth century saw totalitarian regimes emerging even out of democracies. Or that we used scientific advances to build nuclear weapons and factory farms.

The fact that our time is so unusual gives us an outsized opportunity to make a difference. Few people who ever live will have as much power to positively influence the future as we do. Such rapid technological, social, and environmental change means that we have more opportunity to affect when and how the most important of these changes occur, including by managing technologies that could lock in bad values or imperil our survival. Civilisation’s current unification means that small groups have the power to influence the whole of it. New ideas are not confined to a single continent, and they can spread around the world in minutes rather than centuries.

We don’t need to predict every detail, nor could we if we tried. But if we want to make the future better, we need to identify actions that have positive effects on balance over very long timescales.

Consider some state of affairs that people could bring about, like the nonexistence of the glyptodonts. We can assess the longterm value of this new state of affairs in terms of three factors: its significance, its persistence, and its contingency.

Significance is the average value added by bringing about a certain state of affairs. How much worse is the world, at any one time, because the glyptodonts are extinct? In assessing this, we would want to attend to all relevant aspects of the glyptodonts’ extinction: the intrinsic loss of a species on the planet, the loss to humans who could have used their shells or eaten their meat, and the impact on the ecosystems the glyptodonts inhabited.

The persistence of a state of affairs is how long that state of affairs lasts, once it has been brought about. The nonexistence of the glyptodonts may be exceptionally persistent, starting 12,000 years ago and lasting until the end of the universe. It would only fail to be exceptionally persistent if, at a future time, we were to bring them back…

The final aspect of the framework is contingency. This is the most subtle part of the framework. In English the word “contingency” has a few different meanings; in the sense I’m using it, an alternative term would be “noninevitability.” Contingency represents the extent to which a state of affairs depends on a small number of specific actions. If something is very contingent, then that change would not have otherwise occurred for a very long time, or ever. The existence of the novel Jane Eyre is very contingent: if Charlotte Brontë had not written it, that precise novel would never have been written by someone else. Agriculture is less contingent because it emerged in multiple locations independently.

If something is very noncontingent, then the change would have happened soon anyway, even without the individual’s action. Knowledge of calculus was not very contingent because Leibniz independently discovered it just a few years after Newton did. Considering contingency is crucial because if you make a change to the world but it’s a change that would have simply happened soon afterward anyway, then you have not made a longterm difference to the world.

Looking back, it’s clear that, for many of my teenage choices, what mattered most was not the fun I had at the time—whether buildering was a thrill (it was) or whether studying medicine at Edinburgh involved better parties. Rather, what mattered most was the impact of these choices on the rest of my life, whether I was risking death or altering the values that would guide my future self.

There are three aspects to expected value. First, probabilities. Rather than thinking that a three-of-a-kind poker hand is “very unlikely,” Liv knows that the chance of getting one, before any cards are dealt, is about 5 percent; if the first two cards she’s dealt are a pair, this probability rises to about 12 percent. Though both probabilities are small, the difference between them can easily be enough to affect your decisions at the poker table…

The second aspect of expected value is assigning values to outcomes. For professional poker players, this is comparatively easy: they can just look at their financial returns. But financial returns are not in general the right measure of value. If you need £1000 to pay for a life-saving operation, then the difference in value for you between getting nothing and getting £1000 is much greater than the difference in value between getting £1000 and getting £2000. The value that we assign to outcomes should be based on whatever it is we ultimately care about, such as people’s wellbeing…

This brings us to the third aspect of expected value theory, which is measuring how good or bad a decision is by its expected value. This can be intuitive: in the two-drugs example I just gave, the first drug is the better choice; death is more than ten times as bad as a mild headache, so a 10 percent risk of death is sufficient to outweigh a guarantee of a headache. We can calculate the expected value of a decision as follows. First, we list each possible outcome of the decision. Next, we assign a probability and a value to each outcome, which we then multiply together. Finally, we add up all the probability-times-value products.

Much the same will be true for the issues that I cover in this book. I’m not saying that we should be confident that value lock-in or major catastrophe will occur this century. What I am saying is that their chance of occurring is very real—certainly more than 1 percent, and certainly greater than many everyday risks, like dying in a car crash. When combined with how much is at stake, the expected value of trying to ensure a good future is enormous. When we’re applying the significance, persistence, and contingency framework, we should therefore be thinking about expected significance, expected persistence, and expected contingency.

Often, some event can have highly significant, persistent, and contingent effects if there is a period of plasticity, where ideas or events or institutions can take one of many forms, followed by a period of rigidity or ossification. The dynamic is like that of glassblowing: In one period, the glass is still molten and malleable; it can be blown into one of many shapes. After it cools, it becomes rigid, and further change is impossible without remelting.

A period of plasticity also commonly occurs when some idea or institution is still new. For example, the US Constitution was written over just four months—a moment of great plasticity—and amended eleven times in its first six years of operation. After that, though, it became more rigid. Between 1804 and 1913, only three amendments were passed, all immediately following the Civil War: they abolished slavery, granted citizenship to African Americans and formerly enslaved people, and prohibited race from influencing the right to vote. Today, the Constitution is again very rigid: it’s only been amended once in the last fifty years, and that amendment—to prevent increases in congressional salaries from taking effect until the next term of office—was first proposed in 1789.

The abolition of slavery was an example of a values change, by which I mean a change in the moral attitudes of a society, or in how those attitudes are implemented and enforced. In my view, the abolition of slavery was one of the most important values changes in all of history. Over the course of this chapter and the next, I’ll argue that changing society’s values is particularly important from a longtermist perspective. This chapter will look at the significance and contingency of values changes; the next chapter will discuss their persistence.

Values changes are significant because they have major impacts on the lives of people and other beings. But from a longtermist perspective, they are particularly significant compared to other sorts of changes we might make because their effects are unusually predictable.

If you promote a particular means of achieving your goals, like a particular policy, you run the risk that the policy might not be very good at achieving your goal in the future, especially if the world in the future is very different from today, with a very different political, cultural, and technological environment. You might also lose out on the knowledge that we will gain in the future, which might change whether we even think that this policy is a good idea. In contrast, if you can ensure that people in the future adopt a particular goal, then you can trust them to pursue whatever strategies make the most sense, in whatever environment they are in and with whatever additional information they have. You can therefore be fairly confident that you have made the achievement of that goal more likely, even if you have no idea at all what the world will be like when those future people act.

For these reasons changing values has particularly great significance from a longterm perspective. Looking to the past, we see that such changes have had an enormous impact on the lives of billions of people. Looking to the future, if we can improve the values that guide the behaviour of generations to come, we can be pretty confident that they will take better actions, even if they’re living in a world very different from our own, the nature of which we cannot predict.

In other cases, there are multiple peaks on the landscape, such as when there are different ways of adapting to the same environment. Beavers and platypuses both make slow-moving creeks and rivers their home, but they have very different traits. When there is more than one peak, we say there are multiple equilibria. This introduces contingency into evolution, since which peak an organism ends up climbing will depend on where it starts on the fitness landscape, how that landscape is shaped, and the randomness inherent to genetic mutation.

Slavery is so abhorrent that, before getting to grips with the historical scholarship on the topic, I assumed that abolition must have been inevitable. But now I’m not at all sure. Though it’s impossible to know for certain, it’s entirely plausible to me that, were the tape of history rerun a hundred times with slightly different starting conditions, in a significant proportion of those reruns, there would still be legal slavery in many or most countries in the world, even at today’s level of technological development.

This suggests that, as longtermists, when trying to improve society’s values, we should focus on promoting more abstract or general moral principles or, when promoting particular moral actions, tie them into a more general worldview. This helps ensure that these moral changes stay relevant and robustly positive into the future.

We are very far from creating the perfect society, and until then, in order to drive forward moral progress, we need morally motivated heretics who are able to endure ridicule from those who wish to preserve the status quo.

They thought that, if you made a lifelong commitment to self-improvement, you could transform spiritually into a sage. They likened cultivating your character to craftsmanship: cutting bone, carving a piece of horn, or polishing a piece of jade.

Confucius’s Analects still sells hundreds of thousands of copies annually. Every day, quotes from these sources influence political decision-making around the world.

In this chapter I discuss value lock-in: an event that causes a single value system, or set of value systems, to persist for an extremely long time. Value lock-in would end or severely curtail the moral diversity and upheaval that we are used to. If value lock-in occurred globally, then how well or poorly the future goes would be determined in significant part by the nature of those locked-in values. Some changes in values might still occur, but the broad moral contours of society would have been set, and the world would enact one of only a small number of futures compared to all those that were possible.

Writing, for example, was crucial, enabling complex ideas to be transmitted many generations into the future without inevitable distortion by the failures of human memory. The persistence of religious values, or moral worldviews like Confucianism, would not have been possible without writing as a technology.

Would we wield such unprecedented power responsibly? Worryingly, the pursuit of value lock-in has been common throughout history. We saw that when the Qin took control of China, they undertook a programme to systematically eradicate competing schools of thought; similarly, the Han systematized Confucian teachings to the detriment of competing schools. The Mohists, too, desired to lock in their own values indefinitely, if only they had the power. They saw moral disagreement as the biggest problem in the world and thought that the solution was to ensure that everyone had the same values.

When we look at history, we see that the predominant culture in a society tends to entrench itself, eliminate the competition, and take steps to replicate itself over time. Indeed, many moral views regard their own lock-in as desirable.

This is known as the “alignment” problem. It’s discussed at length in other excellent books, like Superintelligence, Stuart Russell’s Human Compatible, and Brian Christian’s The Alignment Problem, so I won’t go into it in depth here.

For this reason, even if you think that the absence of future generations is not a moral loss or that the end of civilisation would be a good thing (issues that I discuss in Part IV), it’s still very important to avoid AI takeover or the lock-in of bad values. There will be future generations of intelligent beings either way, and by preventing the takeover of the world by an AI with bad values, you are changing how good or bad the future is over the course of civilisation’s life span. That’s the main effect, rather than any impacts on civilisation’s life expectancy.

Those values could be narrow-minded, parochial, and unreflective. Or they could be open-minded, ecumenical, and morally exploratory.

By definition, the size of a country’s economy is given by its GDP per person and its population size. And the United States’ current population size is due, in part, to the high rates of immigration from Europe to North America from 1607 onwards and especially after 1830. In the future, countries that maintain high rates of immigration and cultural assimilation will grow in size and power; indeed, journalist Matt Yglesias recently proposed that, in order to maintain global influence, the United States should radically increase immigration, aiming to have a population of one billion people.

On average, atheists have few children compared to the religious, especially fundamentalists and those in poorer countries. Over time, this matters. According to the Pew Research Center, by 2050 the proportion of people with no religious affiliation (which includes atheists, agnostics, and people who do not identify with any religion but who may hold some religious or spiritual beliefs) will decrease from 16 percent to 13 percent of the world population. The primary reason for this is the higher fertility rate among religious groups; conversions in and out of a religion play a surprisingly small role in total numbers.

But if we are guilty of gross moral errors ourselves, then locking in our present values would also be a disaster.

Instead, we should try to ensure that we have made as much moral progress as possible before any point of lock-in. Political philosophers often argue over what an ideal state would look like. I think we should accept that we don’t know what the ideal state would be; the primary question is how we can build a society such that, over time, our moral views improve, people act more often in accordance with them, and the world evolves to become a better, more just place.

Charter cities are often promoted by those who want to see more economically liberal policies. But there is no necessary connection between these two ideas. For almost every social structure we can imagine, we could have a charter city based on that idea; there could be Marxist charter cities and environmentalist charter cities and anarchist communitarian charter cities. We could find out, empirically, which of these brings about the best society. And, in addition to creating a diversity of formal institutions, we could try to cultivate a diversity of cultures, too.

This last point—that we need to structure global society so that cultural evolution guides the world towards better values and better societal structures—highlights an issue facing the design of a morally exploratory world that I’ll call the lock-in paradox. We need to lock in some institutions and ideas in order to prevent a more thoroughgoing lock-in of values. One challenge is that these institutions and ideas will be morally controversial; for example, from many fundamentalist religious perspectives, the idea that we would encourage or even allow a diversity of worldviews might be regarded as abominable. Similarly, the idea that the path to the correct moral view is via reflection and good-faith debate, rather than studying the scripture of a holy book, is not one that everyone would accept.

The lock-in paradox thus resembles the familiar paradox of tolerance— the necessity for liberal societies to defend themselves against intolerant views that would undermine their freedom, even if doing so requires curtailing the very tolerance they want to preserve.

In a recent article called “Are Ideas Getting Harder to Find?,” economists from Stanford and LSE analysed this phenomenon quantitatively. Across a range of industries, across firms, and in the aggregate economic data they found the same thing: progress becomes harder and harder. Based on their numbers, in order to double our overall level of technological advancement, we need to put in, conservatively, four times as much research effort as we did for the previous doubling. To illustrate, suppose (simplistically) that initially it took 10 person-years of “research” to double the world’s level of technological advancement: to move from knowing only how to make a stone axe to knowing how to make both an axe and a spear. In order to get the next doubling of technological progress, it would take 40 person-years of research. The next doubling would take 160 person-years, then 640 person-years, then 2,560 person-years, and so on.

So to think about whether we can sustain technological progress, we have to think about whether we can keep exponentially growing the number of researchers. Consider that there are two ways to do this. First, you can increase the share of the population that is devoted to research. Indeed, we’ve been doing a lot of that, so that’s been the source of most of US technological progress in the last few decades. Technology-driven growth of US per-capita incomes has averaged about 1.3 percent per year. A full percentage point of that comes from increasing the fraction of the population doing R&D and from improving the allocation of talent, such as by reducing gender and racial discrimination.

The second way by which you can increase the number of researchers is by increasing the total size of the labour force: that is, you can grow the population. Over the last few decades, population growth has contributed about 0.3 percentage points to the United States’ technologically driven per-capita growth rate.

The deeper you’ve fallen, the harder it is to get out, and the expected length of stagnation would be greater.

Here are two reasons why this might happen. First, perhaps, as political economist Benjamin Friedman argues, people are more morally motivated in times of economic growth. When the economy is growing, everyone can be better off than they were in the past. This means, Friedman argues, that citizens will worry less about how their life compares to the lives of people around them and will be more supportive of generous, open, and tolerant social policies. And if you look at the historical record, he claims, countries tend to make moral progress—becoming fairer, more open, and more egalitarian—during higher-growth periods, and they tend to morally regress during periods of stagnation.

A second reason ties back to our earlier discussion of cultural evolution. When technological innovation is possible, there are great economic gains to be had from critical thinking and scientific inquiry; and since economically successful cultures gain more members, cultural evolution currently selects for traits conducive to science. As a side effect, so this argument goes, we apply our critical capacities to moral issues, too, and therefore make moral progress. In a stagnant world, the economic reasons to engage in critical thinking and scientific inquiry would be much weaker. Instead, other values would be selected for, such as those favouring hierarchy and conformity, which have guided so many societies in the past.

As New Yorker journalist Larissa MacFarquhar noted in her profile of him, “The driving force behind Parfit’s moral concern was suffering. He couldn’t bear to see someone suffer—even thinking about suffering in the abstract could make him cry.

But given the benefits of having children and raising them well, I do think that we could start to once again see having kids as a way of positively contributing to the world. Just as you can live a good life by being helpful to those around you, donating to charity, or working in a socially valuable career, I think you can live a good life by raising a family and being a loving parent.

In a currently unpublished large survey of over 8,500 people, psychologists Matt Killingsworth, Lisa Stewart, and Joshua Greene added a twist to the experience sampling approach. At random times, they asked participants to write down what activity they were doing and how long it would last, and then respond to the question, “If you could, and it had no negative consequences, would you jump forward in time to the end of what you’re currently doing?” That is, they asked participants to imagine having the option of simply not experiencing—though still doing—whatever activity they were engaged in at that moment.

Again, people skipped around 40 percent of their day, and on average, people were happier during the times they kept than they were unhappy during the times they skipped. Taking both duration and intensity into account, the negative experiences were only bad enough to cancel out 58 percent of people’s positive experiences.

In the Aymara language, the more important feature of time is what we know and what we don’t. We can see the present and the past; they are laid out before us. We can therefore have direct knowledge of them in a way we can’t know the future—anything we know or believe about the future is based on inference from what we have experienced in the present or the past. The implicit philosophy is that, when making plans for the future, we should take much the same attitude as if we were walking backwards into unknown terrain.

When confronted with the empirical and evaluative complexity that faces us, it can be easy to feel clueless, as if there’s nothing at all we can do. But that would be too pessimistic. Even if we’re walking backwards into the future—and even if the terrain we’re walking on is unexplored, it’s dark and foggy, and we have few clues to guide us—nonetheless, some plans are smarter than others. We can employ three rules of thumb.

These three lessons—take robustly good actions, build up options, and learn more—can help guide us in our attempts to positively influence the long term.

Maintaining a diversity of cultures and political systems leaves open more potential trajectories for civilisation; the same is true, to an even greater degree, for ensuring that civilisation doesn’t end altogether.

Perhaps some changes to the world’s institutions and cultures would be valuable trajectory changes. Either of these would be enormously important to identify. These and other crucial issues are worked on at places like the Global Priorities Institute, the Future of Humanity Institute, and Open Philanthropy.

Often the focus on consumption decisions is accompanied by a failure to prioritise. Consider, for example, the recent wave of advocacy for reducing plastic. The total impact this has on the environment is tiny. You would have to reuse your plastic bag eight thousand times in order to cancel out the effect of one flight from London to New York. And avoiding plastic has only a tiny effect on ocean plastic pollution. In rich countries with effective waste management, plastic waste very rarely ends up in the oceans. Almost all ocean plastic comes from fishing fleets and from poorer countries with less-effective waste management.

Emphasising personal consumption decisions over more systemic changes is often a convenient move for corporations. In 2019 Shell’s chief executive, Ben van Beurden, gave a lecture in which he instructed people to eat seasonally and recycle more, lambasting people who eat strawberries in winter. In reality, in order to solve climate change, what we actually need is for companies like Shell to go out of business. By donating to effective nonprofits, we can all make this kind of far-reaching political change much more likely.

Another way to improve the world is to talk to your friends and family about important ideas, like better values or issues around war, pandemics, or AI. This doesn’t mean that you should promote these ideas aggressively or in a way that might alienate those you love. But discussion between friends has been shown to be one of the most effective ways to increase political participation, and it is also probably a good way to get people motivated to work on some of the major problems of our time.

Rather than feeling locked in to one career path, you would see it is an iterative process in which you figure out the role that is best for you and best for the world. The value of treating your career like an experiment can be really high: if you find a career that’s twice as impactful as your current best guess, it would be worth spending up to half of your entire career searching for that path. Over time, it will become clearer whether you have found the right path for you. For many people, I think it would be reasonable to spend 5 percent to 15 percent of their career learning and exploring their options, which works out to two to six years.

For many people, personal fit can mean the best way of contributing is through donations: you work in a career you love and excel at, and even if the work itself is not hugely impactful, you can make an enormous difference with your giving. This was true of John Yan. After learning about effective altruism and thinking about his career options, he decided to continue as a software engineer and donate a significant fraction of his income to effective charities as a member of Giving What We Can.

You can do this by writing, organizing, talking to people you know, or getting involved with organisations such as 80,000 Hours and the Centre for Effective Altruism, where movement building is a component of their work.

Positive moral change is not inevitable. It’s the result of long, hard work by generations of thinkers and activists. And if there’s any change that’s not inevitable, it’s concern for future people—people who, by virtue of their location in time, are utterly disenfranchised in the world today.

If we are careful and far-sighted, we have the power to help build a better future for our great-grandchildren, and their great-grandchildren— down through hundreds of generations. But we cannot take such a future for granted. There’s no inevitable arc of progress. No deus ex machina will prevent civilisation from stumbling into dystopia or oblivion. It’s on us. And we are not destined to succeed.

One robust account of the concept of “sageliness” in Chinese philosophy can be found in Feng (1997,6-9).” (notes to footnote 4, Chapter 4, p.75)

Schein et al. 2020; Green and McClellan 2020.” (notes to footnote 24, Chapter 10, p.234)