Effective Altruism is about asking questions. It’s being judged by its loudest minority.

Months after I read an Effective Altruism book, I booked a flight from my diplomatic assignment in west Africa to California for the movement’s 2017 conference. The core ideas are powerful.

The movement is bigger than people realize. Much of it isn’t speaking up in this moment when the Trump Admin and the media is singling out EA and AI safety.

Most of EA looks nothing like the headlines

Here is some of what that thinking has done:

• It has saved about 340,000 lives, mostly young children in poor countries (via GiveWell). Its top charities hand out bednets, seasonal malaria medicine, and vitamin A, and pay families to vaccinate their babies.
• It was the first big funder of the YIMBY movement. Coefficient Giving backed YIMBY Law, YIMBY Action, and the national YIMBYtown conference when it wasn’t clear building more housing could catch on.
• It helped ban the pesticides people most often use to end their lives. In poor farming countries, swallowing pesticide is a common suicide method. In Sri Lanka, bans cut suicides by ~70%.
• It’s getting lead out of paint and turmeric. In low income countries, roughly 1 in 2 children have lead poisoning.

• It funded the final trial of the first cheap, highly effective malaria vaccine for kids.
• It pays for testing sewage to catch new diseases early. When the next pandemic starts, every day of warning counts!
• It’s helping bring germ-killing UV lights to market. Someday these air sanitizers could mean fewer colds and flus.
• It made giving poor people cash a gold-standard. USAID began testing some of its programs against simply giving poor people money.

EA asks: How many people does this help? Is anyone else already working on it? Would it happen anyway without our money? Does it actually work when you test it? EA’s shorthand for this is “important, neglected, and tractable.”

Those questions often lead to ideas that sound strange at first. Giving poor people cash and YIMBY seemed far fetched! As someone who was bewildered at how little people cared about the global poor, this felt like a missing puzzle piece.

This same willingness to calculate impact leads to odd-sounding research like what global scenarios might lead to permanent suffering, wild animal welfare, or how much insects feel. I’m not drawn to that work myself. But universities have strange research fields too, like someone studying 4th-century Germanic Iron Age. It’s important to the field built around it, and could uncover the next things humanity should know.

AI safety started with the same math

For years, AI safety was a small field doing technical work, like tracking how fast AI was improving. In 2023, I tracked which countries had AI rules.

As AI got better, AI safety got big and well funded. It built its own pipelines for bringing new people. Many join because they care about AI and its risks, not because of EA. In practice, AI safety is becoming a separate movement.

Most AI safety work is boring and invisible

Most people I know in AI safety work on concrete problems: computer chips, cybersecurity, nuclear risk, and making sure export controls aren’t circumvented. Others log AI failures, map cyberattacks, or study how models work and when they break. Many would rather not be called “AI safety”.

This narrow research doesn’t go viral, but it makes up much of Coefficient Giving’s AI grants. In my AI safety circles, if you ask how likely “doom” is, you get a nuanced answer about specific worries, like power grids, nuclear weapons, bioweapons, or the health of democracy.

The “doomers” are a loud minority

One wing of AI safety thinks AI is more than 50% likely to wipe out all humanity. That’s the view of the book If Anyone Builds It, Everyone Dies. This wing doesn’t speak for most AI safety. A cofounder of Effective Altruism, Will MacAskill, found the book and its arguments disappointing.

Some in that wing truly believe the odds are 50% or higher. But after many conversations, I think more of them see a real but smaller chance, but convey certainty to spur action.

That doesn’t feel grounded or honest to much of AI safety, or to the rest of EA.

The cost of urgency

Sometimes that urgency turns into pushback against people working on smaller risks. An attack on the power grid would be terrible, but it wouldn’t end humanity, so some treat it as less important. I got similar pushback when I explained why unskilled lone wolves won’t be able to build engineered pandemics. One reply to me said anything that makes biosecurity seem less urgent will lead to lives lost.

The beliefs are sincere but they lead to censorship, silencing, and intentionally overstating claims in search of impact.

That’s not . It’s not okay with a lot of people in EA. It also worries outsiders who want action on catastrophic risk. Many in AI safety agree, but speaking up could make them look disloyal or cost them funding. The only safe thing is to push urgency.

You can see this tension. An early OpenAI employee and prominent EA warned that EA needs more self-restraint, so it doesn’t accidentally cause harm. The former COO of Rethink Priorities, a major EA research organization, wrote a heartfelt post cautioning the community to be suspicious when the most impactful choice (AI work) happens to also be the best paid and highest status. He left AI policy to start an organization studying how farming affects insects and other small animals. He stuck to his own calculations despite people saying moving to the Bay Area, where the impact and money is.

In EA’s early days, people shared tips on living cheaply so they could give more away. Money and status weren’t really influences here. Now they are, along with urgency, politics, and echo chambers. All of these work against the core ideas.

Who is speaking up?

The people speaking up these months will skew toward those who . A large share of EA never did. The people working on malaria, vaccines, and lead are just as much a part of EA. Many of them disagree with 80,000 Hours, EA’s main career advice group, and its pivot to one-sided talk about AI risk. Many don’t even buy the idea that we can reliably impact people in the far future, which EA calls longtermism.

Partly because of these fractures, EA has no clear leader. Will MacAskill comes closest, but he appeals more to people who support the movement’s pivot to AI or the far future. The doomiest people don’t follow him either. He argues for less work on avoiding catastrophe and more on building a good future.

What holds EA together

EA was never meant to be a list of answers. It’s a way of choosing what to work on. Bednets, YIMBY grants, and AI safety are where those questions led at different moments to different people.

Even as EA’s biggest institutions drifted toward AI, some people are pulling back to the core questions with new organizations and books. The non-AI work didn’t stop: an EA nonprofit incubator keeps launching new charities on drug-resistant infections, global health, and development.

By nature this movement struggles when defending itself, or anything apart from its core.

Maybe the core is the only thing we all agree on: ask what’s important, neglected, and tractable, stay open to weird ideas, actively seek to be proven wrong, and put humility and honesty above everything else.

That’s worth protecting and I will always defend that. I’d fly across the world for it in every universe.

Positive Sum is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.

I’m testing out ads. This is a job ad from a nonprofit I like.

Egg farms kill male chicks right after they hatch, because males don’t lay eggs. Innovate Animal Ag helped bring a technology to the US that checks if an egg is male before it hatches, so those chicks are instead never born. I am a big fan of this project.

Now they are working on building systems to help commercial farms monitor tools and medicines that help improve the lives of farm animals. They’re looking for a technical generalist to lead the designing and building of this system that integrates computer vision, hardware, and machine learning. No need for prior animal-related experience.

The title is Head of Instrumentation & AI. Click here to apply or refer people.