I am a law professor. I teach and write mostly about AI governance, climate change, torts, and law & economics.
Gabriel Weil
Karma: 190
“And the truth is, as even Gabriel Weil accepts, it’s super unlikely that the training and deployment of highly capable advanced AI would fit the longstanding legal definition of an “abnormally dangerous activity”.”
This is not an accurate characterization of my views. Here’s a relevant quote from the linked paper:
“Training and deploying frontier AI models are clearly not activities of common usage, at least with current technology, given the enormous computational resource requirements of these systems.117 This means that, under the Restatement Third’s test, the key issue is whether training or deploying advanced AI systems satisfies the first criterion of creating a foreseeable and highly significant risk of harm even when reasonable care is exercised. This question is likely to be controversial. While I think the available evidence does support the conclusion that reasonable care may be insufficient to reduce the risk of catastrophic AI misalignment or misuse to below levels that would qualify as “highly significant,” recognizing any software development project as abnormally dangerous would represent a substantial doctrinal innovation.”
In the following paragraph, I further clarify: “To be clear, I do think an accurate understanding of the risks of advanced AI systems supports strict liability for the training and deployment of advanced AI systems, but it is important to recognize that this is not the likely outcome of a mechanical application of existing doctrine to harms caused by these systems.”
In a later section of the paper, I say “If courts are persuaded by the arguments summarized in part I that advanced AI systems do indeed pose a significant risk of causing human extinction, they should recognize training and deploying those systems as an abnormally dangerous activity. To be sure, treating software development as abnormally dangerous would represent a significant extension of existing strict liability, but it is entirely consistent with the doctrinal rationale for the abnormally dangerous activities category.”
I don’t see how you can read me as accepting that “it’s super unlikely that the training and deployment of highly capable advanced AI would fit the longstanding legal definition of an “abnormally dangerous activity”.”
On key point 5, I was offering to help secure funding, not asking for help.
How Technical AI Safety Researchers Can Help Implement Punitive Damages to Mitigate Catastrophic AI Risk
Indeed, this post is primarily addressing U.S. tort law. I should have made that clearer.
That said, if you want to see more, I’d direct you to the full draft law review article, which is available on SSRN: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4694006
Tort Law Can Play an Important Role in Mitigating AI Risk
If people suddenly stopped eating meat and other animal products, there would be a costly and economically disruptive unwinding of the industry that, in the best case, would still entail substantial short-term animal suffering. Given the political muscle of the animal agriculture industry in many countries, there would also likely be substantial efforts to prop up the industry, even in the face of a collapse in demand. Meanwhile, the past suffering of animals, which is most analogous to already-emitted carbon, could not be undone.
It’s true that climate would not halt if we achieved net-zero emissions, but there’s less “momentum” in the climate system than most laypeople assume. Absent further steps, it would take a long time for the climate to be restored to its pre-industrial condition, but the vast majority of the expected future climate-induced harms would be avoided. In fact, a rapid return to pre-industrial temperatures would be a bad thing, such much of the harm associated with climate change flows the pace of change.
“but it seems to me that this would be way more costly for individuals than giving up on meat”
This isn’t obvious to me. First, I think it’s likely that many vegans underrate the costs that going vegan would impose on many non-vegans. Second, even going vegan doesn’t completely eliminate one’s individual animal suffering footprint, since there are non-dietary channels in which individual consumption causes animal suffering. Going full Jain seems like it would be comparably burdensome to completely eliminating one’s gross carbon footprint.
I do think I agree that rapid decarbonization through individual consumption decisions would be more economically costly than rapid phaseout of animal agriculture, but I think that says more about the short-term tractability of the two problems than it says about the role of individual consumption in solving them. In any case, neither problem is, as a practical matter, going to be solved primarily via individual consumption choices, so I’m not sure too much hinges on this question. Locally, the effects are quite comparable.
I directly address this objection in the essay (it’s the first if of my “other objections and responses”). Do you mind spelling out why you found this unconvincing?
I’m not sure this is a fair comparison if what you want is an apples-to-apples comparative cost-effectiveness analysis. Not eating chicken is the lowest-hanging fruit in terms of personal sacrifice for animal welfare. Switching from eating beef to beans for 3 days would give you orders of magnitude less bang for sacrifice in animal welfare terms. Conversely, going without AC is hardly the most cost-effective personal sacrifice for climate change mitigation. For that, I’d pick something like biking to work instead of driving (the magnitude of the sacrifice is going to vary a lot from person to person; I really like biking, so it’s not much of a sacrifice to me) or taking a train instead of flying, when the latter would be more convenient.
It also seems strange to only count the mortality costs of climate change. This seems especially striking given that my understanding is that it isn’t really the mortality that’s the core issue in farm animal welfare since the relevant counterfactual in the going-vegan scenario is fewer farm animals being born, to begin with.
Finally, I just want to emphasize that I’m really trying to bracket the question of cause importance. I really don’t feel like I have a handle on how much negative welfare a chicken on a factory farm experiences. Climate damages are somewhat easier to value, but there is still a pretty wide dispersal in estimates of the social cost of carbon. However, I think you can take my core point on board regardless of your view on the relative importance of the two causes. That is, the role of the contributions that individuals can make to solving each problem are roughly analogous. Now, maybe you think that farm animal welfare is just so much more important as a cause area than climate change mitigation that it justifies the discrepancy in EA behavior, but then it’s at least worth being explicit about that. Given that low neglectedness is the main factor typically cited by EAs for emphasizing climate change and that there are not significant diminishing marginal returns (on the extensive margin) to individual sacrifice, that itself would require an update in what I understand to be the EA consensus.
“It bakes in and begs the question that climate change is similar to other EA causes—but it is not through the ITN framework that EAs use.”
I specifically tried to bracket the question of cause importance for a few reasons. I don’t think I have any particular expertise or insight on how to value animal suffering. My core point about the similarity in the potential for impact via personal consumption choices in both domains is analytically distinct from the importance of the cause areas. And I didn’t want agreement or disagreement with my post to hinge on readers own views about the relative importance of the two causes.
“It lampshades this lack of argument by mentioning neglectedness, and then appealing to uncertainty, but that’s not nearly enough treatment for this critical claim.”
I emphasized neglectedness this (though I did also briefly discuss importance and tractability, because lack of neglectedness is typically the feature of climate change mitigation that EAs rely most heavily on in deprioritizing climate change relative to other cause areas. In light of that, I think it’s important to note that (a) existing investments in climate change mitigation mostly do not come in the form of personal consumption decisions; and (b) there is little reason to expect diminishing marginal returns to more people taking some steps to minimize their carbon footprints (that is, on the extensive margin; on the intensive margin, or individual people making further investments in personal consumption decarbonization, decreasing marginal returns are to be expected).
In the typical case of meat-eating, the animal suffering has already happened by the time you choose to consume the product of it. At the point of consumption, there’s nothing you can do to change that. Forgoing meat consumption merely reduces the expected future quantity of animal suffering. Paying someone else to reduce their meat consumption by an equivalent or greater amount does the same thing.
Yep, I’m just saying that the equilibrium change in the quantity consumed will be less than the individual’s foregone consumption, not that it will be zero. How much less depends on the elasticities of supply and demand.
3 votes
Overall karma indicates overall quality.
Total points: 1
Agreement karma indicates agreement, separate from overall quality.
Yes. If the risks are insurable (and more so if insured), then we don’t need punitive damages to mitigate them. We can rely on the prospect of compensatory damages in the event those risks are realized. Punitive damages are only needed to protect against the risk of practically non-compensable harms, for which a compensatory damages award would not be practically enforceable.