17da95c73e3d7b0c6728d11117fd10b6
Things I do:
Research agenda: bit.ly/artificial-moral-progress (Technical research + strategy/governance exploration to prevent value lock-in; keen to discuss collaborations)
Previously co-led AI Alignment: A Comprehensive Survey
My causes:
AI risk (s/x-risk, harms to nonhumans, socio-economic, …)
EA epistemic health (looking for collab on epistemic health infrastructure: bit.ly/website-EHI)
Animal advocacy
Interesting idea, thanks for doing this! I agree it’s good to have more approachable cause prioritization models, but there’re also associated risks to be careful about:
A widely used model that is not frequently updated could do a lot of damage by spreading outdated views. Unlike large collections of articles, a simple model in a graphic form can be spread really fast, and once it’s spread out on the Internet it can’t be taken back.
A model made by a few individuals or some central organisation may run the risk of deviating from the view of majority EAs; instead a more “democratic” way (not too sure what this means exactly) of making the model might be favored.
Views in EA are really diverse, so one single model likely cannot capture all of them.
Also, I think the decision-tree-style framework used here has some inherent drawbacks:
It’s unclear what “yes” and “no” means.
e.g. What does it mean to agree that “humans have special status”? This can be refering to many different positions (see below for examples) which probably lead to vastly different conclusions.
a. humans have two times higher moral weight than non-humans
b. all animals are morally weighted by their neuron count (or some non-linear function of neuron count)
c. human utility always trumps non-human utility
for another example, see alexrjl’s comment.
Yes-or-no answers usually don’t serve as necessary and sufficient conditions.
e.g. I think “most influential time in future” is neither necessary nor sufficient for prioritizing “investing for the future”.
e.g. I don’t think the combined condition “suffering-focused OR adding people is neutral OR future pessimism” serves as anything close to a necessary condition to prioritizing “improving quality of future”.
A more powerful framework than decision trees might be favored, though I’m not sure what a better alternative would be. One might want to look at ML models for candidates, but one thing to note is that there’s likely a tradeoff between expressiveness and interprettability.
And lastly:
In addition, some foundational assumptions common to EA are made, including a consequentialist view of ethics in which wellbeing is what has intrinsic value.
I think there have been some discussions going on about EA decoupling with consequantialism, which I consider worthy. Might be good to include non-consequentialist considerations too.
Four podcasts on animal advocacy that I recommend:
Freedom of Species (part of 3CR radio station)
Covers a wide range of topics relevant to animal advocacy, from protest campaigns to wild animal suffering to VR. More of its episodes are on the “protest campaigns” end which is less popular in EA, but I think it’s good to have an alternative perspective, if only for some diversification.
Knowing Animals (hosted by Josh Milburn)
An academic-leaning podcast that focuses on Critical Animal Studies, which IMO is like the academic equivalent of animal advocacy. Most guests are academics in philosophy, humanities and social sciences. (and btw, one episode discussed wild animal suffering, and I liked that episode quite a lot)
The Sentience Institute Podcast
EA-aligned. Covers topics ranging from alt proteins to animal-focused impact investing to local animal advocacy groups to digital sentience.
Animal Rights: The Abolitionist Approach Commentary (by Gary L. Francione)
A valuable perspective that’s not commonly seen in EA. Recommended for diversification.
Off-topic: I also recommend the Nonlinear Library podcasts; they turn posts on EA Forum and other adjacent forums (LW, AF) to audio. There’re different versions that form a series, including a version containing all-time top posts of EA Forum. There’s also a version containing the latest posts meeting a not-very-high karma bar—I use that version to keep track of EA news, and it saved me a lot of time.
I’d also like to reiterate the arguments Larry Temkin gave against international aid, since the post doesn’t cover them. I’m not sure if I’m convinced by these arguments, but I do find them reasonable and worth serious consideration.
Opportunity cost of local human resources: International aid agencies tend to hire competent local people in the country they operate in (e.g. Sub-Saharan African countries), but these competent people could otherwise serve in important roles for the development of the local society.
Corruption: Lots of international aid funds are corrupted by government officials and local thugs (and such corruption is often covered up). Often, the corrupted funds then go on to fuel the agenda of thugs or malevolent political actors, which is harmful to the people of the aided nation.
In addition to the two empirical arguments above, he also gave normative arguments, but those arguments are commonly raised and rather well-known by EAs, so I don’t repeat them here.
Hypothesis: in the face of cluelessness caused by flow-through effects, “paving path for future progress” may be a robust benefit of altruistic actions.
Epistemic status: off-the-cuff thoughts, highly uncertain, a hypothesis instead of a conclusion
(In this short-form I will assume a consequentialist perspective.)
Take slavery abolition as an example. The abolition of slavery seems obviously positive at the object level. But when we take into account second-order effects, things become less clear (e.g. the meater-eater problem). However, I think the bad second-order effects (if any) can plausibly be outweighed by one big second-order benefit: that the abolition of slavery paves the way for future moral progress, including (but not limited to) those around our treatment of animals. For example, it seems likely to me that in a world with slavery, it would be much harder to advocate for the rights of human minorities, of animals, and of digital sentience.
I guess this applies to many other cases too, including cases irrelevant to moral progress but relevant to some other kind of progress. This hypothesis might not change how we act by much, as we usually tend to ignore hard-to-evaluate second-order effects. This hypothesis may provide a reason why sometimes an action is justified despite of seemingly negative second-order effects, but I also worry that it may be abused, as a rationalization for ignoring flow-through effects.
Thank you for this critique!
Just want to highlight one thing: comments to this post are sometimes a bit harsh, but please don’t take this to mean we’re unwelcoming or defensive (although there may be a real tendency to overly argue for ourselves). The style of discussion on the forum is sometimes just like this :)
Summary: This is a slightly steelmanned version of an argument for creating a mass social movement as an effective intervention for animal advocacy (which I think is neglected by EA animal advocacy), based on a talk by people at Animal Think Tank. (Vote on my comment below to indicate if you think it’s worth expanding into a top-level post)
link to the talk; alternative version with clearer audio, whose contents—I guess—are similar, but I’m not sure. (This shortform doesn’t cover all content of the talk, and has likely misinterpreted something in the talk; I recommend you to listen to the full talk)
Epistemic status: An attempt at steelmaning the arguments, though I didn’t really try hard—I just wrote down some arguments that occur to me.
The claim: Creating a mass social movement around animals, is more effective than top-to-bottom interventions (e.g. policy) and other interventions like vegan advocacy, at least on current margins.
This is not to say policy work isn’t important. Just that it comes into the picture later.
My impression is that the track record of mass movements in creating change is no less impressive than that of policy reforms, but EA seems to have completely neglected the former.
A model of mass movements:
Analogous to historic movements like the civil rights movement in the US, and recent movements like Extinction Rebellion. Both examples underwent exponential growth, which will be explained in the next bullet point.
You start with a pool of people in the movement, and these people go out and try to grab attention for the movement, using tactics like civil disobedience and protests. Exposure to the ideas leads to more people thinking about them, which in turn leads to more people joining. With the enlarged people pool, you start the cycle again. This then leads to an exponentially growing pool.
After the movement is large enough and has enough influence, policy reforms and other interventions aimed at the top of society will become viable.
Research showed that few, if any, movements failed after reaching a size threshold of 3.5% of the entire population.
Many movements died down because their base number of exponentiation is smaller than 1, but successful movements can have much higher base number.
Other interventions like vegan outreach and policy work may also have similar exponential growth, but it’s plausible that their base numbers are much less likely to be >1 (or to be very high) when compared with mass social movements.
Strategies for mass movements:
Strategy is super important!
Start from your ultimate goal (e.g. stop animal exploitation), and then set milestones for achieving this goal, and then design concrete actions and campaigns in service of milestones.
One key point is “escalation”—how to make the movement grow exponentially starting from the initial pool
You need to be momentum-driven: convert the attention you get to new movement members, seize more attention with your enlarged membership, and repeat the cycle
You need to force people in the general public to take sides, possibly by non-violent disruptions and making salient sacrifices (e.g. arrests)
You may need to show concrete demands (rather than abstract ones) that resonate with people
Another key point is “absorption”—when large numbers of new members join the movement, how to rapidly and effectively absorb them
Decentralized movements can absorb more rapidly. (e.g. people trained can go off independently and train other new people)
There’s no silver bullet; we still need deep thinking and discussions and coordination to guide our strategy.
Do you think it’s worth expanding into a top-level post? Please vote on my comment below.
Don’t know how to use Airtable, but a quick googling led me to this. The last reply (by kuovonne) in the linked thread seems useful.
I’m excited to see this post, thank you for it!
I also think much more exploration and/or concrete work needs to be done in this “EA+AI+animals” (perhaps also non-humans other than animals) direction, which (I vaguely speculate) may extend far beyond the vicinity of the Project CETI example that you gave. Up till now, this direction seems almost completely neglected.
Hi Martyna, maybe this post and its comments can interest you.
Also, something else that comes to mind: Andrew Critch thinks that working on Human-Robot Interaction may be very useful to AI Safety. Note that he isn’t solely talking about robots, but also human-machine interaction in general (that’s how I interpret it; I may well be wrong):
HRI research is concerned with designing and optimizing patterns of interaction between humans and machines—usually actual physical robots, but not always.
Not sure whether other AI Safety researchers would agree on the importance of HRI, and not sure if your current career path is very relevant to this. Anyway, just sharing something that might be useful :)
Comment from author: Note that I lean slightly towards the term “animal advocacy”, so it’s possible that my analysis contains a slight bias towards this term.
I agree that the second- and third-order effects of e.g. donating to super-effective animal advocacy charities are, more likely than not, larger than those of e.g. volunteering at local animal shelters. (though that may depend on the exact charity you’re donating to?)
However, it’s likely that some other action has even larger second- and third-order effects than donating to top charities—after all, most (though not all) of these charities are optimizing for first-order effects, rather than the second- and third-order ones.
Therefore, it’s not obviously justifiable to simply ignore second- and third-order effects in our analysis.
Thanks for the post, it’s really exciting!
One very minor point:
In China, tofu is a symbol of poverty—a relic from when ordinary people couldn’t afford meat. As such, ordering tofu for guests is often seen as cheap and disrespectful.
I agree that this is somewhat true, but stating it like this seems a bit unfair. Ordering tofu for guests seems fine to me; It only gets problematic when you order way too much of it—in the same way as ordering nothing but rice for guests is extremely disrespectful. (Conflict of Interest: I’m a tofu lover!)
Anyway, I really like your idea! Good luck :)
Thanks for the answers, they all make sense and upvoted all of them :)
So for a brief summary:
The action that I described in the question is far from optimal under EV framework (CarlShulman & Brian_Tomasik), and
Even it is optimal, a utilitarian may still have ethical reasons to reject it, if he or she:
endorses some kind of non-traditional utilitarianism, most notably SFE (TimothyChan); or
considers the uncertainty involved to be moral (instead of factual) uncertainty (Brian_Tomasik).
Apologies for posting four shortforms in a row. I accumulated quite a few ideas in recent days, and I poured them all out.
Summary: When exploring/prioritizing causes and interventions, EA might be neglecting alternative future scenarios, especially along dimensions orthogonal to popular EA topics. We may need to consider causes/interventions that specifically target alternative futures, as well as add a “robustness across future worlds” dimension to the ITN framework.
Epistemic status: low confidence
In cause/intervention exploration, evaluation and prioritization, EA might be neglecting alternative future scenarios, e.g.
alternative scenarios of the natural environment: If the future world experienced severe climate change or environmental degradation (which has serious downstream socioeconomic effects), what are the most effective interventions now to positively influence such a world?
alternative scenarios of social forms: If the future world isn’t a capitalist world, or is different from the current world in some other important aspect, what are the most effective interventions now to positively influence such a world?
...
This is not about pushing for certain futures to realize. Instead, it’s about what to do given that future. Therefore, arguments against pushing for certain futures (e.g. low neglectedness) do not apply.
For example, an EA might de-prioritize pushing for future X due to its low neglectedness, but if they think X has a non-trivial probability to realize, and its realization has rich implications for cause/intervention prioritization, then whenever doing prioritization, they need to think about “what I should do in a world where X would be realized”. This could mean:
finding causes/interventions that are robustly impactful across future scenarios, or
finding causes/interventions that specifically target future X.
In theory, the consideration of alternative futures should be captured by the ITN framework, but in practice it’s usually not. Therefore it could be valuable to add one more dimension to the ITN framework: “robustness across future worlds”.
Also, there’re different dimensions on which futures can differ. EA tends to have already considered the dimensions that are related to EA topics (e.g. which trajectory of AI is actualized), but tends to ignore the dimensions that aren’t. But this is unreasonable, as EA-topic-related dimensions aren’t necessarily the dimensions in which futures have the largest variance.
Finally, note that in some future worlds, it’s easier to have high altruistic impact than in other worlds. For example in a capitalist world, altruists seem to be at quite a disadvantage to profit-seekers; in some alternative social forms, altruism plausibly becomes much easier and more impactful, while in some other social forms, it may become even harder. In such cases, we may want to prioritize the futures that have the most potential for current altruistic interventions.
Statement: This shortform is worth expanding into a top-level post.
Please cast upvote/downvote on this comment to indicate agreement/disagreement to the above statement. Please don’t hesitate to cast downvotes.
If you think it’s valuable, it’ll be really great if you are willing to write this post, as I likely won’t have time to do that. Please reach out if you’re interested—I’d be happy to help by providing feedback etc., though I’m no expert on this topic.
Proposal: I think building epistemic health infrastructure is currently the most effective way to improve EA epistemic health, and is the biggest gap in EA epistemics.
My definition of epistemic health infrastructure: a social, digital, or organizational structure that provides systematic safeguards against one or more epistemic health issues, by regulating some aspect of the intellectual processes within the community.
They can have different forms (social, digital, organizational, and more) or different focuses (individual epistemics, group epistemology, and more), but one thing that unites them is that they’re reliable structures and systems, rather than ad hoc patches.
(note: in order to keep the shortform short I tried to be curt when writing the content below; as a result the tone may come out harsher than I intended)
We talk a lot about epistemic health, but have massively underinvested in infrastructure that safeguards epistemic health. While things like EA Forum and EAG and social circles at EA hubs are effective at spreading information and communicating ideas, to my knowledge there has been no systematic attempt at understanding (and subsequently improving) how they affect epistemic health.
Examples of things not-currently-existing that I consider epistemic health infrastructure:
(Not saying that these are the most valuable ones; just that they fall into this category. They are examples and are only examples.)
mechanisms to poll/aggregate community opinions (e.g. a more systematized version of Nathan Young’s Polis polls) on all kinds of important topics, with reliable mechanisms to execute actions according to poll results
something like the CEA community health team, but focusing on epistemic health, with better-defined duties and workflows, and with better transparency
EA forum features and (sorting/recommending/etc.) algorithms aiming to minimize groupthink/information cascades
(Many proposals from other community members about improving community epistemic health also fall into this category. I won’t repeat them here.)
I plan to coordinate a discussion/brainstorming on this topic among people with relevant interests. Please do PM me if you’re interested!