Who in the SoGive team is working on it?

Well-positioned

Weighting factor:

None, previously Robert Harling and Matt S

This is dependent on when the new RCT is published - doesn't seem to be available yet. I've asked the authors for an update. (we may also want to draw on broader evidence for context, e.g. https://journals.plos.org/plosmedicine/article?id=10.1371/journal.pmed.1003827) Some discussion around whether it woudl be possible to do something very quick and easy on this which just hilights the fact that there is new evidence coming out

The max score for "well-positioned" is 5, but this seems to undersell how well-positioned we are, given that the work is mostly done

We started this work because there appeared to be RCT quality evidence of an intervention which prevented severely bad outcomes. It seems likely that many donors would be excited to fund something which provably prevents rape.

Hard to determine a meaningful tractability score for this; in some senses the work essentially follows a standard GiveWell playbook for how to analyse a charity. Determining the moral weights could be hard both in the normal sense (it's difficult to know the correct answer) as well as being emotionally difficult topics to work with.

There's a substantial amount of general interest in the topic of sexual violence, but we know of minimal activity in terms of high quality cost effectiveness analysis of charities working in this area.

In some sense we are extremely well positioned, in that we have already done most of the work, and all that's needed is to incorporate the findings of the new RCT and then publish

Some work answering the RFP from Damon Binder and ASB (note the deadline is given as some time in the past in summer 2022, but [redacted] they may still be interested in seeing this question answered)

Getting a quantitative handle on GCBRs seems pretty unambiguously high value

Given the lack of directly applicable data, producing a high confidence answer could be tricky. However producing work that does the best job we could hope for given the available data seems straightforward, given that we are more than capable of good reasoning transparency. Also, of the items listed under "what a successful proposal might look like", some of them look harder than others, but picking off a few items and doing them well should be straightforward

On the one hand, given how valuable it is, one might imagine that this must be done to death already. On the other hand, given how many really valuable things haven't yet been researched, and given than Open Phil has been trying to get this work done and apparently hasn't had success, this may in fact be neglected

We are more than capable of doing good reasoning transparency, and have been discussing GCBRs with experts for some years

Africa: Explore giving opportunities relating to climate in Africa (other than CATF's work in this area)

Africa is growing in terms of its carbon footprint, and its energy systems and infrastrcuture are being built right now, which means we can avoid path dependency problems by getting it carbon friendly from the outset

In general, finding the right interventions is likely to be tricky -- Africa is a large and complex place. Finding high imapct orgs is also difficult; ease of doing business is generally low in Africa, so it takes exceptional people to succeed in this context

I think Johannes might have indicated he might think about this

We are fairly well positioned on climate work

Counterfactuals vs credit-sharing/Shapley values

There's a tension between considering things on counterfactual/at-the-margin basis and a credit-sharing/shapley values basis. This expanded on in the doc (see doc links section)

It's really quite rare for an organisation to achieve its impact in isolation. (e.g. global health organisation will likely collaborate with the ministry of health, a research org is probably part of a community of research orgs, a policy org is likely part of a community of orgs, and obviously interacts with the government/institution that it's trying to influence)

The concepts are fairly subtle

You can occasionally see some references to this in the works of EA orgs (I think 80k have alluded to this distinction before) but I haven't seen anyone write about this carefully

Thinking about this topic is something we have to do anyway as part of our past research. We haven't actually solved the problem yet, but our past experience leaves us well primed to think about this.

https://docs.google.com/document/d/1BvBVqXPSHfQXcYocgdGFh2VVE6sW0Brx6SxrlSKyztY/edit

Calculate a SCC which includes tail risks

Having a social cost of carbon which doesn't omit most of the risks would be very valuable for cause prioritisation

Existing integrated assessment models are complex. Understanding what is and isn't already in them could take material effort, and so there's a potential for double counting. There may be ways round this, especially if the tail risks turn out to be materially larger than the risks currently captured by the models in expectation.

There's lots of literature on integrated assessment models, and it's not rare to find a critique which observes that tipping points (or something else) is missing from the models, but work to quantify the things which are missing appears to be neglected

We are fairly well positioned on climate work

Which is the right lens for reviewing mental health interventions?

At first glance it might seem that the nature of the intervention would be important (e.g. is it CBT, or IPT, or Rogerian therapy, etc). However there is some research which appears to suggest that this doesn’t actually matter that much. A candidate for a potentially more important lens is the cost structure. This might mean comparing: Highly qualified psychologist / Less well qualified counsellor / Volunteer-driven or under-qualified-driven (task-sharing) services / Software/app-based approaches

Mental health appears to be a popular topic; this is less central to the topic than a review of Strong Minds

Lots of the research backing this is already there, seems straightforward to interpret

We don't know of anyone tackling this particular question, but HLI seem to be thinking along similar lines

We are well positioned for this

Lessons learnt from Enmod and how we can ensure BWC doesn’t fail (an example of failed international governance on environmental modification).

Getting an insight on how to make the BWC work effectively would be great, and it seems likely that we will never solve GCBRs without either the BWC or something similar. But there's a limit to how much we can learn from one case study in failure (especially since there are lots of ways to fail and few ways to succeed)

There is probably plenty of relevant information easily accessible on the internet

Given how straightforward / tractable it (presumably) is, it's probably been done before

We have good background on bio, and so are well positioned to draw out the relevant implications

Redteaming John Halstead's report on climate change

Review this carefully; including how the modelling bits are done. (Suggested by [redacted], late 2022) It's available from the WWOTF website, or here's a direct link: https://drive.google.com/file/d/14od25qdb4sdDoXVDMoiSrTwuzYAMSpxK/view

Halstead's report is pretty extensive and has the potential to be a substantial reference resource for the EA community. If we are able to identify any issues, this could be valuable for the community

Arguably there's a lack of tractability in the sense that the report is lengthy, however otherwise redteaming is relatively easy

As far as we know nobody is doing this

We are fairly well positioned on climate work

Criteria for assessing neartermist interventions through a longtermist lens

(e.g. don’t include impact on economic growth in the criteria in the sense economic growth makes the longterm future better (but possibly include it on other grounds), do include impact on x-risk, do include impact on values)

It seems both that the "I" of ITN should be much higher than 5, and that the "T" of ITN should be much lower than 1. It probably seems that the T should be dragged down more than the I should be pulled up, but the T is already weighted more than I anyway, so arguably this is already captured

If successful, this research would be a massive development for longtermism

Giving this the lowest possible score arguably doesn't fully capture how difficult this is. At first glance, this looks extraordinarily difficult, possibly doomed to failure

It seems unlikely anyone else is doing this, but with good reason.

Could argue that our team has been reviewing lots of different interventions, and therefore are well positioned, but arguably that claim doesn't really pass much muster. The work is mostly about working out difficult/impossible(?)-to-know implications of actions.

Broadly applicable

Effective think tanks - case studies

Case studies of effective think tank efforts

Although this was first mentioned to us in a bio context, there are lots of cause areas for which this would be valuable. I.e. even if we focus on one specific angle (e.g. thinking about the GCBR context) the results would still hvae some applicability in other contexts (albeit somewhat less applicable)

This seems pretty straightforward, although standard impact assessment difficulties will no doubt arise (eg considering counterfactuals and assigning credit between different actors)

We don't know of anyone else working on this, but it seems like that sort of thing which other people would have thought about

Moderately well positioned in that we have been interacting with think tanks quite a bit in recent years

Need reviewing in context of FP work - I didn't realise Pratham were working with J-PAL on TaRL Africa: https://founderspledge.com/stories/teaching-at-the-right-level-africa-high-impact-funding-opportunity

Education appears to be a popular topic

The work essentially follows a standard GiveWell playbook for how to analyse a charity. A lower bound for the impact comes from the theory fo change: intervention --> better educational outcomes as measured by test scores --> higher income in later life. If that is sufficient to lead to a high impact outcome, this becomes a relatively tractable piece of work

Founders Pledge has done some work on this. More generally, there is interest in education and academia has produced lots of work on the effects of education (although much of this work is not decision-relevant)

We are well positioned for GiveWell-esque analysis

None but Tommy, Ines, Melissa were interested

Ultimate problem is this: To control the spread of pathogens we need updated building design. So need research into the diffusion of this updated building design (imagining we’ve solved the ideal design re ventilation and airflow); how much can be achieved with just voluntary standards? How much can be achieved with mandatory standards? To get to this, this research is the following useful piece of analysis: What can we learn from past experience like energy efficiency standards or fire suppression?

Applicability to bio is clear. Questionable whether building design is the most important thing in a bio context, but it's definitely a valuable angle. Also useful for climate

Simply doing a historical review would be straightforward.

We don't know of anyone else working on this; doesn't seem like the sort of thing that would be neglected though

We don't have particular advantages when it comes to buildings, but it doesn't seem likely that this will hold us back too much

Explore 'Giving information on the benefits, costs, and quality of education'

If it really is as cost-effective as it looks, this would be a highly valuable finding. The assessment factors in a chance that it won't be (because most things aren't as good as they look)

Most of the work should be a straightforward lit review

Could argue that it can't be that neglected given that it was mentioned in the JPAL best buys leaflet. But it doesn't seem to get mentioned much

We are well positioned for this

Previously: Violet B, Danny Bressler reviewed

Danny B has reviewed and made material amendments to the report; more accurate, but maybe less readable; some discussion about whether the climate models need to refelct indirect effects?

The analysis is couched in terms of the SoGive Gold Standard, which means that it's essentially inapplciable in a context outside of SoGive. Furthermore, given that the scope doesn't invovle really challenging the models, we would have to acknowledge that the IAMs don't properly capture the reality

By limiting the scope and only leveraging the existing models, this makes the work easy

This sort of comparison across different cause areas is rare outside of EA

We are well positioned for climate work

Clean cookstoves have the potential to be valuable from a health, climate, and financial perspective

There is plenty of relevant literature

There is plenty of relevant literature

We are well positioned for this

"Graduation approach" - is BRAC high impact?

This paper compared Livelihood, Cash Transfer and Graduation Approaches. It found that " Overall, lump-sum cash transfers are found to have the highest impact-cost ratio, followed by livelihood and Graduation programs. However, Graduation approach has the most rigorous evidence of long-term (at least a year after end of intervention) impacts" https://www.researchgate.net/profile/Munshi-Sulaiman-2/publication/309492963_Making_Sustainable_Reductions_in_Extreme_Poverty_A_Comparative_Meta-Analysis_of_Livelihood_Cash_Transfer_and_Graduation_Approaches/links/5813836e08aedc7d8961e35e/Making-Sustainable-Reductions-in-Extreme-Poverty-A-Comparative-Meta-Analysis-of-Livelihood-Cash-Transfer-and-Graduation-Approaches.pdf

Lots of interest in lifting people out of poverty; it certainly seems that we *ought* to have some better way of tackling poverty than "just give people cash"

At least the first steps should be straightforward; simply review the Sulaiman 2016 paper linked to in the brief description.

Graduation approaches are well known in the international development community. The Sulaiman paper seemed to be the only literature available on the topic, but that's based on a few seconds of looking on Google scholar

We are well positioned for this

Mental health appears to be a popular topic in general

There seems to be nothing particularly more difficult about this than anything else

Plenty of good work from HLI, + a bit from FP, and lots of academic interest in mental health interventions

We are well positioned for this style of analysis

Should nets be given away or sold?

Moscibrodzki et al 2018 [1] found that purchased bed nets were more likely to be correctly used than a free net. [1] https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6191916/ More generally, there were several interesting pieces in this google scholar search: https://scholar.google.co.uk/scholar?hl=en&as_sdt=0%2C5&q=malaria+nets+purchased+free&btnG=

Has a reasonable chance of unseating one of the most robustly supported EA charity recommendations

I suspect that the literature on this topic will exist, but won't be comprehensive enough to fully answer the question

There is some literature on the topic in academia. Seems to have limited attention on the GiveWell website

We are well positioned for this

Review the methodologies of the standard setters for offsetting orgs (e.g. Gold Standard and Verra).

These standards are very widely used, so if it turns out that the methodologies are poor, then this becomes useful information to several people. However it's unclear how many of these donors can be persuaded to donate instead to a "non-offset-like" charity like CATF, and in the absence of something to point them to instead, there's a reasonable chance that the donor may simply continue with ineffective offsets. Of course, some of them will be persuaded, so it's harsh to give this minimal importance

We haven't looked at them yet, but it may be that shortcoming will be easy to find fairly easily/quickly.

Few people in EA are interested in exploring this (for understandable reasons -- most offsets seem ineffective) and (based on anecdotal evidence) people outside of EA seem to be surprisingly trusting of these third party providers. That said, presumably someone must be scrutinising them

We have given lots of thought to tree planting interventions and have lots to say on them

Development Media International

There's a good chance that this will prove to be high impact, in that they used to be a GiveWell stand out charity, and our previous review of their work was erring in the direction of being more positive than GiveWell

The work essentially follows a standard GiveWell playbook for how to analyse a charity. Furthermore, there is plenty of existing GiveWell work to leverage

There is already some GiveWell work on them

We did some work on DMI a few years ago, although that's possibly out of date by now

One approach to take would be to simply redteam the report done by Rethink. This would be relatively easy (I've already a reasonable amount of this in the comment I wrote). A harder version would try to give some credit to the nonclimate benefits of anti-deforestation work (e.g. biodiversity (including the benefits for science of having more species available), pandemic risk, etc). Rethink report: https://forum.effectivealtruism.org/posts/LmjsyFvQavLQnYSTp/the-redd-framework-for-reducing-deforestation-and-mitigating

Some upward nudge to this score because fo the rethink report

Should be straightforward; there is plenty of literature, and redteaming is normally easier than working frmo a blank page

There is lots of literature on REDD+ (albeit less in EA)

We are well positioned for climate work

Implications if we are in a simulation

If we're living in a simulation, what are the implications for ethics and optimal philanthropy.

Arguably not that important -- seems unlikely that anyone will really make different decisions on the back of this research. However conceivably it might trigger some new strands of work. (Note: if the work were "Convincingly claim that we almost certainly *are* living in a simulation, and therefore should act differently in the following ways", this would arguably be more important, although even most people would likely find the claim a bit difficult to believe, and would probably only allow it to influence them modestly. However this piece of work isn't about arguing how likely this is; it's more about saying, *if* we are living in a simulation, what does that mean.

Requires thinking independently about very weird things, without much existing literature to serve as a guide

Seems that very few people are thinking about this

No particular reason to think we are well positioned for this (except that as members of the EA community, our people are more exposed to these ideas than most)

Success stories in tech governance

Success stories in tech governance e.g. ozone

Could be quite broadly applicable. Also arguably somewhat tenuously applicable

This seems pretty straightforward, although standard impact assessment difficulties will no doubt arise (eg considering counterfactuals and assigning credit between different actors)

We don't know of anyone else working on this, but it seems like that sort of thing which other people would have thought about

No particular reason to think we are well positioned

Due Diligence by Roodman outlines how microfinance tends to focus on microcredit ebcause it's more popular. However it seems (Poor Economics by Banerjee and Duflo) that microinsurance is more valuable

Assuming this is found to be impactful, this could be a useful insight

There is plenty of relevant literature

There is plenty of relevant literature

We are well positioned for this

Investigating whether we agree with GiveWell's modelling of insecticide resistance for AMF

Could be a fairly material update to one of the main GiveWell-recommended charities

Nothing is particularly difficult here, in the sense that GiveWell has been transparent about their reasoning and set everything out in spraedsheets which are in the public domain. However the construction of the spreadsheets is aprticularly thorny

Not only is GiveWell (obviously) thinking very hard about this, but they have invited the wider EA community to scrutinise their models

We understand GiveWell's models very well, and have been reviiewing them for some years

Seems likely that this should be a high impact cause area. Apart from HLI, it's not getting a lot of attention in EA

There seems to be nothing particularly more difficult about this than anything else

Plenty of good work from HLI, + a bit from FP, and lots of academic interest in mental health interventions

We are well positioned for this style of analysis

There's a good argument for why this is interesting from the persepctive of responding to critiques from Angus Deaton et al. HOwever beyond this, it's unclear how this is going to be useful -- it would be surprsing if GiveWell decided to update their models to incorporate these considerstions, which means it's unlikely to feed through to givewell's thinking

There are multiple nuanced consdierations

We don't know of anyone else working on this in a GiveWell/EA context, but there is substantial literature on the link between clientelism and aid

Our team has good backgroudn on this and has already been amking strong progress

Recreate GiveWell's models in Guesstimate or Squiggle

Sounds cool, because it addresses an issue, which is that GiveWell's models vary signficantly in their uncertainty (at least when it comes to comparing deworming with other interventions). However once the work is done, it seems unlikely it would lead to different conclusions. It might help to highlight how much uncertainty there is about deworming, but those who are familiar with GiveWell's work already know this

Conceptually straightforward -- just take the models which are already there and redo them in another format

Don't know of anyone else interested in doing this

We know GiveWell's models well

Explore the relationship between climate and conflict (this feeds into "Calculate a SCC which includes tail risks")

Form a view on the likelihood and severity of conflict/war; Halstead has some useful content on this: https://drive.google.com/file/d/14od25qdb4sdDoXVDMoiSrTwuzYAMSpxK/view

Understanding indirect effects such as this helps with comparing climate with other causes

Getting some sort of handle on this should be tractable. Quantifying tail risks is hard

There is relevant literature on the topic. This more quantitative exploration is more unusual

We are fairly well positioned on climate work

Explore the relationship between climate and food systems (this feeds into "Calculate a SCC which includes tail risks")

Climate change could interfere with food systems in several ways (gradual climate change -> climate not suitable for crops ; extreme weather events damage crops ; climate leads to pathogens which damage crop). A really valuable piece of research would form a view on how likely these outcomes are and how severe. Again Halstead's work has relevant content on this

Understanding indirect effects such as this helps with comparing climate with other causes

Getting some sort of handle on this should be tractable. Quantifying tail risks is hard

There is relevant literature on the topic. This more quantitative exploration is more unusual

We are fairly well positioned on climate work

Form a view on how tipping points should be incorporated in the SCC (this feeds into "Calculate a SCC which includes tail risks")

There is some literature on this arleady. Climate change could cause more forest fires, turn soils from a carbon sink to an emitter, interact with the oceans, and the arctic permafrost. It would be interesting to see which tipping points are most material, and which studies account for them. (Again Halstead has some content on this)

Understanding indirect effects such as this helps with comparing climate with other causes

Getting some sort of handle on this should be tractable. Quantifying tail risks is hard

There is relevant literature on the topic. This more quantitative exploration is more unusual

We are fairly well positioned on climate work

Is Good Food Institute high impact as viewed purely through a climate lens

Risk that GFI might be increasingly well funded now, so not sure this is a top priority

Slightly downweighted this assessment based on an unresearched impression that GFI does good work but is fairly well funded now

Expand AMF CEA model to reflect the harm to the mosquitoes

Quick model (all numbers based on guess/off the top of my head): No of mosquitoes killed per night per net (3) x duration of net (1.5years, c500 days) x duration of suffering per death (1 hour) = 1500 hours of insect suffering per net. (ignores any other insects/whatever that might be killed by the nets, ignores ecological impacts). Number of nets to save a life = c1000. So c1.5 million hours of insect suffering per human life saved. To try to keep the model contained, the scope of this task is just limited to the direct suffering to the insect caused by the insecticide. Several other considerations could be incorporated (e.g. effects on ecology) but those are out of scope for this task. Insects are being prioritised here over (say) schistosomes because of (a) recent evidence [1] of insects feeling pain, which we don't have for schistosomes (as far as I know). Would also be good to assess the impact of economic development (eg in the GiveDirectly context) on Wild Animal Suffering, but this is more intractable. [1] https://forum.effectivealtruism.org/posts/yPDXXxdeK9cgCfLwj/short-research-summary-can-insects-feel-pain-a-review-of-the

If a mosquito has as much moral value as a human, then the very quick (~made-up) model arguably suggests that the mosquito suffering cancels out the human life saved. However, it seems more likely that the insect has several orders of magnitude less moral value than a human, so this will likely be a rounding error

Most of the work is straightforward and just involves leveraging an existing GiveWell model. The hard bit is determining the amoutn of moral value to give to the insect.

It doesn't seem that anyone else is looking at this?

We are well positioned for this, as we know GiveWell's models well, and we have given lots of thought to moral weights

This is applicable across lots of areas of research

The research is quite interdisciplinary and includes lots of subtle considerations

There is a huge amount of work on discount rates already, e.g. in the climate arena. Arguably some neglectedness in terms of how discount rates apply in an EA context, although even this is questionable

We have enough finance experience, which means relevant background thinking about discount rates

Suicide/mental health helplines

Likely ineffective from a preventing suicide perspective
Could be effective from the perspective of increasing short term wellbeing, assuming cost per call is low enough

Somewhat lower importance, factoring in the expectation that this has a good chance of being ineffective

Should be relatively straightforward to perform, although there might be an absence of hgih quality evidence, which may mean that the research might need a few more judgement calls

The amount of academic literature about this seems to be modest, as far as can be seen from not having reviewed it carefully yet. There are some studies conducted within charities, although these appear to be superficial and small quantity, as far as can be seen (not having done a full review yet)

We are well positioned for this

Can we find out more about infant mortality from heat? Even if the total numbers of infant deaths are lower, the DALYs saved are so much higher per person that it might be material?

It prboably is true that the difference between infant deaths and elderly deaths could make a difference, but overall it seems unlikely that the bottom line impact will be huge, given that the elderly are probably more affected by heat stress than infants (source: guess). Furthermore, it seems more likely that adding in massive soruces of mortality/impact such as tail risks is much more relevant than tweaking the model with regard to age.

Unclear whether we can get the data we would need to do this well, but it should be possible to come up with a reasonable esimate of the age distribution of deaths

We don't know of anyone else working on this, but lots of people think about climate models, so maybe someone else is thinking about tihs? (on the other hand, given that Bressler's mortality model seemed, oddly, to be ground breaking, maybe it really is neglected?) Bressler just counts number of deaths, doesn't factor in age

We have fairly good understanding of climate models in house

To what extent does work on “nearterm” AI safety considerations help with existential AI safety/alignment

One possible failure mode with various types of policy intervention is that a policy proposal gets watered down (e.g. through making compromises with different factions) and ends up looking a bit different to the original proposal. I.e. maybe a think tank (or someone) aimed to bring about a policy change that was trying to achieve (EA-oriented) AI alignment, but it might end up getting watered down and turning into neartermist AI safety (e.g. algorithmic bias, etc). In case of this, it could be valuable to understand whether this constitutes failing gracefully (i.e. it's almost as good or half as good as the original intent), or if it has close to none of the value of the original proposal.

Overall seems pretty intractable. A relatively tractable first step could be spend some time redefining the problem more clearly, and then conduct some sort of expert survey; this could be a publishable and moderately tractable first step -- it might not fully answer the issue, but it could be reasonable traction.

We don't know of anyone else working on this, but it's totally possible someone is without us knowing

We aren't particularly well positioned on this in the sense that we have done relatively little AI alignment work; on the other hand we are well connected with people in the space

Cause area exploration: Community-led Video Education

We only know of one charity working on this (Medical Aid Films). This [1] is a pre-pub notice for an RCT evaluating this intervention. The metrics they are following are (a) the proportion of women who report a GDM diagnosis by 32 weeks of pregnancy and (b) glycaemic control (fasting glucose and HbA1C) in women with GDM at ~34 weeks of pregnancy. These are meaningful health metrics, but they are instrumentally valuable, not intrinsically. This one [2] is just assessing health-related knwoledge, which is even more distant from ultimate outcomes. [1] https://trialsjournal.biomedcentral.com/articles/10.1186/s13063-021-05435-x [2] https://journals.sagepub.com/doi/full/10.1177/17455057221104297

Low score captures the concern that there doesn't seem to be good enough evidence of effectiveness (based primarily on quick google scholar search) although at least there is some literature relating to Medical Aid Films.

Should be no harder than a typical GiveWell-esque analysis. Could argue that it should be treated as low tractability, in the sense that doing this properly might require us to go and conduct a RCT. But marking this score down for that might be double counting

There is already some releavnt literature on Medical Aid Films, but more is needed

We are well positioned for GiveWellian work. If the work were to be expanded to involve conducting an RCT, then we should lower the score, but I didn't lower the score because I thought that would be double counting

Cause area exploration: soil degradation

Hard to say how important this is

Probably no reason to believe it's materially harder than any other cause area

There's quite a lot of interest in this

We would probably be better positioned for this if we had more biology knowledge in hosue

Hard to say how important this is

Probably no reason to believe it's materially harder than any other cause area

There's quite a lot of interest in this

We would probably be better positioned for this if we had more biology knowledge in hosue

Need comments from IG, plus further review

Need comments from IG, plus further review

Need comments from IG, plus further review

Need comments from IG, plus further review

Nobody (there was some previous work on this, but it changed direction and the original question is still unanswered)

For donors who want to donate to a tree-planting charity (e.g. we have tried to persuade them to support something else, but they are focused on trees), what is the best choice?

It seems that tree planting is going to underperform other interventions anyway, so knowing which si best is still only going to cause donors to reach a suboptimal outcome. However there does seem to eb material variation in tree planting opportunities, so there is some value in guiding donors to the higher impact options. This may be somewhat offset by the fact that some donors use them for offsets, so if they choose something cost ineffective, they just pay more (of course, this assumes that they are basing their paymnet on a legitimate model of cost effectiveness of tree planting)

Trees are actually rather complex!

Not neglected. Eg gold standard, verra etc consider this. Also Giving Green have written about this

We are well positioned for climate work

https://docs.google.com/spreadsheets/d/1joMWic13FrR_H97Rq6AehLVRVq77eNeuMzZpmkPXUpU/edit#gid=1943926155

https://colab.research.google.com/drive/1uMVM0mkeWZRuFk6LI2VMRDpTPftCgLHM?usp=sharing&authuser=1#scrollTo=2Hom9zMwHH9Mhttps://colab.research.google.com/drive/1uMVM0mkeWZRuFk6LI2VMRDpTPftCgLHM?usp=sharing&authuser=1#scrollTo=2Hom9zMwHH9M

SoGive house view on bio anchors

The question of timelines is quite important to the question of how much weight to place on AI alignment/safety work

In some ways this is tractable -- it is simply a case of reading carefullly work which has already been done. On the other hand the work is lengthy and nuanced

Several other people have thought carefully about this, not just Ajeya/Open Phil themselves, of course, but also various other commentators on less Wrong.

We aren't particularly well positioned on this in the sense that we have done relatively little AI alignment work.

Cause area exploration: biodiversity

A bit hard to say how important this is until we've done the work, but loss of biodiversity might have material implicatinos for humans, and might also have material intrinsic importance

This is probably not straightforward especially if the work involves a moral weights component, to decide about the intrinsic value of species not going extinct

There's a lot of interest in this

We would probably be better positioned for this if we had done more on animals, or had more biology knowledge in hosue

Value of cancer research/biomedical research

Expert survey would be good work, but it would take too much time, so repace with desk-based research

Seems highly likely we will conclude that cancer research is not high impact

It's clear that there's a lot of moneygoing into cancer research, but it's not clear how tractability it is (e.g. maybe the flood of money is making it more tractable, conceivably)

Few people are thinking about whether to donate to cancer research (as construed in EA terms) but research does exist about the cost effectiveness of cancer research

We already have some relevant thinking to build on

Expand cost-effectiveness models for (say) GiveWell recommended charities (and others?) to reflect ecosystem effects (e.g. what's the impact of killing lots of mosquitoes on the wider ecosystem)

Quick back of envelope model, just relates to mosquitoes, suggests that this is immaterial (Should be considering, e.g. worms as well) (all numbers based on guess/off the top of my head): No of mosquitoes killed per night per net (3) --> ~1000 mosquito deaths per year per net. No of nets in the world = 1 billion (rough guess, probably slightly high) --> 1 trillion deaths per year. Total number of mosquitoes in the world = 100 trillion (source: googled it quickly). Suggests that nets are killing 1% of mosquitoes per year. So probably not that material

It seemed this research project was being overly rewarded for being neglected -- it's probably neglected for good reason

Quick model (see brief description) suggests this will be immaterial

Should be fairly straightforward

Nobody seems to be researching this, probably because everyone else has done the same quick calculation; seems unfair to nudge the overall rating as much as this because of neglected, given that the neglectedness is probably for a good reason

We are well positioned for this, as we know GiveWell's models well

I need to check in with Jeremy and Ishan to be clear on what the current scope of the work is before it can be scored