[My views are my own, not my employer's. Thanks to Michael Dickens for reviewing this post prior to publication.]
[More discussion here]
Summary
Spreading ethical offsetting is antithetical to EA values because it encourages people to focus on negating harm they personally cause rather than doing as much good as possible. Also, the most favored reference class for the offsets is rather vague and arbitrary.
There are a few positive aspects of using ethical offsets, and situations in which advocating ethical offsets may be effective.
Definition
Ethical offsetting is the practice of undoing harms caused by one's activities through donations or other acts of altruism. Examples of ethical offsetting include purchasing carbon offsets to make up for one’s carbon emissions and donating to animal charities to offset animal product consumption. More explanation and examples are available in this article.
Against offsetting
I think ethical offsetting is antithetical to EA values, and have three main objections to it.
1) In practice, people doing ethical offsetting use vague and arbitrary reference classes.
2) It's not the most effectively altruistic thing to do.
3) It spreads suboptimal and non-consequentialist memes/norms about doing good.
1) The reference class people pick for ethical offsets is arbitrary.
For example, let's say I cause some harm by buying milk that came from a cow that was treated poorly, and I want to negate the harm. I have a bunch of options.
I cannot undo the exact harm done by my purchase once it's happened, but I could (try to) seek out that specific cow and try to do something nice for her, negating the harm I caused for that specific cow's utility calculus. I could donate some money to a charity that helps cows, negating my harmful effect on the total utility of cow-kind. I could donate some money to a charity that helps all farmed animals, negating my harmful effect on farmed animal-kind. Or I could donate to whatever charity I thought did the most good per dollar, negating my negative impact on the universe most cost-effectively but less directly.
People seem to settle on a sort of broad cause-area-level offsetting preference (e.g. donating to help farmed animals). While reference class seems intuitive, it's ultimately arbitrary*.
2) Ethical offsetting isn't the most effectively altruistic thing.
You should do the things you think are most effectively altruistic, and you should donate to the charities you think are most effective. If you eat dead animals and don't believe animal charities are the most effective charities, I don't think you should donate to them.
Like everything else, ethical offsetting has opportunity costs; you could use that money to donate to the best charity, which is often different from the charity you’re using for ethical offsetting. It causes a harm relative to the world where you donate only to the most effective charity.
Even if you think the charity you donate your offsetting money to is the most effective, I don’t think it’s helpful to do ethical offsetting. Much of the suffering in the world isn’t directly caused by anyone, so an offsetting mindset increases the probability that you’ll miss big sources of suffering down the line. It causes a bias towards addressing anthropogenic harms, rather than harms from nature.
3) Ethical offsetting spreads anti-EA memes and norms
Ethical offsetting reinforces a preoccupation with not doing harmful things (instead of not allowing harmful things to happen, and taking action when they do). But EAs should (and usually do) focus on the sufferers, not themselves.
By encouraging others to offset, we set norms oriented around people’s personal behavior. We encourage an inefficient model of charity that involves donating based on one’s activities, not one’s abilities or the needs of charities that help neutralize various harms. We miss the chance to communicate about core EA ideas like cause prioritization and room for more funding by establishing a framework that has little room for them.
There are some other dangers involved in ethical offsetting, although I haven’t seen much evidence they actually occur: Offsetting may also encourage unhealthy scrupulosity about the harms we inevitably contribute to in order to function (although it could also help alleviate anxiety about them). And as Scott Alexander points out, offsetting could lead people to think it’s acceptable to do big harmful things as long as they offset them. This could contribute to careless and destructive norms about personal behavior.
Caveats
Offsetting is better than nothing. There may be situations in which ethical offsetting is the biggest plausible ask one can make. In such situations, I think bringing up the idea of ethical offsetting may be appropriate. And it may be an interesting conversation starter about sources of suffering and ways of alleviating them.
I've previously discussed my concerns about the obstacles to changing one's mind about cause prioritization, and I can imagine ethical offsetting at the cause area level being used to remind oneself about various causes of suffering in the world and the organizations working to stop them. This could make it easier to change one’s mind about what’s most effective. It seems somewhat plausible that offsetting would help make the community better at updating and better informed.
It may be really psychologically beneficial for some people, similar to the way donations for the dubiously-named fuzzies (donations for causes that are especially personally meaningful to the donor rather than maximally effective) sometimes are.
I think the argument that we should focus on doing lots of good rather than fixing harms we cause could drive destructive thoughtlessness about personal behavior, so I’m wary about making it too frequently. I’m most worried about this concern.
*The reference class schelling point is stronger with carbon offsets, where the harmful thing is adding some carbon dioxide to the atmosphere. Carbon dioxide molecules are pretty interchangeable. If you remove as many as you added, you neutralize the harm from your emissions-causing action very directly, which is intuitively appealing.
All suffering may be equally important, but not all forms of harm are the same, or even similar. How similar the harm you offset is to the harm you cause can vary a lot. Few other types of offsetting I’ve heard of allow the opportunity to create a future so similar to the one where the harmful activity had never been done.
I think "do as much good as possible" is not the best framing, since it means (for example) that an EA who eats at a restaurant is a bad EA, since they could have eaten ramen instead and donated the difference to charity. I think it's counterproductive to define this in terms of "well, I guess they failed at EA, but everyone fails at things, so that's fine"; a philosophy that says every human being is a failure and you should feel like a failure every time you fail to be superhuman doesn't seem very friendly (see also my response to Squark above).
My interpretation of EA is "devote a substantial fraction of your resources to doing good, and try to use them as effectively as possible". This interpretation is agnostic about what you do with the rest of your resources.
Consider the decision to become vegetarian. I don't think anybody would think of this as "anti-EA". However, it's not very efficient - if the calculations I've seen around are correct, then despite being a major life choice that seriously limits your food options, it's worth no more than a $5 - 50 donation to an animal charity. This isn't "the most effective thing" by any stretch of the imagination, so are EAs still allowed to do it? My argument would be yes - it's part of their personal morality that's not necessarily subsumed by EA, and it's not hurting EA, so why not?
I feel the same way about offsetting nonvegetarianism. It may not be the most effective thing any more than vegetarianism itself is, but it's part of some people's personal morality, and it's not hurting EA. Suppose people in fact spend $5 offsetting nonvegetarianism. If that $5 wasn't going to EA charity, it's not hurting EA for the person to give it to offsets instead of, I don't know, a new bike. If you criticize people for giving $5 in offsets, but not for any other non-charitable use of their money, then that's the fallacy in this comic: https://xkcd.com/871/
Let me put this another way. Suppose that somebody who feels bad about animal suffering is currently offsetting their meat intake, using money that they would not otherwise give to charity. What would you recommend to that person?
Recommending "stop offsetting and become vegetarian" results in a very significant decrease in their quality of life for the sake of gaining them an extra $5, which they spend on ice cream. Assuming they value not-being-vegetarian more than they value ice cream, this seems strictly worse.
Recommending "stop offsetting but don't become vegetarian" results in them donating $5 less to animal charities, buying an ice cream instead, and feeling a bit guilty. They feel worse (they prefer not feeling guilty to getting an ice cream), and animals suffer more. Again, this seems strictly worse.
The only thing that doesn't seem strictly worse is "stop offsetting and donate the $5 to a charity more effective than the animal charity you're giving it to now". But why should we be more concerned about making them give the money they're already using semi-efficiently to a more effective charity, as opposed to starting with the money they're spending on clothes or games or something, and having the money they're already spending pretty efficiently be the last thing we worry about redirecting?
Aren't you kind of not disagreeing at all here?
The way I understand it, Scott claims that using your non-EA money for ethical offsetting is orthogonal to EA because you wouldn't have used that money for EA anyway, and Claire claims that EAs suggesting ethical offsetting to people as an EA-thing to do is antithetical to EA because it's not the most effective thing to do (with your EA money).
The two claims don't seem incompatible with each other, unless I'm missing something.