Really enjoyed this piece. It is somewhat painful to read, given that I believe most of my professional life did more harm than good.
I do think that partially rationalizing torturing billions of sentient beings every year for more corn in silos in case of a nuclear winter - that's really a stretch.
Thanks for sharing, Matt!
I started following a plant-based diet roughly 4 years ago mostly due to finding out about the badness of factory-farming (and also because I think it mitigates global warming, and is healthier). Meanwhile, I have gotten confused about the overall impact of a plant-based diet, given the uncertain effects on wild animals and in the longterm. I think I continue plant-based because (descending order of importance):
This is a linkpost to the article Counterproductive Altruism: The Other Heavy Tail from Daniel Kokotajlo and Alexandra Oprea. Some excerpts are below. I also include a section at the end with some hot takes regarding possibly counterproductive altruism.
First, we argue that the appeal of effective altruism (henceforth, EA) depends significantly on a certain empirical premise we call the Heavy Tail Hypothesis (HTH), which characterizes the probability distribution of opportunities for doing good. Roughly, the HTH implies that the best causes, interventions, or charities produce orders of magnitude greater good than the average ones, constituting a substantial portion of the total amount of good caused by altruistic interventions. Next, we canvass arguments EAs have given for the existence of a positive (or “right”) heavy tail and argue that they can also apply in support of a negative (or “left”) heavy tail where counterproductive interventions do orders of magnitude more harm than ineffective or moderately harmful ones. Incorporating the other heavy tail of the distribution has important implications for the core activities of EA: effectiveness research, cause prioritization, and the assessment of altruistic interventions. It also informs the debate surrounding the institutional critique of EA.
Assume that the probability distribution of charitable interventions has a heavy-right tail (for example, like the power law described in the previous section). This means that your expectation about a possible new or unassessed charitable intervention should include the large values described above with a relatively high probability. It also means that existing charitable interventions whose effectiveness is known (or estimated with a high degree of certainty) will include interventions differing in effectiveness by orders of magnitude. We contend that this assumption justifies well-known aspects of EA practice such as (1) effectiveness research and cause prioritization, (2) “hits-based-giving,” and (3) skepticism about historical averages.
What if the probability distribution of altruistic interventions includes both a left and a right heavy tail? In this case, we cannot assume either that (1) one's altruistic interventions are expected to have at worst a value of zero (i.e. to be bounded on the left side) or (2) that the probability that a charitable intervention is counterproductive or harmful approaches zero very rapidly.
Many catastrophic interventions — whether altruistic or not — generate large amounts of (intentional or unintentional) harm. When someone in the world is engaging in an intervention that is likely to end up in the heavy left tail, there is a corresponding opportunity for us to do good by preventing them. This would itself represent an altruistic intervention in the heavy right tail (i.e. one responsible for enormous benefits). The existence of the heavy-left tail therefore provides even stronger justification for the prioritization research preferred by EAs.
Another conclusion we draw from the revised HTH is that the value of a class of interventions should be estimated by considering the worst as well as the best. Following such analysis, a class of interventions could turn out to be net-negative even if there are some very prominent positive examples and indeed even if almost all examples are positive. This sharply contradicts MacAskill's earlier claim that the value of a class of interventions can be approximated by the value of its best member.
If we are right about the existence of the left tail, certain interventions (even well-intentioned ones) are or can be expected to be extremely net-negative. Furthermore, even certain classes or subclasses of charitable interventions (e.g. foreign aid, food aid, or billionaire philanthropy) can be net-negative as a whole. In these cases, the most good an effective altruist can do may not be to launch new charitable ventures of her own or even to donate to the most effective charities. As noted above, the most efficient intervention might be to stop oneself or other people from launching massively negative interventions.
In this section, we begin by reconstructing three arguments in favor of a single right heavy tail that EAs have sketched: (i) the argument from examples of extreme values, (ii) the argument from nore systematic observational studies; and (iii) the argument from inefficient markets. For each of the arguments presented, we note that they should be extended to the existence of a heavy left tail.
Consider any big goal you wish to achieve—the sort of goal that would put your intervention far out in the right tail if you were to achieve it. There is some chance that the goal will be reached anyway without your effort, due to the effort of someone else. There is also a chance—perhaps a smaller chance, but a chance nonetheless—that your effort will cause an effective intervention not to happen or to be less effective than would have been the case without your action. For example, perhaps if you had not chosen to work towards this goal, someone more competent would have noticed the need and taken up the project in your absence. Thus, your choice to work on the project has a chance of backfiring, and if it does, it is a failure of the same magnitude as your success would have been.
Given that calculating the effectiveness of even a narrow range of philanthropic interventions we are considering [e.g. distributing anti-malaria bednets] typically involves multiplying together a large number of independent variables, we should expect the distribution of philanthropic interventions by effectiveness to be at least log-normal [which "are typical when data points are the product of many independent inputs"].
Without specific research into effectiveness, your uncertainty about how effective they are will range over many orders of magnitude. Lining them up side-by-side in your position of ignorance, they might look something like Figure 4 below.
Minimising downside is a common theme in effective altruism. However, I still found the article interesting as a reminder that heavy left/harmful tails are often neglected, and hidden behind the status quo. What looks robustly beneficial ignoring left tails might not be so once one accounts for them. In other words, left tails may conceal crucial considerations. Some hot takes (from me, not the article):
To be clear, I am not arguing for factory-farming, global warming, shorter lives, deforestation, and resistance to invasions, nor against remittances, and foreign aid. I am just trying to illustrate what is considered robustly beneficial may have a real chance of being harmful. Relatedly, there is the concept of complex cluelessness.
Less theoretical example: FWIW im not sold on 'more than anyone' but the top 2-3 current AI labs are all downstream of AI safety!
Great point! It does look like left tails are everywhere in the AI safety space.
Though you need to consider the counterfactual where the talent currently at OAI, DM, and Anthropic all work at Google or Meta and have way less of a safety culture.