Effective Altruism: When Doing “Good” is No Good

Effective altruism began with a sensible question: if you are going to help people, why not try to do as much good as possible with the resources you have? From that simple idea grew a movement, a network of wealthy donors and institutions, and eventually a philosophy extending from malaria prevention to the possible extinction of humanity through artificial intelligence.

The original case was straightforward. Money and time are limited, so charitable decisions involve trade-offs. If $1,000 spent on one program produces little measurable benefit while the same amount spent elsewhere can prevent serious illness or save lives, there is a strong reason to prefer the second program.

Peter Singer's famous drowning-child argument supplied some of the philosophical background. If you would ruin an expensive pair of shoes to save a drowning child directly in front of you, why should geographical distance greatly reduce your obligation to save a child elsewhere? Organisations associated with effective altruism attempted to turn that intuition into practical decisions. GiveWell compared charities by cost effectiveness. Giving What We Can encouraged people to pledge part of their incomes. 80,000 Hours applied similar reasoning to career choices.

The early results were often impressive. Malaria prevention, vitamin A supplementation, direct cash transfers and other global-health interventions could be compared using reasonably good empirical evidence. Effective altruism forced charities to confront a question that should always have been asked: what does the money actually achieve?

That remains the strongest part of the movement. There is nothing objectionable about comparing interventions, measuring results and directing resources towards programs that demonstrably work. Philanthropy should not be exempt from evidence merely because its intentions are good. The difficulty begins when this useful principle becomes a complete moral philosophy.

Effective altruism places considerable weight on impartiality. A human life should not count for less merely because that person lives thousands of kilometres away. But the apparently simple principle that everyone should count equally contains a substantial moral assumption. Human beings ordinarily believe they have special responsibilities towards their children, families, neighbours and political communities. Effective altruism tends to treat these attachments as biases that should not outweigh greater benefits available elsewhere.

Sometimes that criticism may be justified. There is no obvious reason why suffering becomes morally unimportant simply because it occurs on another continent. But it does not follow that all special obligations can be reduced to an impartial calculation of total welfare. The question is not merely what good I could possibly produce, but what responsibilities I actually have.

The problem became clearer with "earning to give." Instead of working directly for a charitable organisation, an effective altruist might pursue a highly paid career and donate a substantial proportion of the income. In principle, this could produce far more benefit than personally working for a charity.

But it also exposed a weakness in the philosophy. If outcomes dominate the calculation, questionable means can become easier to excuse when sufficiently large benefits are expected at the end. The collapse of FTX therefore damaged effective altruism for more than reputational reasons. Sam Bankman-Fried had publicly associated himself with the movement and with earning to give. His case raised an old objection to strongly consequentialist moral theories: what prevents a sufficiently valuable expected outcome from being used to excuse conduct that should never have been permitted in the first place?

Longtermism pushed the problem further. If future human beings count morally, and the future could contain enormous numbers of them, then preventing human extinction might outweigh almost every present concern. Even a very small probability of catastrophe can generate an enormous expected loss when multiplied by vast numbers of possible future lives.

The arithmetic can be performed. The difficulty lies in deciding what should go into it. How should extremely uncertain probabilities be estimated? How much moral weight should be given to people who do not yet exist? Is failing to create a possible future person equivalent to the death of an existing person? How should present suffering be weighed against hypothetical suffering centuries from now? Small changes in assumptions can produce very different conclusions. The mathematics therefore does not settle the philosophy. The philosophical assumptions determine what the mathematics is being asked to calculate.

This became especially important with artificial intelligence. Effective-altruist money helped develop AI-safety research at a time when comparatively few people were thinking seriously about catastrophic risks from advanced AI. That may prove to have been valuable. Rapidly advancing technology capable of affecting entire societies plainly deserves serious safety research.

But accepting the need for AI-safety research does not require accepting every estimate of existential risk. Probabilities attached to unprecedented future events are extremely difficult to calibrate. A claim that there is a 10 per cent chance of AI causing human extinction may look quantitative, but the presence of a percentage sign does not make the estimate equivalent to the probability calculations used in mature empirical sciences.

This exposes a broader weakness in effective altruism. Measurement naturally favours things that can be measured. Malaria prevention can produce reasonably concrete figures: treatments delivered, infections prevented and mortality reduced. Other social goods are much harder to place in a spreadsheet. Strong institutions, social trust, family obligations, political legitimacy, cultural continuity and human dignity resist simple comparison. That does not make them unreal.

Once importance, tractability and neglectedness become dominant criteria for allocating resources, the method can begin determining what counts as morally significant. Problems that fit the model rise to the top. Problems that resist quantification can disappear from view.

The same difficulty applies to effective altruism's confidence in rational self-correction. The movement has often emphasised openness to evidence and willingness to change one's mind. Those are admirable principles. But effective altruism also developed its own institutions, funding networks, career paths and specialist vocabulary. Like every intellectual movement, it therefore acquired incentives that could influence which problems were regarded as important and which answers were considered reasonable. No methodology can simply assume that its own criteria are neutral.

This is why criticism of effective altruism should distinguish between its different parts. It would be foolish to dismiss careful evaluation of malaria programs merely because one disagrees with longtermism or particular claims about artificial intelligence. Conversely, the demonstrated success of evidence-based global-health programs does not validate every philosophical extension of the effective-altruist framework.

The political reaction against effective altruism can make the same mistake in reverse. Treating malaria charities, AI-risk researchers, wealthy technology donors and longtermist philosophers as one political faction obscures important differences between them. A method for comparing charities is not identical to a theory about human extinction, even if both developed within the same movement.

What, then, should survive? The practical core is worth keeping. Compare interventions. Demand evidence. Publish assumptions. Admit uncertainty. Revise programs when evidence shows that they do not work. Ask what another million dollars would actually achieve rather than assuming that every worthy cause uses money equally well.

What should be rejected is the assumption that everything morally important can ultimately be reduced to the same optimisation problem. Human beings make choices under scarcity, and consequences matter enormously. But consequences are not the whole of morality. Duties, relationships, rights, institutions and limits on permissible conduct matter as well.

Effective altruism began by asking philanthropy a question it badly needed to hear: if you genuinely want to help, why aren't you trying to achieve the greatest benefit with the resources available? That remains an excellent question. The mistake was assuming that it was the only one.

https://80000hours.org/2020/08/misconceptions-effective-altruism/