2017 Donor Lottery Reportpost by AdamGleave · 2018-11-12T19:26:32.136Z · EA · GW · 8 comments
Selection Process Summary of conclusions ALLFED perspective downsides metrics from ALLFED Global Catastrophic Risk Institute (GCRI) AI Impacts Wild Animal Suffering Research (WASR) None 8 comments
I am the winner of the 2017 donor lottery. This write-up documents my decision process. The primary intended audience are other donors: several of the organisations I decided to donate to still have substantial funding gaps. I also expect this to be of interest to individuals considering working for one of the organisations reviewed.
To recap, in a donor lottery many individuals make small contributions. The accumulated sum is then distributed to a randomly selected participant. Your probability of winning is proportional to the amount donated, such that the expected amount of donations you control is the same as the amount you contribute. This is advantageous since the winner (given the extra work, arguably the "loser") of the lottery can justify spending substantially more time evaluating organisations than if he or she were controlling only their smaller personal donations.
In 2017, the Centre for Effective Altruism ran a donor lottery, and I won one of the two blocks of $100,000. After careful deliberation, I recommended that CEA make the following regrants:
- $70,000 to ALLFED.
- $20,000 to the Global Catastrophic Risk Institute (GCRI).
- $5,000 to AI Impacts.
- $5,000 to Wild Animal Suffering Research.
In the remainder of this document, I describe the selection process I used, and then provide detailed evaluations of each of these organisations.
I am a CS PhD student at UC Berkeley, working to develop reliable artificial intelligence. Prior to starting my PhD, I worked in quantitative finance. This document is independent work and is not endorsed by CEA, the organisations evaluated, or by my current or previous employers.
I assign comparable value to future and present lives, place significant weight on animal welfare (with high uncertainty) and am risk neutral. I have some moral uncertainty but would endorse these statements with >90% probability. Moreover, I largely endorse the standard arguments regarding the overwhelming importance of the far future.
Since I am mostly in agreement with major donors, notably Open Philanthropy, I tried to focus on areas that are the comparative advantage of smaller donors. In particular, I focused my investigation on small organisations with a significant funding gap.
To generate an initial list of possible organisations, I (a) wrote down organisations that immediately came to mind, (b) solicited recommendations from trusted individuals in my network, and (c) reviewed the list of 2017 EA grant recipients. I shortlisted four organisations from a superficial review of the longlist. Ultimately all the organisations on my shortlist were also organisations that immediately came to my mind in (a). This either indicates I already had a good understanding of the space, or that I am poor at updating my opinion.
I then conducted a detailed review of each of the shortlisted organisations. This included reading a representative sample of their published work, soliciting comments from individuals working in related areas, and discussion with staff at the organisation until I felt I had a good understanding of their strategy. In the next section, I summarise my current views on the shortlisted organisations.
The organisations evaluated were provided with a draft of this document and given 14 days to respond prior to publication. I have corrected any mistakes brought to my attention, and have also included a statement from ALLFED; other organisations were provided with the option to include a statement but chose not to do so. Some confidential details have been withheld, either at the request of the organisation or the individual who provided the information.
Summary of conclusions
I ranked ALLFED above GCRI as I view their research agenda as having a clearer direct path for impact and greater room for growth. GCRI’s work intentionally spans a wide range of catastrophic risks. I was most impressed by their work on modelling nuclear risk. However, while having better models is useful for cause prioritisation, I am sceptical of its ability to directly influence policy makers decisions, especially elected officials. By contrast, I find it plausible that ALLFED will make significant progress on developing alternative foods, mitigating a significant fraction of the negative effects of a nuclear war. The areas I am most uncertain of in this evaluation are the possible downside risks of ALLFED, including outreach mishaps and moral hazards.
I ranked GCRI above AI Impacts as AI Impacts core staff are adequately funded, and I am sceptical of their ability to recruit additional qualified staff members. I would favour AI Impacts over GCRI if they had qualified candidates they wanted to hire but were bottlenecked on funding. However, my hunch is that in such a situation they would be able to readily raise funding, although it may be that having an adequate funding reserve would substantially simplify recruitment.
I rank AI Impacts above WASR due to a long-term future outlook. I find WASR slightly more compelling by the lights of a near-term animal welfare centric outlook than I find any of the other organisations under a long-term future outlook, but the difference is not substantial enough to be relevant given my fairly low levels of moral uncertainty.
If I had an additional $100k to donate, I would first check AI Impacts current recruitment situation; if there are promising hires that are bottlenecked on funding, I would likely allocate it there. Otherwise, I would split it equally between ALLFED and GCRI. In particular, I recommend a proportionally greater allocation to GCRI than I made. My donation to ALLFED increased their 2018 revenue by 50%: although they have capacity to utilize additional funds, I expect there to be some diminishing returns. I am also happy to discuss my thoughts further with other donors considering supporting these organisations.
ALLFED is a non-profit conducting research and outreach for alternative food sources to be used in the event of a mass agricultural catastrophe. Their primary goal is to develop alternative food sources to allow a large fraction of humanity to survive a nuclear winter scenario. Even in the worst case of a full-scale US-Russia nuclear interchange, the majority of the human population would be outside the blast radius of nuclear detonations, living in remote rural areas or in non-combatant countries. However, there would be many indirect negative effects that impact everyone, including destruction of industrial capacity, government disruption and agricultural failures from nuclear winter. ALLFED is attempting to provide technical solutions to the last of these problems, by developing food sources that can be grown even with limited sunlight.
They have around 3.5 full-time equivalent staff, consisting of several part-time employees and a diverse set of volunteers. Their co-founder and primary researcher, Prof. David Denkenberger, splits his time between ALLFED and teaching at the University of Alaska Fairbanks. Sonia Cassidy is in charge of operations, with a background in social enterprise and business continuity.
They have a limited budget, with a projected revenue of $215,000 in 2018 (including my donation). The most likely use of additional funding would be to hire additional junior employees/contractors, or provide research grants to academics to work on related topics. Several of their current volunteers they would like to hire as contractors on a part-time basis. David could also buy himself out of teaching, and so be able to dedicate almost all his time to ALLFED.
ALLFED is a young organisation in an untested cause area. There are three main areas of uncertainty: (a) the likelihood of nuclear exchange and consequent nuclear winter; (b) the tractability of alternative foods; (c) the managerial capacity for ALLFED to scale as an organization.
There are few publicly available quantitative models of the risk of nuclear war. Barrett et al (2013) is the most comprehensive study I'm aware of, although its estimates should likely be updated downwards due to the historical absence of nuclear war. It seems likely that a full-scale US-Russia nuclear exchange would trigger a nuclear winter, although this has not been adequately studied. Overall, I think the chance of nuclear winter happening is low but still high enough to be one of the top five global catastrophic risks. Although nuclear security as a whole attracts significant government attention and investment, nuclear winter mitigation is highly neglected, and I am not aware of any organisation besides ALLFED working on alternative foods.
Although it is hard for me to assess as an outsider, there appear to be a number of tractable research pathways to developing alternate foods. One of these is methane (natural gas) digesting bacteria, a technology being developed at an industrial scale as a feed source for animal agriculture by startups including Unibio and Calysta. ALLFED are currently experimenting with growing methane digesting bacteria at the household scale. This would be most influential in a no-sun, no-industry scenario (note that gas will continue to bleed out of wells), but is also a hedge against a high cost of retrofitting existing chemical plants.
The outcome of these experiments will give considerable insight into ALLFED's ability as an organization to conduct novel research in this space. While I have been impressed by David's other research output, he is not an agronomist nor biologist, and has a limited track-record in this area. Finan Adamson, a recent hire who will be working on the methane-digestion experiment, does have a background in biology but has little research experience. Accordingly, I find it likely that ALLFED will initially fail to make progress in this area.
ALLFED is small with the majority of their research still being conducted by David. As such, there are serious questions regarding their ability to recruit and manage new hires. Furthermore, many of the skills needed by ALLFED (such as generalist researchers and operations capacity) are in high-demand at other EA organisations, raising a question as to the opportunity cost of any hires they make. ALLFED are, however, optimistic about being able to recruit from outside the EA talent pool.
I am of the opinion that there is in fact a large amount of untapped talent, both within the effective altruism community and amongst other researchers who would be interested in this space. Effective altruism is not so much talent constrained as organisation constrained: there need to be more environments where junior hires can develop their skills. David is happy to spend his time mentoring new researchers in this area, and so ALLFED is well placed to help fill this gap.
ALLFED has a "two-pronged strategy", performing R&D to improve long-term resilience, but also performing outreach now in case a disaster happens soon. Examples of outreach:
- Talking to senior civil service officials.
- Individual discussions with senior agronomics researchers, including exploring collaboration opportunities.
- News coverage, e.g. this interview in Science.
I tend to think of outreach to government officials as being premature: at this time ALLFED has little concrete advice to offer government officials or the broader public. ALLFED disagree with this, arguing that governments having any response plan to a 10% or 100% agricultural shortfall would substantially improve the outcome. In particular, ALLFED feels that at the 10% level, there a a number of shovel-ready interventions, such as pre-commitments to trade between nations and utilising agricultural food residues and domestic food waste (via municipal collection) as food for ruminant digesters. Moreover, ALLFED hope that by raising awareness of alternative foods amongst policymakers, this may spur governments into funding research or conducting their own investigations into alternative foods. However, ALLFED agree there has been no tangible results from government outreach so far, although they feel the response of officials has been broadly positive.
I am also concerned about media coverage, as in general I think it is challenging to have a high-quality public discussion about low-probability high-impact events such as global catastrophic risks, with coverage tending to be either alarmist or dismissive. In the case of alternative foods, I would be concerned about it becoming perceived by relevant fields as an agenda being pushed by outsiders. There is also a risk of becoming associated with fringe communities such as survivalists. Both of these could harm the long-run development of alternative foods, and possibly even hinder action on other global-catastrophic risks. However, ALLFED has recently shown encouraging signs of integrating with the agronomy community: they now have a publication in Agriculture, and a researcher in sustainable agriculture, Dr Shackelford, has joined their board.
An additional risk is that development of alternative foods might cause moral hazard amongst nuclear decision makers. The threat of a nuclear winter had a large influence on the perception of nuclear war amongst the public and policy makers. Indeed, Gorbachev stated in a 2000 interview that "models made by Russian and American scientists showed that a nuclear war would result in a nuclear winter that would be extremely destructive to all life on Earth; the knowledge of that was a great stimulus to us, to people of honor and morality, to act in that situation." This feels to me like a second-order consideration: even with alternative foods, a nuclear winter would still mean the destruction of nations involved in the nuclear interchange, and the death of hundreds of millions. But it does provide an additional reason to clearly communicate both the capabilities and limitations of alternative foods to the public and policy makers.
I view ALLFED as being an exciting seed funding opportunity. However, I would like to see substantially more evidence of progress before considerable funding (beyond $1 million) is provided. In particular, I would want to see improvement on some (but not all) of the following metrics:
Progress on concrete alternative foods research: e.g. the methane digesting bacteria project above. This work needn't have a positive outcome: I'd be almost as excited about negative results if valuable lessons are learned from it.
Development of a concrete research agenda, reflecting what they believe are the most impactful and tractable experiments to conduct in the near-future. This is in comparison to their existing work, which has focused on high-level cost effectiveness analyses for alternative foods and long, superficially developed lists of possible interventions.
I recommend this in particular since, despite reading their public material and having several discussions with ALLFED, I am still uncertain what their immediate next steps are. Creating a concrete research proposal would at the least make recruitment and fundraising easier, and might also help set an internal direction for the organisation. Of course, I would expect and encourage deviations from this proposal as ALLFED learns more about the area.
Recruitment or setting up collaborations with researchers from relevant backgrounds, e.g. in biology or agriculture.
I would also want to check that the following mistakes do not occur. To clarify, I do not expect ALLFED to make these mistakes; I merely believe these are the most salient risks for this cause area:
- Low-quality output that could hamper development of the field, e.g. cost-effectiveness analyses with serious errors or research output that is viewed as low-quality by those in the relevant field.
- Outreach (whether to policymakers, academics or the media) that leaves a bad impression of alternative foods or global catastrophic risks.
- Recruitment or support of individuals who have a poor track record in the above areas.
I am somewhat more excited about ALLFED than GCRI since their research agenda seems more directly impactful and there is a clearer pathway for growth. However, I see more downside risks to ALLFED, and in particular would expect GCRI to be in a better position to work productively with governments. ALLFED has a large team of volunteers, which increases reputational risks. I view support for ALLFED at this stage as mostly a test of the tractability of R&D in this area, and to enable them to continue to build relevant collaborations.
Statement from ALLFED
David Denkenberger, the founder of ALLFED, provided the following comments on a draft of this report:
Thank you, Adam, for both the grant and your careful consideration (and our thanks also to everyone who took part in the EA Lottery in the first place). We appreciate the feedback and the process, as it has been useful to work on this over the course of several months. The discussions we have had have helped to crystalise ideas.
As “a young organisation in an untested cause area,” which we are, we also appreciate this opportunity to highlight the cause area itself as much as our particular organisation. It is our belief that the area of alternative foods development has not yet received nearly as much attention as it deserves, especially given its cost effectiveness and the potential to contribute to both recovery from catastrophes as well as many problems facing the world today.
The key challenges of researching, testing, and developing viable alternative food solutions, while also providing a solid organisational structure, are valid and something we remain conscious of. For this reason, from the beginning, we have been committed to steady, sustainable growth, which allows for more and more research as well as for sufficient time to put a strong organisational framework in place (with a new internship programme the most recent addition to it).
We completely agree that media handling is a delicate matter at the best of times, but this is particularly so when it comes to sensitive and/or new subjects. At the same time, we consider it essential that any organisation, regardless of its field of work, be capable of handling such interest in an appropriate manner, especially if there is a high possibility of miscommunication and misinterpretation. As such, at ALLFED, we prefer to be able to engage with enquiries in a constructive manner, rather than not at all. We encourage others to contact us when in need of our assistance or expertise, particularly in the event of a catastrophe.
In terms of ALLFED’s research agenda, we have started scoping out the most promising experiments at various funding levels. On the lower end, there is approximately $50,000 for household scale natural gas eating bacteria experiments. On the higher end, there is >$1 million for flexible biorefinery design that could turn crop leaves into fuel or food which would involve experts in biofuels, mushrooms, chickens, and ruminants. Projects such as these will be implemented according to the resources available. We have already established several potential partnerships around this, so people with transferable skills can be redirected towards this effort with more funding. We would welcome further enquiries and opportunities for collaboration.
Collaboration, hope, and care are at the heart of ALLFED’s vision and our work. We believe that with advanced planning, research, collaboration, and communications it is both plausible and possible to provide for all (or most) humans and to help preserve biodiversity in the event of a catastrophe. The EA Lottery grant is the next step towards this.
Global Catastrophic Risk Institute (GCRI)
GCRI is a research-oriented decentralised think tank, with 1.5 full-time equivalent paid staff members, and some voluntary research associates. Their only full-time employee is Seth Baum, with two to three paid part-time employees including their co-founder, Tony Barrett. GCRI's work intentionally spans a wide range of catastrophic risks, but nuclear weapons and artificial intelligence are two recurring themes.
They currently have a very small budget, with a forecast revenue of $165k for 2018 (including my donation). This gives them limited runway: although there will be some money leftover for 2019 (which can be extended by Seth taking a paycut), it will not last more than a year. GCRI hope to raise $500k to cover salaries for their current employees for the next two years, with leftover to hire more support.
Many of the risks investigated by GCRI are neglected, especially within the effective altruism community. Although some risks (e.g. nuclear security) have attracted considerable attention from other communities, I believe it is still worthwhile engaging. For one, GCRI can provide a more long-term focus than other actors. Additionally, if GCRI does uncover promising interventions in this area, they are in a good position to communicate it to the effective altruism community, shaping future donations and work.
I have generally been impressed by Seth Baum's research output. In particular, their model for the impact and probability of nuclear war is a substantial improvement on any publicly available models. The model itself is still fairly crude, but previous work had taken a qualitative approach and not moved far beyond case studies.
My main criticism is GCRI's output has in the past sometimes emphasised quantity over quality. In particular I would have liked to see GCRI leadership provide greater mentorship and exercise more editorial control over the work by GCRI's associates. Low-quality research can reflect badly both on GCRI, and on the rest of the field. However, with the recent overhaul of GCRI's affiliates program, I expect this aspect to improve. Seth indicated that their uncertain funding situation has pushed them towards keeping a steady level of research output, rather than letting ideas mature, so it is also possible that having a greater runway would alleviate this issue.
It is also unclear to me whether GCRI should continue to exist as an independent organisation, as opposed to Seth joining an organisation such as CSER or FHI pursuing a similar agenda. Joining a larger organisation could provide more opportunities for research collaboration, and greater operations support. However, Seth believes that relocating would hamper his interaction with US policy communities.
Overall I am moderately excited about supporting the work of GCRI and in particular Seth Baum. I am pessimistic about room for growth, with recruitment being a major challenge, similar to that faced by AI Impacts. However, Seth is significantly more optimistic than me about the ease of recruitment, and believes that GCRI's distributed nature will allow them to hire people who are unwilling to relocate to work at another organisation doing similar work. He notes that recruitment has never been a focus, since they have not had the funding to grow.
At their current budget level, additional funding is a factor for whether Seth continues to work at GCRI full-time. Accordingly I would recommend donations sufficient to ensure Seth can continue his work. I would encourage donors to consider funding GCRI to scale beyond this, but to first obtain more information regarding their long-term plans and recruitment strategy.
AI Impacts focuses on an important research area, AI forecasting, that could substantially influence the research agendas of technical AI safety researchers and the focus of AI policy. They currently have 2 FTE employees. They have a small budget, with reserves of $70k (as of July 2018) and a forecast revenue of $480k for 2018. Their forecast revenue for 2019 is considerably lower, as $350k of the revenue in 2018 was intended to be used over two years.
The area seems moderately tractable. I broadly agree with the assessment of their founder, Katja Grace, that there are a number of tractable low-level questions (e.g. what are the length of axons in the brain?), but it is unclear how much light they shed on medium-level (e.g. how important is communication vs compute?) and high-level (e.g. how much parallelism can distributed training exploit?) questions.
I have found Katja's output in the past to be insightful, so I am excited at ensuring she remains funded. Tegan has less of a track record but based on the output so far I believe she is also worth funding. However, I believe AI Impacts has adequate funding for both of their current employees. Additional contributions would therefore do a combination of increasing their runway and supporting new hires.
I am pessimistic about AI Impacts room for growth. This is primarily as I view recruitment in this area being difficult. The ideal candidate would be a cross between an OpenPhil research analyst and a technical AI or strategy researcher. This is a rare skill set with high opportunity cost. Moreover, AI Impacts has had issues with employee retention, with many individuals that have previously worked leaving for other organisations.
I should note that Katja was substantially more optimistic than me about recruitment. My view (mentioned previously in the evaluation of ALLFED) that the space is more "organisation constrained" than "talent constrained" also suggests recruitment may be tractable, conditioned on AI Impacts being willing to dedicate significant time to recruiting and mentoring employees.
Given their small size, it's unclear to me whether AI Impacts should continue to exist as a separate organisation (similar to the consideration for GCRI). Their work would be a good fit for FHI, although location preferences are likely to rule out any such merger.
Overall, the thing holding me off on making a stronger endorsement is recruitment. It's worth checking in to see what hires they are currently considering: I'd want to support AI Impacts if there was someone they were excited about and they did not have the funding. It might be worth funding them so they have cash on the sidelines in case the right person comes along.
Conflicts of interest: Katja Grace, the founder of AI Impacts, lives in the same house as me. She did not at the time the decision was made to recommend this grant.
Wild Animal Suffering Research (WASR)
Wild Animal Suffering Research (WASR) is a non-profit with 2 FTE (1 full-time and 2 part-time) employees, seeking to bootstrap research into wild animal suffering. Around 1.5 FTE is spent on in-house research, both on fundamental topics such as welfare evaluation and possible interventions. The remainder of the time is spent on management, and outreach to academics in related fields.
The primary audience for their research output are EAs and animal advocates, and it is written with the goal of better guiding their own next steps. I have found their research a useful resource for better understanding wild animal suffering. It tends towards distilling existing research, rather than conducting primary research. They may try to get some of their output published in peer-reviewed venues, but this is not a key goal.
WASR have conducted outreach to academics, and plan to soon offer grants to a small number of academics. They have created a database of faculty members working in relevant fields, and e-mailed a shortlist. When I spoke to them (in early June), they had contacted 34 academics and heard back from 12. WASR will need substantially more funding to be able to provide grants to more than one or two academics, with $50,000 being the minimum grant size most academics would find useful. They have also recently ran a small grants competition aimed at independent researchers.
WASR has clearly spent significant time considering strategy, and have been commendably transparent, publishing their strategic and research plans. I broadly support their strategy, although I would favor two modifications. I believe more care should be taken during outreach at this early stage. It is hard to know what framing will most engage academics or other stakeholders attention. If a bad first impression is made, it may be hard to recover, potentially hampering field-building efforts for WASR and other organisations in the future. WASR are aware of this and are confident they have left a positive or neutral impression. In particular, they have not received any direct negative feedback from individuals they have contacted. However, we do not know what impression the majority of academics who did not respond formed.
My second recommendation would be to prioritize supporting individuals who can champion this field of research from within existing academic or other research institutes. It is challenging to seed a field as an outsider, and in cases such as cryonics and nano-technology this has backfired. I am uncertain of this recommendation, however. In particular, philanthropists have a better track-record of seeding fields, so if WASR were to become a major grant-maker in this space this could be a viable (albeit expensive) method of field-building.
Additionally, WASR argued that PhD students in relevant fields (such as zoology or bioethics) have limited autonomy, so a junior researcher might have little ability to work on relevant topics. In particular, they believe that merely providing funding for PhD students is unlikely to be sufficient, unless there is buy-in from senior researchers. Accordingly, they favor a more long-term relationship building approach. This seems plausible to me, but my hunch is that a PhD student inside academia might still be better placed to build these relationships than an external organisation. My intuition here is heavily influenced by the norms in my own research field, Computer Science, where PhD students can have substantial academic freedom if they have an external fellowship and an open-minded adviser. This may be less true in other fields.
Overall I think WASR is a well-run organisation with a clear strategy and a short but encouraging track record. I would encourage those with a near-term animal welfare centric worldview to support them. Under my own worldview, I did not find them competitive with the other organisations, and so recommended a small grant of $5,000.
Comments sorted by top scores.