207 | William MacAskill on Maximizing Good in the Present and Future
What this covers
Will MacAskill, a philosopher and co-founder of the Effective Altruism movement, makes the case for longtermism—the view that humanity's decisions now should weigh heavily the interests of vastly larger populations that might exist across future centuries or millennia. The conversation moves through the philosophical foundations of this position, examining why distance in time should not diminish moral concern any more than distance in space does, and why existential risks like engineered pandemics, misaligned artificial intelligence, and great-power conflict warrant urgent societal attention. MacAskill argues that most actions reducing extinction risk also yield immediate benefits, so concern for the distant future rarely pits itself against helping those alive today.
The discussion engages several entangled problems. Population ethics poses deep puzzles: Derek Parfit's work revealed that every attempted framework—whether total utilitarianism, critical level view, or alternatives—produces counterintuitive impossibility results. MacAskill sketches how to navigate these using Moral Uncertainty Compromise, holding partial credence across competing moral theories rather than committing wholly to one. The conversation also addresses specific neglected dangers: the risk of value lock-in if bad ideologies entrench themselves indefinitely, the demographic and technological fact that we live during an unusually pivotal century of rapid change, and the democratic deficit that future generations cannot vote on policies affecting them. MacAskill touches on practical implications too—career choice, charitable giving, and the weakness of arguments against having children based solely on climate impact—grounding the philosophical case in how individuals might act on these ideas.
MacAskill argues for longtermism: because the future may contain vastly more people than the present and our actions can pivotally shape it, we have strong moral reason to prioritize reducing existential risks and steering civilization toward a good trajectory, treating future people's interests as comparable to those of distant strangers today.
- Spatial discounting and temporal discounting are both morally unjustified; distance in space or time does not diminish a being's moral worth
- Existential risks (engineered pandemics, AI, great-power war) are non-negligible (~1 in 6 this century) and current preventive effort is vastly underfunded
- Most actions that reduce extinction risk also yield large near-term benefits, so longtermism rarely conflicts sharply with present-focused good
The far future contains vastly more value than the present, making its protection one of humanity's most important moral obligations.
- Sustained ~2% annual economic/technological growth cannot persist indefinitely (10,000 years of it would imply an economy 10^89 times today's, exceeding available atoms), so we live through an unusually fast period of technological progress relative to both past and future, during which dual-use technologies make this an especially perilous and pivotal time.
“Continuation of 2% per year growth for 10,000 years, means that the economy would be, I think, 10 to the power 89 times as big as it is today... it's just not plausible that we're gonna have that much technological advancement. So things have to slow.”
- Just as someone's well-being does not diminish morally because they live far away in space, the fact that someone lives a century, a millennium, or a million years in the future should not reduce the moral importance of their interests; we owe future people at least what we owe present-day strangers.
“it just seems obvious that the well-being of someone doesn't diminish if they move farther away from you in space... And similarly, just the fact that someone lives in a century's time or a thousand years time or even a million years time should not matter when we're considering kind of how important are their interests?”
- Longtermism is the view that we should take seriously the sheer scale of the future and the high stakes of shaping it, identifying pivotal events in our lifetime and navigating them well to bring about a just, prosperous, long-lasting future spanning potentially thousands, millions, or billions of years.
“long-termism really means long-term. Not just thinking, "Oh yeah, years, decades into the future," but really for how long we might live, which I think is thousands, millions or even billions of years”
- Humanity's future could unfold in two starkly different ways: either we continue advancing technologically and morally and spread to the stars where human agency shapes the cosmos for trillions of years, or we cause our own extinction and leave the universe empty and lifeless; preventing extinction is thus one of the most important ways to influence the long-term future.
“there's another possible future, which is that humanity does not get to that point of developing technologically, instead we cause our own extinction. And then the story of the universe would be much as the story of the past... the universe would be empty, it would be lifeless, it would be dead.”
Existential and catastrophic risks from pandemics, engineered pathogens, misaligned AI, and value lock-in are sizeable, neglected, and demand urgent preventive action.
- Beyond natural pandemics, the increasing ability to create or upgrade engineered pathogens poses a real existential threat; MacAskill estimates roughly a 1% or lower chance this century of a pandemic reaching the scale of killing everybody, which though small is a higher probability of death than dying in a fire or by drowning and warrants serious societal action.
“I think it's 1% or less that it could reach within the coming century, that it could reach the scale of killing everybody.”
- Societies learn safety through trial-and-error feedback (planes became safe because of past crashes), but rare high-consequence events like century-scale pandemics or novel technologies provide no such feedback loop, so we systematically under-prepare; we therefore need clear-thinking, numerate, morally motivated people to act on these risks before they materialize, as with the atomic bomb.
“Pandemics kill many, many more people per year on average than plane crashes... But are we prepared for them? No, and it's because in terms of natural pandemics... that's like a one in a century event, we don't get to learn from feedback.”
- The probabilities of existential risk are sizeable—comparable to dying in a car crash and higher than a plane crash—with Toby Ord estimating roughly a one-in-six chance of existential catastrophe within our lifetime, a figure MacAskill broadly accepts; aggregated forecasting platforms like Metaculus provide one way to make such estimates more reliable.
“my, again, colleague, Toby Ord, puts the overall chance at about one in six of existential risk in our lifetime. And I don't significantly disagree with him on that.”
- A major neglected risk is value lock-in: bad values—a fascist ideology, or a misaligned AI pursuing valueless goals—could take control and entrench themselves indefinitely, ending moral progress prematurely and producing a dystopian rather than flourishing future.
“I call value lock in, this idea that certain bad values could take control and then just entrench themselves, keep themselves there indefinitely.”
Utilitarianism—measuring well-being through preferences and summing it across all conscious beings—provides the most plausible moral framework.
- Utilitarianism (well-being-focused consequentialism) is more plausible than other consequentialist views because we have direct awareness of the goodness of good experiences and the badness of suffering, whereas the purported goodness or badness of a natural landscape or artwork feels like a mere judgment rather than something directly perceived.
“it seems to me I have direct awareness of this, of goodness, when I have a good experience. I have direct awareness of badness when I suffer. I don't have direct awareness of the goodness or badness of a painting”
- Well-being can be measured by appealing to people's carefully considered preferences and the trade-offs they would make—such as time trade-offs (indifference between two days skiing or one day at the beach before dying)—which, if they satisfy certain axioms, allow numbers to represent preferences on a ratio scale; the numbers merely represent fundamental preference orderings rather than being intrinsic quantities.
“the way that I would measure wellbeing is by appealing to people's like carefully considered preferences and the trade offs that they would make. My particular favorite method is time trade offs.”
- Although there is no settled theoretical account of how to compare well-being across people, such comparisons are clearly possible in practice—a shoulder pinch versus being horrifically beaten up is obviously worse for the latter—and experiences like headaches fall in similar ranges across people, so approximate interpersonal comparison enables meaningful utilitarian summation of total well-being.
“if you are saying that it's, you know, me pinching you in the shoulder and... someone else getting horrifically beaten up, there's no comparison between those things. Can't say for whom is that worse? I'm just like, come on.”
- Consequentialism holds you should do what brings about the best outcomes and that this is all that matters, which is unintuitive in two ways: you may not give yourself greater weight than others (so disposable income should go to where it does most good), and promoting good outcomes is always right even when the means—like killing one to save a hundred—conflict with common-sense morality.
“Consequentialism says you should do what's best, where what's best is what brings about the best outcomes. And it says that that's all that matters”
Moral uncertainty requires holding credences across competing theories and synthesizing their implications rather than committing to a single view.
- Given credence in both nihilism (moral error theory) and moral realism, you should act as a moral realist, because if nihilism is true nothing matters—including any reasons to benefit yourself—so acting morally loses nothing, whereas if realism is true you have done genuine good; since MacAskill puts his credence near 50/50, this is not a low-probability Pascalian wager.
“if you have got some degree of belief in both nihilism and the moral realist view... There's no chance of making a mistake by doing the moral realist thing, 'cause if nihilism is true, nothing matters.”
- Work on how to handle tiny probabilities of enormous amounts of value has produced impossibility results: no plausible view on the topic has generally plausible implications, so every position carries some highly counterintuitive consequences—though MacAskill notes the actual existential risks we face are not in this tiny-probability regime.
“the bigger results have been impossibility results, so just showing that no plausible view... In fact, proving that no plausible view... Sorry, no view you can have on this topic has implications that are just generally plausible.”
- Uncertainty about the future is often used as a justification for inaction but is a bad one; just as uncertainty about climate impacts justifies more precaution because things could be worse than expected, uncertainty about future generations and high-stakes risks gives us more reason to take a cautious, responsible approach rather than winging it.
“Uncertainty is often used as a justification for inaction, but I think it's a very bad justification.”
- Because we are uncertain which moral theory is correct, we should hold some degree of belief in a variety of moral views (consequentialism and non-consequentialism) and, when acting, choose the best synthesis or compromise between them; concern for the long term emerges from such a compromise rather than from pure consequentialism alone.
“you should have some degree of belief in a variety of different moral views, consequentialism and non-consequentialism. And when you're taking action you should try and act in a way that's the best synthesis or compromise between these different moral views.”
All sentient beings, including non-human animals and future digital minds, have equal moral worth independent of species or temporal location.
- There is nothing distinctively special about Homo sapiens; all animals with consciousness have moral worth, and future beings—whether evolved humans or artificial intelligences—would still be morally worthwhile so long as they pursue good values like knowledge, love, friendship, and happiness.
“I think there's nothing distinctively special about Homo sapiens. I think that all animals with consciousness have moral worth”
- We have some additional reasons to weight present people more—special relationships (family, friends) and reciprocity (debts owed for education, benefits from society and the state)—but these obligations only justify partiality up to a point and don't conflict with using disposable income to help distant or future people, whom we still owe at least what we owe present strangers.
“I do think we have some extra reasons to care about people in the present generation compared to people in future generations. And in the same way, I have special reasons to care about people near and dear to me compared to distant strangers.”
Every population ethics view faces paradoxes; a critical-level compromise avoids the repugnant conclusion but no perfect solution exists.
- Taking a compromise across views of population ethics yields a critical level view: above the zero level (indifference between existing and never having existed) there is a higher critical threshold, and only bringing into existence a life above that threshold counts as good, so adding barely-worth-living lives is not good and may be bad—this rejects unrestricted dominance addition and avoids the repugnant conclusion.
“There's just this higher level, and it's only above that higher level that it's good to bring someone into existence. So bringing someone to existence with a life that's barely worth living is not a good thing. It might even be bad.”
- Intuitions suggest the total value of good lives may be bounded with diminishing returns—people reluctant to gamble a flourishing 100-trillion-person utopia for a 50% chance of three such utopias versus extinction—but if value is bounded then present decisions absurdly depend on how many good or bad lives existed in the past or elsewhere in the universe, so every view of aggregation faces a paradox.
“if you think there's an upper bound to value, then it really matters how... When you make decisions now, it really matters how many people have lived in the past or how many people are alive in different places in the universe... and that just seems absurd to me.”
- Total utilitarianism implies the repugnant conclusion: a sufficiently vast population of lives barely worth living has greater total well-being than a smaller population of blissful lives; Parfit found this repugnant, but a step-by-step argument (dominance addition, then equalizing, iterated, with transitivity) shows you must reject one of several highly compelling principles or accept the conclusion—so every population-ethics view faces an impossibility result.
“you'll get this conclusion that the very, very large population with trillion, trillion, trillion, trillion people with lives that are just barely worth living, that has total greater wellbeing, then the much smaller, but still a very large population of people with total bliss lives. And that seems repugnant, says Derek Parfit.”
High-impact careers, substantial charitable giving, and informed political engagement are among the most effective ways individuals can improve the world.
- Because you spend roughly 80,000 hours working over a lifetime, choosing a career aimed at high-impact problems—reducing pandemic risk, safely guiding AI, reducing war risk—can do enormous good; alongside donating at least 10% of income to the most effective charities and politically well-informed voting, this is among the most effective ways for ordinary people to improve the world.
“you spend 80,000 hours working in the course of your life, maybe even more. So I set up an organization called 80,000 Hours that encourages people to pursue those careers that do good”
- The influential view that having children is bad because of climate impact is weak and possibly harmful because it counts only the harms a child imposes and ignores the benefits—children contribute to society, pay taxes, can become moral changemakers, and drive the innovation that, historically, raised humanity above subsistence farming; the same logic favors larger populations and makes the projected 2050 population peak and decline a real challenge.
“I think that argument is really quite weak and perhaps even harmful, and I think because it only looks at one side of the ledger. It looks at the harms that a child imposes on the world, but not the benefit too”
Temporal discounting and democratic representation both systematically undervalue future generations, leaving their interests undefended.
- Economists' rate of pure time preference (typically ~2% per year), which treats a benefit or harm as intrinsically worth less the further in the future it occurs, becomes absurd over long timescales—it would imply preventing one death in 10,000 years is more important than preventing a genocide of a million people in 11,000 years—so we should not discount future well-being in this way.
“if you have this rate of pure time preference that economists sometimes use of 2% a year, then you think it's more important to save the one life in 10,000 years than the million lives in 11,000 years. And that is absurd.”
- Democracy has a major flaw for longtermism: if future generations outnumber the present roughly a thousand to one, then current voters represent only ~0.1% of those affected by today's policies, and because future people can't represent themselves, any institution built to represent them will be vulnerable to capture by special interests—making genuine representation likely impossible.
“of the people that will be impacted by the policies we choose today, where let's conservatively say that future generations outnumber us a thousand to one... Well, then it's only 1.1% representation of those who are gonna be impacted.”
Moral progress is real and ongoing; our descendants may achieve deeper ethical understanding just as we have surpassed past eras.
- From the perspective of the Roman Empire our values look radically different—we don't own slaves, women hold office—yet this represents genuine moral progress, and our distant descendants could make further progress such that, given enough conversation, they could convince us their different goals are admirable, just as one could convince Aristotle that slavery is wrong.
“from the perspective of the Roman empire, the values that guide society today are very different. We don't own slaves... but this is all a very good thing, 'cause it seems like this is kind of moral progress.”
- Moral philosophy has contributed importantly over the last five years by using formal tools to demonstrate impossibility results across many areas of ethics.
“one of the things I think moral philosophy has contributed most over the last five years, especially as we've started to use some more formal tools, is demonstrating a whole bunch of these areas where you can prove a... impossibility results”
The Fermi paradox and the implausibility of pure time preference over millennia both support taking long-term futures seriously.
- The hypothesis that civilizations become technologically advanced but lose ambition cannot solve the Fermi paradox, because with so many planets and solar systems the required filter would need ~99.99999% of civilizations to become unambitious, which is implausibly high.
“it would have to be about 99.99999% of civilizations decides to be unambitious, and it seems hard for me to believe that it could be that high.”
- Whether one reasons over thousands or billions of years often makes little practical difference; the pressing near-term reasons to reduce catastrophe risk largely coincide with longtermist conclusions, so much of longtermism's action-guidance is compatible with a more modest near-termism that simply avoids things that could wipe us out in coming centuries.
“I don't think it makes an enormous difference whether we're on a time scale of thinking thousands of years into the future or billions of years into the future.”
- The 'horror' of effective altruism is that every charitable choice carries an unreal opportunity cost—funding malaria bed nets in one country means not funding deworming or another country, and someone is dead because of where you donated—but from the point of view of the whole world, splitting one's giving between present and future causes is a negligible reallocation, and current spending on protecting future generations (~0.001% of global GDP) is far too low.
“I fund bed nets to protect children against malaria. That means I've not funded deworming tablets... Some people are dead if you're donating, even just thousands of dollars, because of your decision. And so anything we do has this unreal opportunity cost”