Animal advocacy researcher. Organizer for Effective Altruism Waterloo. Former animal welfare research intern at Charity Entrepreneurship and research fellow at Animal Charity Evaluators.

Background mostly in pure math and machine learning, but also some in statistics/econometrics and agricultural economics. Earned to give in deep learning at a startup for 2 years for animal charities, and now trying to get into effective animal advocacy research. Curious about s-risks.

Suffering-focused, anti-speciesist, prioritarian, consequentialist. Also, I like math and ethics.

My shortform.

Wiki Contributions


DeepMind: Generally capable agents emerge from open-ended play

It seems like this could extend naturally to cooperative inverse reinforcement learning.  Basically, the real world is a new game the AI has to play, and humans decide the reward subjectively (rather than with some explicit rule). The AI has developed some general competence beforehand by playing games, but it has to learn the new rules in the real world, which are not explicit.

EA Forum feature suggestion thread

You can strong downvote on a "open listing" tag to try to get it removed from a post, and then just add a "closed listing" tag. I think once the tag score drops to 0, it gets removed.

DeepMind: Generally capable agents emerge from open-ended play

For what it's worth, I've mostly not been interested in AI safety/alignment (and am still mostly not), but this also seems like a pretty big deal to me. I haven't actually read the details, but this is basically not "narrow" AI anymore, right?

I guess the expressions "narrow" and "general" are a bit unfortunate, since I don't really want to call this either. I would want to reserve the term AGI for AI that can do at least this, but can also reason generally and abstractly, and excels at one-shot learning (although there are specific networks designed for one-shot learning, like Siamese networks. Actually, why aren't similar networks used more often,even as subnetworks?).

Can a Vegan Diet Be Healthy? A Literature Review

Across all studies, there was no evidence to support a causal relation between the consumption or avoidance of meat and any psychological outcomes. However, three studies provided evidence suggesting (contradictory) temporal relations between meat-abstention and depression and anxiety. Michalak, Zhang, and Jacobi (2012) demonstrated that the mean age at the adoption of meat-abstention (30.58 years) was substantially older than the mean age of the onset of metal disorder (24.69 years). These authors posited that mental disorders may lead to the adoption of a meat-less diet. The authors stated that individuals with mental disorders may “choose a vegetarian diet as a form of safety or self-protective behavior” (Michalak, Zhang, and Jacobi 2012, 6) due to the perception that plant-based diets are more healthful or because individuals with mental disorders may be “more aware of suffering of animals” (Michalak, Zhang, and Jacobi 2012, 2). Interestingly, these investigators also found that people with a lifetime diagnosis of psychological disorders consumed less fish and fast food. While these results conflict with previous research on fast food and mental health (Crawford et al. 2011), they support Matta et al.’s results and hypothesis that the exclusion of any food group, and especially meat and poultry, is associated with increased odds of having symptoms of psychological disorders (Matta et al. 2018).

Conversely, in their longitudinal analysis, Lavallee et al. (2019) found that meat-abstention was linked to “slight increases over time” (Lavallee et al. 2019, 153) in depression and anxiety in Chinese students. One important caveat when considering these disparate results on temporal relations may be differences in the factors that led to meat-abstention (e.g., religious practices, health and ethical considerations, or socio-economic status). For example, economically disadvantaged individuals who do not consume meat due to its relative cost may be at risk for ill-health for myriad reasons independent of their lack of meat consumption. Thus, future research examining temporal relations should establish clear distinctions between individuals and populations that abstain from meat consumption due to ethical, religious, and health-related perceptions, or those who do not consume meat for economic reasons.



In 2012, Beezhold and Johnston (2012) conducted a RCT in which 39 self-characterized omnivores (82% female) were assigned to one of three groups: lacto-vegetarian (i.e., avoided all animal foods except dairy), ovo-pescatarian (i.e., avoided meat and poultry but consumed fish and eggs), or omnivore (i.e., consumed meat and/or poultry at least once daily). Their results suggested that restricting meat, fish, and poultry improved some domains of short-term mood states. As detailed in our discussion, this study had major design flaws (e.g., potential observer-expectancy effects) and errors in interpretation and communication (e.g., nonequivalent groups at baseline, failure to recognize regression to the mean).

Can a Vegan Diet Be Healthy? A Literature Review

Dobersek, U., Wy, G., Adkins, J., Altmeyer, S., Krout, K., Lavie, C. J., & Archer, E. (2021). Meat and mental health: a systematic review of meat abstention and depression, anxiety, and related phenomena. Critical reviews in food science and nutrition, 61(4), 622-635.


Objective: To examine the relation between the consumption or avoidance of meat and psychological health and well-being.

Methods: A systematic search of online databases (PubMed, PsycINFO, CINAHL Plus, Medline, and Cochrane Library) was conducted for primary research examining psychological health in meat-consumers and meat-abstainers. Inclusion criteria were the provision of a clear distinction between meat-consumers and meat-abstainers, and data on factors related to psychological health. Studies examining meat consumption as a continuous or multi-level variable were excluded. Summary data were compiled, and qualitative analyses of methodologic rigor were conducted. The main outcome was the disparity in the prevalence of depression, anxiety, and related conditions in meat-consumers versus meat-abstainers. Secondary outcomes included mood and self-harm behaviors.

Results: Eighteen studies met the inclusion/exclusion criteria; representing 160,257 participants (85,843 females and 73,232 males) with 149,559 meat-consumers and 8584 meat-abstainers (11 to 96 years) from multiple geographic regions. Analysis of methodologic rigor revealed that the studies ranged from low to severe risk of bias with high to very low confidence in results. Eleven of the 18 studies demonstrated that meat-abstention was associated with poorer psychological health, four studies were equivocal, and three showed that meat-abstainers had better outcomes. The most rigorous studies demonstrated that the prevalence or risk of depression and/or anxiety were significantly greater in participants who avoided meat consumption.

Conclusion: Studies examining the relation between the consumption or avoidance of meat and psychological health varied substantially in methodologic rigor, validity of interpretation, and confidence in results. The majority of studies, and especially the higher quality studies, showed that those who avoided meat consumption had significantly higher rates or risk of depression, anxiety, and/or self-harm behaviors. There was mixed evidence for temporal relations, but study designs and a lack of rigor precluded inferences of causal relations. Our study does not support meat avoidance as a strategy to benefit psychological health.

Can a Vegan Diet Be Healthy? A Literature Review

Isabel Iguacel, Inge Huybrechts, Luis A Moreno, Nathalie Michels, Vegetarianism and veganism compared with mental health and cognitive outcomes: a systematic review and meta-analysis, Nutrition Reviews, Volume 79, Issue 4, April 2021, Pages 361–381, https://doi.org/10.1093/nutrit/nuaa030




Vegetarian and vegan diets are increasing in popularity. Although they provide beneficial health effects, they may also lead to nutritional deficiencies. Cognitive impairment and mental health disorders have a high economic burden.


A meta-analysis was conducted to examine the relationship between vegan or vegetarian diets and cognitive and mental health.

Data Sources

PubMed, Scopus, ScienceDirect, and Proquest databases were examined from inception to July 2018.

Study Selection

Original observational or interventional human studies of vegan/vegetarian diets were selected independently by 2 authors.

Data Extraction

Raw means and standard deviations were used as continuous outcomes, while numbers of events were used as categorical outcomes.


Of 1249 publications identified, 13 were included, with 17 809 individuals in total. No significant association was found between diet and the continuous depression score, stress, well-being, or cognitive impairment. Vegans/vegetarians were at increased risk for depression (odds ratio = 2.142; 95%CI, 1.105–4.148) and had lower anxiety scores (mean difference = −0.847; 95%CI, −1.677 to −0.018). Heterogeneity was large, and thus subgroup analyses showed numerous differences.


Vegan or vegetarian diets were related to a higher risk of depression and lower anxiety scores, but no differences for other outcomes were found. Subgroup analyses of anxiety showed a higher risk of anxiety, mainly in participants under 26 years of age and in studies with a higher quality. More studies with better overall quality are needed to make clear positive or negative associations.


Some specific important points about the methodology from the paper:

"When a study offered information about matched and nonmatched data, the matched data were used for analysis." but it looks like only one old study had matching.

"Only raw data (unadjusted) were used to perform the meta-analyses, as only 2 publications in the present meta-analysis included adjusted data." However, "Nevertheless, adjustment for confounders did not drastically change results in these 2 studies."

A Sequence Against Strong Longtermism

I don't think there's a consensus on whether physics is continuous or discrete, but I expect that what matters ethically is describable in discrete terms. Things like wavefunctions (or the motions of physical objects) could depend continuously on time or space. I don't think we know that there are finitely many configurations of a finite set of atoms, but maybe there are only finitely many functionally distinct ones, and the rest are effectively equivalent.

I think we've also probed scales smaller than Planck by observing gamma ray bursts, but I might be misinterpreting, and these were specific claims about specific theories of quantum gravity.

Also, a good Bayesian should grant the hypothesis of continuity nonzero credence.

FWIW, though, I don't think dealing with infinitely many possibilities is much of a problem as made out to be here. We can use (mixed-)continuous measures, and we can decide what resolutions are relevant and useful as a practical matter.

Mogensen & MacAskill, 'The paralysis argument'

Are these constraints on doing harm actually standard among non-consequentialists? I suspect they would go primarily for constraints on ex ante/foreseeable effects per person (already or in response to the paralysis argument), so that

  1. for each person and each extent of harm, the probability that you harm them to at least the given extent must be below some threshold, or
  2. for each person, the expected harm is below some threshold, or
  3. for each person, their expected value from the act is nonnegative (or close enough to 0), so only acts which leave no one in particular worse off in expectation than "doing nothing", i.e. weak ex ante Pareto improvements, maybe with a little bit of room.

The thresholds could also be soft and depend on benefits or act as a penalty to a consequentialist calculus, if you want to allow for much more significant benefits to outweigh lesser harms.

It might get tricky with possible future people, or maybe the constraints only really apply in a person-affecting way. Building off 3 above, you could sum expected harms (already including probabilities of existence which can vary between acts, or taking the difference of conditional expectations and weighting) across all actual and possible people, and use a threshold constraint that depends on the expected number of actual people. Where  represents the individual utilities in the world in which you choose a given action and  represents the utilities for "doing nothing",

  1. , or
  2. , for some (or all) values of  such that .

This could handle things like contributing too much to climate change (many possible people are ex ante worse off according to transworld identity) and preventing bad lives. With counterparts, extending transworld identity, you might be able to handle the nonidentity problem, too.

Some constraints might also be only on intentional or reckless/negligent acts, although we would be owed a precise definition for reckless/negligent.

What would you do if you had half a million dollars?

Besides others mentioned, consider also getting in touch with

  1. https://www.cooperativeai.com/
  2. https://emergingrisk.ch/ (I think this is s-risk-focused, given the team running it)
What would you do if you had half a million dollars?

I think robustness (or ambiguity aversion) favours reducing extinction risks without increasing s-risks and reducing s-risks without increasing extinction risks, or overall reducing both, perhaps with a portfolio of interventions. I think this would favour AI safety, especially that focused on cooperation, possibly other work on governance and conflict, and most other work to reduce s-risks (since it does not increase extinction risks), at least if we believe CRS and/or CLR that these do in fact reduce s-risks. I think Brian Tomasik comes to an overall positive view of MIRI in his recommendations page, and Raising for Effective Giving, also a project by the Effective Altruism Foundation like CLR, recommends MIRI in part because "MIRI’s work has the ability to prevent vast amounts of future suffering.".

Some work to reduce extinction risks seems reasonably likely to me on its own to increase s-risks, like biosecurity and nuclear risk reduction work, although there may also be arguments in favour related to improving cooperation, but I'm skeptical.

For what it's worth, I'm not personally convinced any particular AI safety work reduces s-risks overall, because it's not clear it reduces s-risks directly more than it increases them by reducing extinction risks, although I would expect CLR and CRS to be better donation opportunities for this given their priorities. I haven't spent a lot of time thinking about this, though.

Load More