clock menu more-arrow no yes mobile
Origami by Katharine Molloy / Photo by Anand Katakam

Filed under:

You have $8 billion. You want to do as much good as possible. What do you do?

Inside the Open Philanthropy Project.

Dylan Matthews is a senior correspondent and head writer for Vox's Future Perfect section and has worked at Vox since 2014. He is particularly interested in global health and pandemic prevention, anti-poverty efforts, economic policy and theory, and conflicts about the right way to do philanthropy.

In early 2015, I reported and wrote a profile of the Open Philanthropy Project, an offshoot of the charity recommender GiveWell, funded in large part by billionaires Cari Tuna and Dustin Moskovitz.

Open Phil has changed a lot since this article’s writing; it’s given out tens of millions of dollars more in grants, issued and updated long explanations of its philosophy, and expanded (and diversified) its staff considerably. Tuna and Moskovitz’s reported net worth, which they hope to donate almost all of, has since grown to over $11 billion. Open Philanthropy is also formally its own organization, no longer tied to GiveWell.

With that being said, I think the piece still offers a useful glimpse into how Open Philanthropy operates, and into effective altruism as a worldview and a practice. So here’s the original piece, anachronisms and all.


I sat in a San Francisco conference room a few months ago as 14 staffers at the charity recommendation group GiveWell discussed the ways in which artificial intelligence — extreme, world-transforming, human-level artificial intelligence — could destroy the world. Not just as idle chatter, mind you. They were trying to work out whether it’s worthwhile to direct money — lots of it — toward preventing AI from destroying us all, money that otherwise could go to fighting poverty in sub-Saharan Africa.

”Say you tell the AI to make as many paper clips as it can possibly make,” Howie Lempel, a program officer at GiveWell, proposed, borrowing a thought experiment from Oxford professor Nick Bostrom.

The super AI isn’t necessarily going to be moral. Even with positive goals, it could backfire. It could see the whole world as a resource to be exploited for making paper clips, for example.

”Just because it’s very intelligent doesn’t mean it has reasonable values,” Lempel said. “Maybe it starts turning puppies into paper clips.”

”Maybe it would turn the whole universe into paper clips,” cofounder and co-executive director Holden Karnofsky added.

Joining the GiveWell staff in the meeting was Cari Tuna, the president of Good Ventures, a foundation she and her husband, Facebook co-founder Dustin Moskovitz, founded with their roughly $8.3 billion [2018 update: now $11.1 billion] fortune. The couple plans on giving most of that sum away.

”We want to burn down our foundation before we die, and ideally well before we die,” says Tuna. So she and Moskovitz joined forces with GiveWell to form the Open Philanthropy Project, whose mission is to figure out how, exactly, they should spend their billions to do as much good as possible.

That may mean giving cash to poor people in Uganda, or distributing anti-malarial bed nets, but it also might mean funding research into how to prevent AI from killing us all. Or it might mean funding the fight to end mass incarceration in the US. Or it might mean funding biological research.

Open Phil (as the staff calls it, eschewing the OPP acronym) doesn’t know which of these is the best bet, but it’s determined to find out. Its six full-time staffers have taken on the unenviable task of ranking every plausible way to make the world a much better place, and figuring out how much money to commit to the winners. It’s the biggest test yet of GiveWell’s heavily empirical approach to picking charities. If it works, it could change the face of philanthropy.

Better charity through research

The team at Open Phil are effective altruists, members of a growing movement that commits itself to using empirical methods to work out how to do the most good it possibly can.

Effective altruism holds that giving abroad is probably a better idea than giving in the US. It suggests that giving to disaster relief is worse than giving elsewhere. It argues that supporting music and the arts is a waste. “In a world that had overcome extreme poverty and other major problems that face us now, promoting the arts would be a worthy goal,” philosopher Peter Singer, a proponent of effective altruism, writes in his new book, The Most Good You Can Do. In the meantime, opera houses will have to wait.

Effective altruism also implies it’s quite possible that even the best here-and-now causes — giving cash to the global poor, distributing anti-malarial bed nets in sub-Saharan Africa — are less cost-effective than trying to reduce the risk of the world as we know it ending. Hence, the chatter about AI. If it causes human extinction, then billions, trillions, even quadrillions of future humans who otherwise would have lived happy lives won’t. That dwarfs the impact of global poverty or disease at the present moment. As Bostrom writes in a 2013 paper, “If benefiting humanity by increasing existential safety achieves expected good on a scale many orders of magnitude greater than that of alternative contributions, we would do well to focus on this most efficient philanthropy.”

What’s radical about GiveWell and Open Phil is their commitment to do substantial empirical research before deciding on causes. Many other foundations pick issues based simply on the personal whims of the funder. If that whim is to fund medical science (as in the case of Howard Hughes), the world gains. But if it’s to fund a fancy art museum (as J. Paul Getty did with his fortune), then money that could have saved lives was, in the effective altruist view, frittered away.

”The vast majority of donors aren’t interested in doing any research before making a charitable contribution,” Paul Brest, former president of the Hewlett Foundation, wrote in an article praising effective altruism. “Many seem satisfied with the warm glow that comes from giving; indeed, too much analysis may even reduce the charitable impulse.” By contrast, effective altruists are obsessed with doing research into cause effectiveness. Open Phil has a literal spreadsheet ranking a number of different causes it might invest in.

That has earned effective altruism criticism from more traditional corners of philanthropy. Charity Navigator, which tries to ensure that charities’ money goes where they say it’s going, has been particularly opposed. Its CEO, Ken Berger, and consultant Robert Penna penned a venomous takedown in the Stanford Social Innovation Review, in which they replaced every mention of “effective altruism” with “defective altruism.”

”This approach amounts to little more than charitable imperialism, whereby ‘my cause’ is just, and yours is — to one degree or another — a waste of precious resources,” they write. In the comments of the piece, Penna clarified: “We do not believe that it is the role of anyone to say to another that his or her cause is not ‘worthy.’”

To effective altruists, this attitude reeks of moral nihilism. In response to Berger and Penna, Will MacAskill, who founded the effective altruist group 80,000 Hours and co-founded Giving What We Can, proposed a thought experiment. Say you’re standing before two burning buildings, one of which has a family of five trapped inside and the other of which is storing a $20,000 painting for a nearby museum. You only have time to save the family or the painting. What do you do? Save the family, right? Now, how is that different from choosing whether to save lives by giving to the Against Malaria Foundation or to make exhibits a little nicer by giving to the Metropolitan Museum of Art? It’s not, MacAskill claims, and that’s lethal to the argument for philanthropic pluralism: “[Berger and Penna] have to reject the idea that the family of five’s interest in continuing to live is weightier and more morally important than the museumgoers’ interest in viewing an additional painting.”

That’s the other thing about effective altruists: they’re utilitarians. Or, if not utilitarians obsessed with maximizing happiness, they’re consequentialists concerned with maximizing the good in whatever form it takes. “We want to give people more power to live the life they want to live. It’s a consequentialist moral framework,” GiveWell’s Karnofsky says. “Justice as an end in itself, liberty as an end in itself — those aren’t things we’re interested in.”

That lends itself to a particular political bent, which is left of center but technocratic, friendly to markets (when they can be shown to work), and, above all, cosmopolitan. Effective altruists, and GiveWell in particular, go to great lengths to emphasize that doing good abroad is just as valuable as doing it in America, and probably cheaper, as well. They’re sympathetic to the welfare state but far more jazzed about open borders.

Effective altruists tend to share a hyper-analytical personality type. Before visiting the GiveWell offices, I went to a Super Bowl party at Karnofsky’s house. We went around the room saying which team we were rooting for — the New England Patriots or the Seattle Seahawks — and why. Karnofsky said he was rooting for the Pats in light of then-recent allegations that they had purposely deflated their balls to win the AFC championship. Many detractors wanted them to lose as punishment for this offense, and Karnofsky thought it important to disabuse the public of the notion that the world can exact cosmic justice like that: “Trial by combat doesn’t work.” Tim Telleen-Lawton, a GiveWell analyst and roommate of Karnofsky’s, said he was rooting for a tie, as it was the most improbable outcome and thus the most exciting. My explanation — my dad’s a Seahawks fan, so I’m a Seahawks fan — felt a little under-reasoned by comparison.

My reasoning failures aside, the effective altruists’ tendency to rationally analyze everything is endearing, and I should disclose that I’ve been won over. I’m a cosmopolitan utilitarian, too. I’ve given to GiveWell’s recommended charities for years (GiveDirectly is my current favorite). I’m friends with many of the staffers outside work. I talked to Karnofsky and Berger about policy issues in the early going of Open Phil, even musing about what we should name the idiosyncratic set of positions we happen to share (“newtilitarianism” was rejected as an offense against the English language). And I was and remain deeply excited by the prospect of a dedicated team sharing my values doing empirical research to rank policy issues in order of importance — which is exactly what Open Phil is up to.

How Open Phil thinks about causes

Open Phil may not care about justice as an end in itself, but it’s certainly interested in it as a means. Criminal justice reform is one of its top priorities at the moment, not because of the concerns over due process and constitutional liberties that motivate many groups working on the topic, but due to the Open Phil team’s hard-to-dispute observation that prison is really, really awful.

Most Americans live lives far better than those of people in developing countries, but the same can’t necessarily be said for prisoners. That makes interventions involving American incarceration look similarly promising to ones benefiting the global poor. In global health, it’s common to talk in terms of “disability-adjusted life years” (DALYs), which measure a disease’s burden by considering both how many years of life it denies victims and how much worse it makes their lives before they perish. A disease that cuts 10 years off your lifespan and causes 10 years of partial paralysis before that has a higher DALY toll than one that just cuts off 10 years, for instance.

As part of his investigation into the issue, Alexander Berger, the program officer overseeing Open Phil’s policy work, did a quick and dirty analysis of how many DALYs would be saved by reducing incarceration by 10 percent. He posited that the “disability weight” of being in prison (that is, how much it reduces the quality of a life-year) is 0.5. A year in prison is half as good as one on the outside. For context, that’s roughly the same disability weight as having terminal cancer. Once you take the abject awfulness of prison into account, reducing incarceration starts to look like a great way to save hundreds of thousands of DALYs.

Criminal justice is also one of the few points of bipartisan agreement in contemporary politics. Over the past decade or so, conservatives at the state level have come around to the view that prisons and the law enforcement complex are just another form of big government, and deserving of major cuts. There are a number of bipartisan proposals to reduce incarceration on the federal level, as well. It’s at a point where philanthropic involvement could help push through significant reforms.

Open Phil’s spreadsheet of causes ranks policy interventions based on their importance and tractability: how much good they’re capable of producing, and whether philanthropy can actually effect policies that produce that good. Even taking the DALY analysis into account, criminal justice is listed as only moderately important. But it’s unusually tractable.

How Open Phil knows whether a cause is effective

To that end, Good Ventures is already spending a lot on criminal justice. It has given $3 million to the Pew Charitable Trusts’ Public Safety Performance Project, which works with states to develop policies that “reduce incarceration and correctional spending while maintaining or improving public safety and concentrating prison beds on high level offenders,” to quote Open Phil’s review of the group.

GiveWell prides itself on trying to rigorously determine the magnitude of impact of the charities it recommends, and Open Phil took the same approach here. Pew claims incarceration will fall 11 percent in the states they’re working with, compared with rates if they hadn’t intervened; that suggests the project has led Americans to spend 858,000 fewer person-years in prison, at a cost of $29 per person-year. That’s pretty good; wouldn’t you spend $29 to avoid a year in prison? Or, in wonkspeak, $58 for a DALY?

But Open Phil’s review expresses skepticism about that magnitude, as “in many states where PSPP did not provide intensive assistance, prison growth reversed or stagnated on its own.” It concludes, however, “We believe that PSPP increased the probability of reform and/or the quality of reforms in at least several of the states in which it worked.”

This is about as rigorous an evaluation as Open Phil can do. But it’s still considerably less rigorous than analyses GiveWell does of charities. Just look at GiveWell’s analysis of the Against Malaria Foundation, its current top-rated charity. To evaluate the group, GiveWell has to know how effective the anti-malarial bed nets it distributes are. And it does know that: it cites five randomly controlled trials about the effect of bed nets on childhood mortality in its evaluation of the foundation. That means GiveWell knows how many bed nets it takes to save a life, and how big of a donation to the Against Malaria Foundation is needed to buy and distribute that many bed nets. It has a strong quantitative estimate of the real-world effects of giving to the group.

By contrast, Open Phil has very little sense of how many people are spared prison time due to a donation to the Public Safety Performance Project. There are no randomly controlled trials about the effectiveness of the program’s particular approach in enacting criminal justice reforms. While there are well-established best practices for evaluating service delivery programs like the Against Malaria Foundation, none exist for evaluating advocacy efforts.

”Advocacy, even when carefully nonpartisan and based in research, is inherently political, and it’s the nature of politics that events evolve rapidly and in a nonlinear fashion, so an effort that doesn’t seem to be working might suddenly bear fruit, or one that seemed to be on track can suddenly lose momentum,” Steven Teles, who has consulted for Open Phil, and Mark Schmitt once wrote in the Stanford Social Innovation Review. “Advocacy evaluation should be seen, therefore, as a form of trained judgment — a craft requiring judgment and tacit knowledge — rather than as a scientific method.” This, they argue, makes “evaluating particular projects — as opposed to entire fields or organizations — almost impossible.”

Open Phil concedes that it’s not able to estimate the effects of the Public Safety Performance Project’s work precisely. “We have not invested much time in explicit estimates of PSPP’s cost effectiveness,” their review concludes. “We highly value the unquantified benefits of learning from PSPP and we do not believe policy-oriented philanthropy is likely to consist of proven, repeatable interventions with easily quantified expected impact.”

They’re right, of course. You can’t know the exact effectiveness of a policy intervention. But because of that, doing the kind of cause comparisons that Open Phil needs to do could prove maddeningly difficult.

The only ones in the room

Key to Open Phil’s thinking about policy is the idea of leverage. Effective advocacy work isn’t necessarily that expensive, and when it works, its impact can be several orders of magnitude bigger. For instance, $100 million spent lobbying for a health reform bill could produce hundreds of billions of dollars in new health spending. Even if you don’t think the lobbying is guaranteed to work, it’s an attractive-looking investment.

That’s why Open Phil’s other main area of interest on policy is preventing recessions. This is an unusual point of focus for a foundation. Apart from extraordinary measures like the 2009 stimulus, the task of ensuring that the economy doesn’t fall into recession and inflation doesn’t spiral out of control is the almost exclusive province of the Federal Reserve. And the Federal Reserve is generally regarded as un-lobbyable. Since donors haven’t traditionally thought anything could influence the Fed, spending like that hasn’t happened.

That is changing to a degree. A number of think tanks and grant-making groups interested in monetary policy have cropped up since the financial crisis, including the George Soros–funded Institute for New Economic Thinking, the Washington Center for Equitable Growth, and the Hutchins Center on Fiscal and Monetary Policy (whose staff includes former Fed Chair Ben Bernanke). But the advocacy side is extremely sparse. In particular, there aren’t many voices calling for the Fed to worry less about inflation and more about unemployment, which is still a problem seven years after the crisis hit.

Cari Tuna, president of Good Ventures.
Cari Tuna, president of the Open Philanthropy Project and Good Ventures.
Marvin Joseph/the Washington Post via Getty Images

There’s a lot of potential leverage here. Open Phil’s analysis notes that the most recent financial crisis cost the US economy around ten trillion dollars in lost output, not to mention the permanent harm it did to the economy’s ability to grow by reducing investment and forcing people out of the workforce. Even if you assume that a crash like that will happen only twice in a century, that means preventing huge recessions could have an annual impact in the hundreds of billions. The humanitarian benefits swamp those of reducing mass incarceration.

The great spreadsheet of causes lists the tractability of macroeconomic policy as “highly uncertain,” which makes finding grantees a bit of a challenge. So far Good Ventures has given $335,000 to support the Full Employment Project at the Center for Budget and Policy Priorities and $100,000 to back the Center for Popular Democracy’s Fed Up campaign, which aims to organize workers in favor of a looser, more pro-growth monetary policy. Both are, in their way, investments in advocacy.

But Open Phil is even less able to estimate the effectiveness of these investments than it is its investments in criminal justice reform. Pew’s Public Safety Performance Project at least produced an estimate for how many people its work kept out of jail, albeit one that’s nigh impossible to verify. But the Fed Up campaign and Full Employment Project are new endeavors. They don’t have track records to evaluate. That makes the work more exciting. But it also makes it hard to tell if the money is better spent here than elsewhere.

The plague

Policy is only one of four issue areas Open Phil covers. One is global health and development; research there isn’t a priority, since GiveWell has already learned a lot about what works. Another is science, where work is still preliminary. That leaves the fourth and final area: global catastrophic risks, or GCRs. Such as world-destroying AI.

The basic idea — originated by Oxford’s Bostrom in his paper ”Astronomical Waste” — is that human extinction or civilizational collapse would be so bad that even a small risk of it happening is worth expending considerable resources to reduce. There are likely billions, if not trillions or quadrillions, of humans who could live in the future, provided we don’t go extinct. Saving their lives is thus, on this view, massively more important than anything affecting people currently alive.

”Even if average future periods were only about equally as good as the current period, the whole of the future would be about a trillion times more important, in itself, than everything that has happened in the last 100 years,” Nick Beckstead, a GiveWell research analyst who works on Open Phil, wrote in his 2013 philosophy dissertation.

To that end, if there are dangers that pose a real risk of destroying civilization, and steps can be taken to reduce that risk, Open Phil is interested. And the stakes required are quite high, such that major concerns like antibiotic resistance don’t make the cut. “We looked at antibiotic resistance,” Karnofsky says. “What would a world without antibiotics look like? It’d look, like, not that bad. It’d look like the ‘40s or ‘50s. Most of the decline in bacterial diseases happened before we developed antibiotics at all, on any large scale. Most of it is a hygiene thing. Look, it’d really suck to lose antibiotics, a lot of people would die, but I don’t think it counts as a GCR for me.”

That’s not to say that plagues aren’t a major concern for Open Phil. On the contrary, “biosecurity,” or work to make it easier to prevent and control massive outbreaks of the kind that could end civilization, tops Open Phil’s list of GCRs. The risk is compounded by the advent of synthetic biology, a field that works on both creating artificial life and repurposing existing organisms for new ends.

But to count as a GCR, a pandemic would have to do more than kill merely millions of people. It’d need to kill enough people to threaten civilization as we know it, perhaps to such a degree that the survivors won’t be able to make it. “The way I’ve put it is a global disruption of civilization,” Karnofsky says. “Something that didn’t literally wipe out every single person but killed, like, 25 percent of the world’s population would be enormously destabilizing. Today we have this civilization that seems to be making some kind of progress. Based on my understanding of history, it’s very easy to not make that kind of progress.”

He has a point. The current post–Industrial Revolution era of steady economic growth and improving living standards is a gigantic historical aberration. Any shock that threatens to end it could make billions, if not trillions, of people worse off.

Open Phil’s position here is actually markedly less extreme than many effective altruists’. “There’s a certain set of people who basically care about global catastrophic risks because of the potential for existential risk,” Lempel, who manages Open Phil’s work on GCRs, says. “There’s an argument that goes something like, ‘An enormous proportion of all people who will ever live are potential people who will live in the future, and so all the utility that exists is in the future — so the difference between something really bad that winds up not making us go extinct is enormous relative to something that makes us go extinct, so we should only care about things that make us go extinct.’”

That view implies that biosecurity shouldn’t top the list. Massive outbreaks have the potential to do great harm, but they can’t kill everybody. “There are some people on submarines,” Lempel notes. “Those people are going to make it.” There are a couple of reasons Open Phil lists it anyway. For one thing, Lempel says, there’s not unanimity among the staff that future persons have heavy or perhaps even equal moral weight as people today. “Personally, speaking for Howie, and not speaking for GiveWell, I care about future generations a lot. That’s just my value set.” Lempel says. “I think that’s not a consensus at GiveWell. But when we weigh different values, it’s one value set that we are thinking about.”

Biosecurity also appeals because many of the philanthropic steps that will likely be involved — improving response to disease outbreaks in poor countries, increasing hospital sanitation, etc. — would be desirable even if you don’t care much about future people. And the fact that they’re implementable now also means they can be tested, increasing confidence in their ability to avert a future mass catastrophe and motivating continued funding.

”There’s a risk when you set something up that’s only used in case of a crisis,” Lempel says. “Some of the stuff you might think about for AI risk, for example, are things that might be used if there was a really malevolent AI that was developed. You could imagine setting up infrastructure to work on that, five years later nothing’s happened, and it loses its support.”

But Open Phil still has a ways to go before it starts making grants on the issue. “We do not feel that we have a strong sense of the interventions available to a new philanthropist in this field,” its cause evaluation concludes, “but we expect that most work would take the form of research and advocacy.” Biosecurity thus poses a very similar challenge as criminal justice reform and monetary policy. Estimating the magnitude of impact for a philanthropic intervention is difficult bordering on untenable.

What could go wrong

Open Phil will be in a research phase for a while, but soon it will need to start spending down Tuna and Moskovitz’s billions more rapidly.

”The world is getting better, and that means that giving opportunities now are better than they’re going to be 10 or 20 or 30 years from now, hopefully,” Tuna says. “The good you do today compounds over time.”

That suggests Good Ventures’ money needs to be distributed sooner rather than later. That’s a quicker time horizon than many foundations use, and allows for a relatively rapid test of what large-scale giving on effective altruist grounds would look like. If it works, it could prove hugely influential for other donors. Already other Silicon Valley entrepreneurs are signing on; Instagram co-founder Mike Krieger (who got an estimated $100 million from selling the company to Facebook) and his fiancée, Kaitlyn Trigger, have committed at least $750,000 to Open Phil, and Trigger is set to start working there part-time.

That means that at some point in the not-too-distant future, Open Phil will have to decide if criminal justice investments are a better bet than macroeconomic policy ones, and by how much; if macroeconomic policy investments are a better bet than biosecurity; and whether either is better than funding medical research. It will have to start comparing magnitudes — and that’s ridiculously difficult.

Think about what you’d need to know to do a really precise comparison of whether to invest in prison reform or in preventing geomagnetic storms. You’d need to know exactly how many people will be let out of prison due to your grant to the think tank you’re considering funding. You’d need to figure exactly how much worse life in prison is than life on the outside — is it half as good? three-quarters? — and what effect, in either direction, you’re having on crime.

You’d need to know the weather patterns of the sun for the next few centuries, in order to calculate the odds of a coronal mass ejection hitting us. You’d need to know exactly how much damage those ejections will do to our current grid. You’d need to know who needs to be paid what to make telecommunications and electrical systems robust against an ejection.

It’s an impossible task, and Open Phil admits as much. Because of the inherent difficulty in assigning numeric odds to everything from effects of policy investments to global catastrophic risks, it’s moved away from relying too heavily on quantification. “We’re excited about the project of making giving more analytical, more intellectual, and overall more rational,” Karnofsky once wrote. “At the same time, we have mixed feelings about the project of quantifying good accomplished: of converting the impacts of all gifts into ‘cost per life saved’ or ‘cost per DALY’ type figures that can then be directly compared to each other.”

A man receives a cash transfer in Jakarta, Indonesia.
A man receives a cash transfer in Jakarta, Indonesia.
Bay Ismoyo/AFP/Getty Images

So far, so good. But you still need to be able to compare rough magnitudes. I know we can’t quantitatively compare geomagnetic storms to criminal justice reform with any rigor. But I strongly suspect we can’t even qualitatively judge one to be more effective, either. The human brain can only process so much data. Six people can only process so much data.

Quantification can obscure more than it helps, but qualitative evaluations are prone to all kinds of cognitive biases, to subconscious emotions, to instinctive individual political leanings. One striking feature of Open Phil’s policy list is that it looks like my own personal public policy wish list. Maybe it shouldn’t. Maybe what’s going on is that people with a similar personality type are latching on to one set of causes and ignoring others that are equally good but less amenable to our temperaments. The less quantitative the process gets, the higher the potential for arbitrary factors to corrupt it.

This is especially true when most of the people in the room look alike. Out of GiveWell’s staff of 18, just three (plus Tuna) are women. There is not a single person of color on staff. [2018 update: this is no longer true.] When your stated purpose is to rank the world’s problems by importance and solvability, this really matters. Consciously or not, it influences your views on the importance of, say, women’s education in poor countries, or reducing police mistreatment of minorities, or even criminal justice reform, which did make it onto the list, but not on racial justice grounds.

”There’s a group of staff members with whom I’m almost daily sharing articles on issues that are totally separate from GiveWell issues, like identity politics,” research analyst Eliza Scheffler says. “We’ll often talk about gender.” But there’s no reason, Scheffler says, not to consider identity issues as causes worth addressing. She recalls seeing a talk by a gay activist from rural Kenya, who was forced to undergo numerous rounds of reparative “therapy.”

”His life sounds horrible,” Scheffler says. “I think we need to be able to try to compare that ... I worry that we’re missing out on negative utility, but I also feel pretty confident in the way we are approaching it.” She’s right. Cost-benefit analysis is not an inherently racist or sexist practice. But you need to be aware of costs to which your position in society might blind you.

”We’re not where we want to be when it comes to diversity, and we’re working on it,” Karnofsky says. “Based on accepted offers, our incoming class will represent a step forward on this front — about evenly split on gender with a significant minority of people of color — though we recognize that we still have a long way to go.”

The sheer difficulty of the process raises the question of whether Good Ventures should just give to the causes GiveWell has already identified, and whose effectiveness is much easier to measure. For example, GiveDirectly, one of GiveWell’s top charities, gives cash directly to poor people in Kenya and Uganda. Cash transfers to the poor are among the most-studied topics in development economics — and if that weren’t enough, the charity was subject to a randomized evaluation that found major positive results for families receiving the cash. GiveWell knows what it does, and knows that it works. [2018 update: a more recent followup study of GiveDirectly was more mixed.]

”The things that we found so far in Open Phil, some of them I’m really happy that we’ve been able to provide funding to,” GiveWell cofounder Elie Hassenfeld says. “But I don’t yet feel like, ‘Oh wow, this is so amazing that GiveDirectly should never get another dollar.’ If anything, the fact that we have done the Open Phil work for a year or two and haven’t found anything that is so amazing makes me feel better about GiveWell than I did two years ago.” Karnofsky has expressed similar thoughts.

Tuna says she doesn’t understand her colleagues’ pessimism. “I am still optimistic that we can do better than just giving money to poor people,” she says. “But in the meantime, we’re doing a lot of just giving money to poor people.”

Still, whether there’s something better than giving money to poor people may not be the hardest question Open Phil faces. The really hard question is how they’d know.

Future Perfect

Our love of orcas is making them miserable

Future Perfect

The first results from the world’s biggest basic income experiment

Future Perfect

Scaling slaughter-free meat is hard. Here’s one way to make it easier.

View all stories in Future Perfect