Announcing the Rethink Priorities Cross-Cause Fund
TL;DR
Today, based on our multi-year prioritization research, we launch the Rethink Priorities Cross-Cause Fund (CCF). The fund pools donors’ contributions and allocates them to high-impact giving funds across Global Health and Development, Animal Welfare, and Global Catastrophic Risks.
Key highlights from this post:
We believe that strategic cross-cause prioritization is an important step in doing good at scale.
This fund is for donors who want their donation to go where a marginal dollar is likely to do the most good across cause areas, all things considered.
We modeled key uncertainties that matter for cross-cause prioritization: moral weights, time discounting, risk attitudes, aggregation across ethical views, AI-related uncertainty, and empirical uncertainty within each giving opportunity.
We present the current recommended allocation of marginal resources across high-impact funds in each of the three cause areas mentioned above.
In addition, we’re introducing you to the first version of the Donor Compass, a tool to help donors explore cross-cause giving, powered by our cross-cause prioritization model. It is a short quiz that outputs custom giving allocations based on the user’s moral and empirical assumptions.
You can dive deep into our rationale and methodology for the CCF in the announcement below, or go straight to our dedicated website to engage.
GO TO THE CROSS-CAUSE FUND WEBSITECross-cause prioritization in effective giving is underdeveloped
To truly do the most good we can, we need to find a way to assess interventions that span vastly different areas of philanthropy.
In an ideal world, we’d do cross-cause prioritization, using a set of universal benchmarks to evaluate specific interventions across a wide range of cause areas. Cross-cause prioritization is extremely difficult, though, because of the many philosophical and empirical complexities involved. To understand why, consider what this kind of comparison actually requires.
Giving is full of moral and empirical uncertainty
When trying to truly ensure we’re doing the most good we can across dramatically different causes, we must — implicitly or explicitly — rely upon a host of moral (or, normative) assumptions regarding questions like:
How do we weigh different kinds of doing good, such as saving lives, alleviating suffering, and promoting economic opportunity?
Should we weigh near-term outcomes more than long-term outcomes?
Should we give disproportionate weight to avoiding causing harm, even if it means sacrificing some upside benefits?
Do we restrict our concern to humans, or extend it across all sentient beings, including various animal species?
Moreover, if we have competing intuitions about answers to any of the above questions, how can we navigate these tensions and determine how to act?
And once we’ve come to some conclusion about these normative and meta-normative questions, we run into many empirical difficulties, such as:
How do we estimate the marginal effectiveness of any given opportunity in any given cause area at any given moment?
How much funding can different interventions absorb productively?
How should uncertainty about the future — in particular, around the potentially transformative impacts of AI — change our assessment of particular causes?
Every allocator faces these questions to some degree. But cross-cause prioritization is unique in demanding that they all be made explicit at once: you simultaneously need within-cause cost-effectiveness estimates and cross-cause conversion factors and a framework for handling moral uncertainty.
Yet EA has historically invested little in cross-cause prioritization. According to our analysis, only 2% of EA prioritization resources were allocated to cross-cause prioritization in 2025. Consequently, the community has lacked the research and advising infrastructure that cross-cause prioritization demands.
How EA has historically handled this
Given these difficulties with evaluating specific opportunities head-to-head across cause areas (cross-cause prioritization), EA has historically tended to take a two-step approach:
Cause prioritization: using heuristics to divide resources among pre-defined cause areas, such as Global Health, Animal Welfare, and Global Catastrophic Risks.
Such heuristics include evaluating the cause area as a whole by its perceived importance, neglectedness, and tractability.
Cause areas are typically defined by their beneficiaries (e.g., humans living now, animals, people in the far future).
Within-cause prioritization: once a certain amount is going to be allocated to a pre-defined cause area, conducting highly detailed research to assess the most impactful interventions to fund in that area, based on its internal standards of evidence and metrics.
This two-step process accounts for the vast majority of EA prioritization effort (98% in 2025). This imbalance is partly historically contingent, as EA started by focusing heavily on global health interventions, then added Global Catastrophic Risks and Animal Welfare as cause areas. The process was largely driven by the availability of research and by the cities and academic communities where early EAs lived, rather than by systematic cross-cause comparison. The community has now largely reified these cause areas: funding bodies, research teams, and grantmaking processes are organized around them, and the dominant approach is to allocate funding by cause area first, then optimize within each.
Accordingly, much of the community debate focuses on cause prioritization or within-cause prioritization. But little effort actually goes to the question that arguably matters most for impact: how can we optimize among interventions across causes?
Why thinking in cause areas isn’t enough
Why not just pick the best cause and fund it?
An often-heard claim in the EA/impact space is that one cause area is clearly superior, so we should concentrate all resources there. We are highly skeptical of this view and believe that concentrating all resources on a single cause requires an extremely high degree of confidence in our normative judgments about moral weights, the effects of our actions, and approaches to risk. We don’t think this is warranted. Our models are inevitably imperfect, and diminishing returns and shifting risk profiles mean the relative value of additional funding in any area changes over time. Epistemic and normative humility both point toward diversification rather than concentration.
At the end of the day, interventions are what’s funded — not causes
What gets funded is not “global health” or “animal welfare” in the abstract, but specific interventions: distributing bednets, lobbying for cage-free legislation, and funding biosecurity research. Yet much of the allocation to individual interventions within a cause area is determined by our rough picture of the cause area as a whole, even though interventions within cause areas differ dramatically in their cost-effectiveness, their risks of having no effect or causing harm, and their impact time horizons.
If these dimensions are important to donors, then the amount that’s given to a specific intervention should depend on its own risk profile, cost-effectiveness, timing, and so on, independently of how others in the same category fare. As it stands, the merits of a specific intervention can be overshadowed by the characteristics of other interventions in that cause area, shaping the decision-maker’s view of the cause area as a whole.
Diminishing returns also occur at the intervention level, and cause-level allocations based on average or historical cost-effectiveness can become stale as the marginal cost-effectiveness of interventions shifts relative to others — potentially necessitating a reallocation that the two-step process might not trigger quickly enough.
What would address the prioritization problem
We believe that the problems above all point in the same direction: we need to compare and prioritize specific interventions across cause areas, rather than treating the cause-level allocation as fixed and optimizing only within each silo.
We need explicit modeling
Without modeling, anyone trying to allocate effectively across causes must hold a bundle of moral weights, risk preferences, empirical estimates, and decision-making algorithms in their head at once, and somehow apply them consistently across every option. In practice, this is impossible. People fall back on heuristics, intuitions, and whichever considerations happen to be most salient. This produces allocations that are opaque (even to the person making them) and that can’t be easily scrutinized, challenged, or improved.
By contrast, explicit modeling forces these assumptions into the open. When we specify our moral and empirical inputs in a model, we can see exactly how each one shapes the output. We can ask: what happens if I discount the far future more or less? How much does the allocation shift if I value animal welfare 2x more? How consequential are disagreements between two people about moral weights for their recommended portfolios? Being explicit makes it possible to productively and transparently disagree about important normative issues, quantify their impact on decision-making, and identify a clearer path forward for allocating scarce resources.
Of course, models don’t capture everything that matters. But we think they’re worth building because there’s little reason to believe the implicit models running in someone’s head are doing any better. Putting our uncertainties on the table, naming our assumptions, and testing how the recommendation moves under different inputs is the most honest way we have found to prioritize across causes that don’t share a common unit of measurement.
The barriers to cross-cause prioritization are lower than they once were
Though cross-cause prioritization is difficult, several recent developments have made the challenge much more approachable. RP’s cross-cause prioritization researchers have a track record of producing foundational, decision-relevant work when given resources:
The Moral Weight Project, one of the first systematic efforts to compare welfare capacities across species, provides a framework for human-animal moral weight conversions.
The CURVE sequence developed tools for reasoning about how risk attitudes, time discounting, and empirical uncertainty could affect cause prioritization.
The CRAFT sequence, including the Portfolio Builder and Moral Parliament tools
All of the above underlie our Cross-Cause Fund and Donor Compass.
The space has also benefited from:
Organizations like GiveWell, Animal Charity Evaluators, and more that publish their cost-effectiveness analyses.
Philosophers developing sophisticated procedures to incorporate risk aversion and normative uncertainty into decision-making.
Developments in AI that have made complex modeling endeavors much more feasible.
As such, the barriers to doing cross-cause research are substantially lower than they were even a few years ago.
Our solution: an explicit and transparent cross-cause prioritization model
The model
Our Cross-Cause Prioritization model allows for:
Comparing top giving opportunities across cause areas and success metrics. Global health funds are measured in life-years saved, disability averted, and income gains; animal welfare funds in years of suffering avoided across species; global catastrophic risk funds in expected lives saved from mitigated catastrophes. The model takes these different success metrics, applies explicit moral weights, and lets us compare a malaria net, a cage-free hen campaign, and an AI policy grant on a single yardstick.
Aggregating across moral and empirical worldviews, rather than picking one and ignoring the rest. Reasonable people disagree about how much weight to give animals, how to value future generations, how risk-averse to be, and how plausible different ethical frameworks are. Rather than pick one set of answers and pretend the others don’t exist, the model represents multiple worldviews simultaneously and combines them using up to nine aggregation methods (e.g., Nash bargaining).
Handling the structural features of giving that intuition easily misses, such as diminishing marginal returns, delayed impact, and risks that interventions might have no effect at all or might backfire.
In short, we can:
Explicitly model key uncertainties that matter for cross-cause prioritization, including moral weights, time discounting, risk attitudes, aggregation across ethical views, AI-related uncertainty, and empirical uncertainty within each giving opportunity.
Cover a wide range of funds that give to specific interventions, from malaria nets to shrimp stunning campaigns to lobbying for AI regulations, and put them on a common scale.
Make assumptions visible, so disagreements can be located precisely and debated.
Giving opportunities currently included in the model
In practice, many donors give through funds rather than to individual interventions. Our model evaluates funds based on the marginal cost-effectiveness of the interventions they support, so the fund-level allocation reflects intervention-level analysis. We evaluate funds rather than specific interventions at this stage, both because funds themselves conduct intervention-level vetting and because we do not yet have the capacity to independently assess a wide range of individual interventions. And in contrast to the highly abstract cause areas, we often have information to evaluate the cost-effectiveness and diminishing returns rates of each particular fund. Finally, we believe these funds can absorb a large amount of money productively and divert it to the highest-priority interventions within their remit.
Nevertheless, we are actively working to broaden the range of opportunities covered by the model, including funds focusing on climate change mitigation and democracy preservation, as well as specific interventions.
The model currently allocates donations across the following funds:
GiveWell All Grants Fund: funds a broad set of global health and development interventions with strong evidence of cost-effectiveness
Lead Exposure Action Fund: worldwide lead exposure mitigation efforts
EA Animal Welfare Fund, The Navigation Fund—General Farm Animal Fund, and The Navigation Fund—Cage-Free Fund: interventions to improve the welfare of farmed and wild animals
Longview Philanthropy’s Nuclear Weapons Policy Fund: strategic and neglected giving opportunities in nuclear safety
Longview Philanthropy’s Frontier AI Fund: key opportunities in AI safety
Sentinel Bio: biosecurity and global catastrophic biological risk reduction
Please note that this model is a perpetual work in progress, which we aim to regularly update as new cost-effectiveness data becomes available, as our research on cause priorities develops, and as we expand coverage to new funds and cause areas.
The Cross-Cause Fund
Powered by the cross-cause prioritization model described above, the Rethink Priorities Cross-Cause Fund pools donations from multiple donors and allocates them across a curated set of high-impact giving funds to get the most impact per dollar. You donate, our research and operations teams do the rest. (We do not charge any fees for managing the Cross-Cause Fund.)
Who is it for?
The CCF is for donors who want their contribution to go where a marginal dollar is likely to do the most good, all things considered, and across cause areas, but who do not have the time to do the necessary research and ongoing monitoring themselves.
This fund might be a great fit for you if you:
want your giving to span multiple high-priority interventions (currently represented by the funds from the GHD, AW, and GCR cause areas, with new funds being added in the upcoming months)
prefer deferring to a research-vetted process rather than monitoring funds, interventions, and cost-effectiveness data yourself
take moral and empirical uncertainty seriously
The Cross-Cause Fund might not be a great fit for donors who:
may be inclined to give to a single cause area and prefer to direct their full donation there
might be bought into a single ethical view
prefer hands-on control over each grant when it’s made; these donors may be a better fit for the personalized Rethink Priorities Cross-Cause Donor-Advised Fund
How does it work?
When you donate to the RP Cross-Cause Fund, your contribution goes into a single pooled account held at CharityVest, which is managed by RP. We allocate pooled contributions across our recommended charitable funds on a regular basis, following our most current recommended split (see below). You can give once, on a recurring schedule, or at any time you choose. You’ll be able to see the grants RP has made from the fund and receive regular impact reporting. For donations over $500,000, you can also submit a preferred allocation (generated via our Donor Compass tool) by emailing us.
CONTRIBUTEQuestions about contributing, or want to discuss your giving in more detail? Contact us at fund@rethinkpriorities.org to learn more or arrange a call.
Our current recommended allocation across cause areas
Below is the current split (by fund and by traditional EA cause area) that we apply to any donations made to the RP Cross-Cause Fund. These splits are determined by running the cross-cause prioritization model based on:
the current cost-effectiveness data for each of the funds included in our model
our current foundational assumptions
multiple, research-backed approaches to navigating moral uncertainty
The current split will always be available on the Cross-Cause Fund homepage.
It was last revised on May 18, 2026, and we’re tracking all updates and improvements to the model here. (This allocation will likely change over time as we adjust our inputs and refine the model.)
Our recommended allocation (based on an assumed budget of $200M) reflects our best all-things-considered judgment on where a marginal dollar does the most good among the funds we considered. As described above, the underlying model explicitly accounts for empirical uncertainty about interventions, moral uncertainty across ethical frameworks, and different attitudes to risk. It considers cost-effectiveness at different funding levels, welfare tradeoffs between humans and animals, and how uncertainty about AI should affect other cause areas.
Donor Compass
What is it?
We are also presenting the first version of the Donor Compass, a quiz-type tool powered by our cross-cause prioritization model that outputs a custom cross-cause allocation based on the user’s beliefs.
It is a tool for anyone looking to explore cross-cause giving, particularly those curious to see how their moral and empirical beliefs will shape downstream allocations.
You can use it to compare RP views and your own, resulting in a personalized allocation recommendation that reflects your moral and empirical commitments.
How to use it
The Donor Compass has two modes:
The default is a short quiz that you can take in just a few minutes. It walks you through a selection of four key questions with preset values, and generates a personalized allocation across our funds. Each answer can also be fine-tuned with more specific inputs for more precise control.
Once you’re familiar with the inputs and want more control, there’s a power-user mode that lets you see all inputs side by side and choose between different worldview aggregation methods (or combine multiple). The definitions for each input are found in the notes section of this spreadsheet, which contains the default inputs for our Cross-Cause Fund.
Here is a walkthrough of how the tool works with real-world examples.
Methodology behind our model and tools
Our methodology has three components: (1) how we select funds for inclusion, (2) how we estimate each fund’s marginal cost-effectiveness and how it changes with funding influxes, and (3) how the cross-cause prioritization model combines those estimates into an allocation across funds. This section provides a brief overview; more details are available in the document linked here.
Fund selection
We aimed to include at least one fund per major cause area focused on by the EA community (Global Health and Development, Animal Welfare, and Global Catastrophic Risks), and applied four further criteria: fund quality (we included only funds we were confident would lead to high-impact grants, and excluded areas we lacked the experience to vet); data availability (we needed cost-effectiveness data we could obtain or model from public sources); decision relevance (we shaped the initial list around the giving decisions we expected to be most useful to early users); and our legal restrictions as a 501(c)(3), which require us to exclude political giving entirely.
Fund-by-fund cost-effectiveness estimation
For each fund, we estimate two things: how cost-effective it is at the margin today, and how quickly that cost-effectiveness declines as the fund receives more money. Our process is to build a cost-effectiveness framework for each fund, solicit data directly where possible, fall back on in-house estimates from existing research and expert views when data isn’t available, and simulate the fund’s marginal impact and risk profile across each dimension. Separately, we construct diminishing returns curves informed by internal assessments, third-party experts, and fund managers.
Funds in the same cluster share outcome metrics. These metrics may be expanded in the future to cover more cross-cause impacts of each fund, but at the moment include:
Global Health and Development funds are scored in years of life saved, years of disability or illness averted, and income gains (standardized as one-year income doublings).
Animal Welfare funds are scored in years of suffering avoided, broken out by species (chickens, fish, shrimp, and non-shrimp invertebrates).
Global Catastrophic Risks funds are scored in expected years of life saved from mitigating catastrophes.
All three cluster models then output a standardized 6-by-8 grid: six time periods (from 0–5 years out to 500+) by eight risk profiles (from risk-neutral expected value to discounting optimistic scenarios to giving significant extra weight to avoiding harms). Moral weights and cross-cause comparisons are applied downstream of this grid, which keeps the underlying impact estimates transparent and lets us swap in different moral views without rebuilding the fund-level models.
Data Viewer
We built a Data Viewer, if you want to inspect the cost-effectiveness data that feeds into Donor Compass recommendations and Cross-Cause Fund allocations. To help you understand the data, we also created a brief overview explaining how to read it.
Defining worldviews
The model takes the fund grids and generates a scalar score for how well each fund performs under each of several worldviews. A worldview is a coherent bundle of moral and empirical commitments, along with an approach to acting in the face of risk and uncertainty. The components that define a worldview in the model include:
Moral weights. How do we value a year of human life relative to a year of disability averted, a one-year income doubling, or a year of animal suffering (for chickens, fish, shrimp, and other invertebrates)? The base unit is one human life-year (weight = 1.0).
Time discounting. A discount factor (0–1) for each of six time periods (0–5, 5–10, 10–20, 20–100, 100–500, and 500+ years),[1] reflecting how much weight a worldview gives to future impacts relative to immediate ones.
Risk profiles. A risk profile that determines how a fund’s distribution of cost-effectiveness collapses into a single score, ranging from risk-neutral expected value to heavily risk-averse.
AI-risk discount. For worldviews that assign weight X to a near-term AI catastrophe, an X% discount is applied to non-AI-risk fund scores, reflecting the chance that AI takes an extreme action that reduces the impact of those interventions.
Because we might be torn on one or more of the above questions, we can assign credences to represent our degree of belief in several competing worldviews. These credences are numbers between 0 and 1, and they collectively sum to 1 across all worldviews.
Outputting an allocation
Once these worldviews are defined and our credences in them are set, the model runs in four steps.
Score each fund under each worldview by applying that worldview’s moral weights, time discounts, risk profile, and AI-risk discount to the fund’s grid.
For each of the nine aggregation methods, allocate the total budget, in $2M increments, to each fund based on how each fund fares under each worldview and our degree of belief in that worldview. After each $2M increment is distributed between funds, each fund’s score is adjusted to reflect the extent to which the allocation reduced its marginal returns.
Because no single aggregation method is universally accepted, we assign a credence between 0 and 1 to each of the nine methods.
Finally, we take our credences in these aggregation methods and the allocations each recommends, and produce a weighted-average recommendation for how the budget should be allocated across funds.
The default inputs that we use can be found here. A more detailed description of the methodology behind assessing the funds and computing the allocations can be found here and here.
Future cross-cause prioritization plans
Our cross-cause prioritization model is a living tool. While it represents our best current methodology, we acknowledge that it does not yet capture every factor relevant to estimating impact. For example, we are not currently modeling certain interventions, such as alternative proteins, which we intend to add in the near future.
A detailed discussion of the model’s current limitations is available in our methodology documentation and the Cross-Cause Fund’s FAQ section. We are committed to rigorously prioritizing these gaps and incorporating the factors most relevant to improving model accuracy.
The pace and depth of this refinement work will scale with the fund’s size. As the volume of donations increases, the value of marginal improvements to our estimates grows correspondingly. We will invest in model development in proportion to the resources we steward.
The short-term improvements will likely include:
Covering areas beyond global health and development, animal welfare, nuclear risk, biosecurity, and AI safety, e.g.:
Democracy protection (non-partisan): We have commissioned an initial review of this area. Full incorporation of a vetted fund is likely more than three months away, but we expect to share interim findings with interested donors before then.
Climate change: We’re about to complete a project assessing the value of climate interventions relative to global health interventions. This analysis should allow us to produce cost-effectiveness conversions for climate-focused funds. We anticipate this will be particularly relevant for mainstream donors who wish to consider our work but also include climate in their cross-cause portfolios.
Vetting more funds for inclusion, e.g., Ambitious Impact funding circles (Mental Health, Meta Charity, Seed Network, Global Health, Strategic Animals), Giving Green, The Life You Can Save, AI Safety Tactical Opportunities Fund, Astralis Foundation, Survival and Flourishing Fund, Animal Charity Evaluators’ (ACE) Recommended Charity Fund, Senterra Funders, Macroscopic Ventures, EA Infrastructure Fund, Coefficient Giving Funds (e.g., Global Growth Fund, Abundance & Growth Fund).
Longer-term opportunities could include:
Adding more specific funds within the broad areas of global health, animal welfare, and global catastrophic risks, including:
Digital minds and AI welfare
Economic growth interventions (in low-income and/or high-income countries)
Wild animal welfare
Additional biosecurity interventions
Funds that are giving to “meta” causes that aim to increase the volume or effectiveness of resources flowing to high-impact charities.
We will communicate any additions to our fund coverage through our regular updates.
Support the research behind the cross-cause prioritization model
We do not charge any fees for managing the Cross-Cause Fund, and Donor Compass is free to use. We chose this because we want to enable impactful cross-cause giving with as few barriers as possible.
This also means our research, development, and maintenance work is completely reliant on donations. The Cross-Cause Fund and Donor Compass exist because individual donors believed cross-cause prioritization research was worth funding. If you also find this work valuable, consider donating here. You can also reach out to us at development@rethinkpriorities.org to discuss options for supporting this work.
Conflicts of interest statement
Rethink Priorities (RP) has received grants from Longview Philanthropy, EA Animal Welfare Fund, Navigation Fund, and Coefficient Giving. RP has also done commissioned research for GiveWell and Coefficient Giving. In addition, one RP staff member serves as an advisor to the EA Animal Welfare Fund but is recused from grantmaking decisions involving RP.
RP is not involved in any grant-making decisions for these organizations, and RP conducts independent cost-effectiveness analyses of any funds we decide to include in our Cross-Cause Fund.
Acknowledgements
The Cross-Cause Fund and Donor Compass are projects built by Rethink Priorities’ Worldview Investigations and Interdisciplinary Research Teams, with support from the Development Team and assistance from the Communications Team, which is responsible for final edits, copywriting, website content, and co-authoring this post. We want to thank all the team members behind this work, as well as the contractors and everyone who provided assistance and feedback at various stages of this project.
- ^
The time periods are inclusive of the lower bound and exclusive of the upper bound.
I’ve known about this for a while. I really liked this write-up. I continue to be extremely impressed with Rethink Priorities.
Thanks so much, Marcus, on behalf of the Team :)
I think this is important work, but I want to flag my biggest concern about the process—the imbalance of the backgrounds of the research team and therefore potential for bias. I asked Claude to rank their areas of interest and prior work, and it shows a heavy bent towards Animal Welfare and Global Catastrophic risk.
Half of the research team have been strong advocates for Animal Welfare work in the past, while none of the team seems to have a special interest in GHD. Two of the team were deeply involved in the Animal Moral Weights project itself.
I think the work is impressive, and it’s great there is a new fund, however I think a cross-cause research prioritisation team like this ideally would have been more balanced in makeup. Research team balance on prior opinions is especially important when the project relies on many assumptions and requires subjective assessments at many junctures. In addition potential conflicts of interest here should probably be stated on the website and in a post like this. Now that there is donation money directly involved, I feel the stakes are higher than when the situation was purely research (although the research alone is very influential).
A couple of other less important declarations perhaps could have been made as well. Given that Lead Exposure Action Fund was the only specific GHD intervention selected, they perhaps should have mentioned that they have been commissioned to do research on lead exposure before, and also that a previous RP researcher is now a program associate at LEAF. The RP CEO was also previously a fund manager on the EA animal Welfare fund which allocates 13% of this fund’s money.
All think tanks will have bias to some extent (Political think tanks are even rated on a spectrum) and I think its important to consider this and state where there might be potential bias.
Hi Nick,
I think there is not a single EA organization I would consider unbiased on this question, including ourselves (despite our ongoing efforts not to be). That is exactly why we publish so much of our methodology and our assumptions openly. One of the main motivations for this work is concern about the effect of bias when assumptions and models are implicit or hidden. We would welcome more experts with broader backgrounds being involved in drafting and improving these estimates, which is part of what we hope this kind of public methodology enables.
On conflicts of interest: we aim to be transparent about potential or perceived conflicts of interest, and we state all present COIs in the post above and on the website (here and here), and we will update these when and if they change. RP is recommending these grants independently and no one outside of RP was involved in choosing the funds for inclusion.* You are welcome to review our methodology on how the funds were chosen and provide critique.
On team composition and priorities specifically, a few points: First, the researchers you list are not, in fact, the principal contributors to the project and it’s harder to pin us down even if you stuck to that methodology. For example, Carmen van Schoubroeck was the project lead for all this work, and started her EA career as a global health and development researcher. Some of the specific inputs that in theory could be biased here (like the choice of aggregation methods) are mine and my background involves starting a global health charity, being an animal welfare fund manager, and incubating and providing operational support to a large number of AI projects. And obviously RP works across all of these areas because I, personally, think it’s a good idea. Secondly, Claude does not have access to the work our researchers have done that’s not published, nor what they actually prioritize in their own giving. Third, our global health and development researchers gave feedback on the modeling assumptions, in the same way we involved researchers with other expertise on other cause areas. Having that option is one of the benefits of being a research organization with teams across causes.
More broadly, I would not consider Claude’s opinions about the priorities of our researchers to be a good method of gauging the quality of our work. I think all of the following:
The old adage that “personnel is policy” is relevant to any exercise that’s as complex as this one
I’m confident that there are ways to improve this work, and specific improvements to be gained by bringing in diverse perspectives (I’d be happy to discuss options here in more detail)
Despite (1) and (2) there are larger areas of improvement or concerns about this work than trying to infer cause area orientation from public work of staff and use that to infer potential bias in our choices for the model
Specifically with regard to (1), I can never tell you with certainty no one who worked on this was not subtly unconsciously biased, but one of the reasons people were chosen to work on this project at all is because of my perception of their lack of bias. This is a considerable filter to working on any cross cause work at RP and I stand behind the credibility of every person who worked on this was doing their best to be unbiased. I think this filtering combined with the process we had of having multiple people review work and getting input from funds and cause area experts should lead to less bias. I could be wrong, of course, but I think it would be more productive to make specific critiques of our work and choices because the specific choices are there to see. It’s not costless in time to do this, so I don’t begrudge anyone for not engaging, but you don’t have to infer if you think our animal moral weights are too high, or the cost-effectiveness of an area is too low, you can see what we chose and say what you think is wrong and why.
I truly wish for myself and RP to do the impartial good but we only ever achieve doing so imperfectly. But, primarily, the way we find out is through specific things we get wrong. So, if you have specific issues with inputs or choices that you think privileged one area over others, please point them out. We’d be happy to revise them if they should be improved. The current model surely isn’t perfect (as we acknowledge) and can benefit from specific, thoughtful criticism about our methodology, which we’ve described and linked to on this site.
*Despite the complete lack of involvement of former staff in our decisionmaking, in general I favor more transparency in EA so ill note you are correct that RP has done research on lead exposure before, and in fact, two former RP staff now work on the Lead Exposure Action Fund. I think our ability to model lead was improved because we’d done work on it and this was a pro for us including it, as we felt more confident we could accurately do so by the time of launch. I was also a fund manager for the EA Animal Welfare Fund. We also have former staff at Longview, and some at other parts of Coefficient Giving unrelated to LEAF. We also have former staff or board members at a number of the funds we list that we plan to consider including Astralis, the AI Safety Tactical Opportunities Fund, Giving Green, and Macroscopic Ventures.
Thanks Marcus for the reply. First this criticism is specifically about potential preferences and bias of the researchers. With a project like this with perhaps hundreds of junctures which require subjective decisions, I think that’s a reasonable discussion to have. I don’t think it’s fair to ask me shift the ground and ask to discuss the research methodology, I’m sure there’ll be plenty of great discussion about that. My point was purely concerns about potential researcher bias.
I completely agree with your 3 points, and specifically that there are larger areas of concern than the cause area orientation of researchers—but I still think it’s somewhat important .
This comment seems to twist the point I was trying to make “More broadly, I would not consider Claude’s opinions about the priorities of our researchers to be a good method of gauging the quality of our work” I was not questioning the quality of your work, I think your work is extremely high quality. Yet even the highest quality work can go in different directions, based on the assumptions and biases of the researchers who generate it. In many imprecise fields like development, economics and political science, the highest quality researchers can argue opposite sides. Stiglitz and Friedman both produced high quality work which often ended up at almost opposite conclusions. I would put cross-cause categorisation in a similar-ish category due to the wealth of assumptions required and uncertainties to be reckoned with. So in research like these yes, I think the make-up of the research team is important and open to scrutiny outside of objective criticism of the work itself. You might disagree with me on this one. Assuming people’s backgrounds is open for discussion, I think a Claude analysis of people’s previous work is reasonable if obviously very imprecise and yes potentially misleading.
I could be wrong here, but I feel like I may have been baited-and-switched a little on who are the major contributors here.. I went to your website and just searched the cross-cause prioritisation team, finding a page with photos and bios. You’ve said that the researchers I listed were not the principal researchers - then who were? That Carmen van Schoubroeck led the work isn’t listed anywhere I don’t think.
I agree with the direction but not the strength of this comment “I can never tell you with certainty no one who worked on this was not subtly unconsciously biased” and agree with your opening statement more “I think there is not a single EA organization I would consider unbiased on this question, including ourselves” We are all biased, often more than we think. Most of us (myself included) have a cess-pit of opinions and angles, despite our best efforts to the contrary. I think within EA we often overrate how objective we are, and we can only have so much objectivity based on our past experience and worldview built up over time. This is why I think for a cross-cause prioritisation exercise, it is helpful to start with a team with a wide range of prior opinions or perhaps very uncertain ones.
Hi Nick—just regarding the team page issue, are you thinking of this page: https://rethinkpriorities.org/our-research-areas/worldview-investigations/ ? It has the people listed in your screenshot from Claude.
As a reference, I got to this from CCF page --> support our work --> under the “our team” section. Notably, the link is from this text:
“The Rethink Priorities Cross-Cause Fund sits on top of work done by our Worldview Investigations Team (WIT), and Interdisciplinary Research Team, groups established specifically to tackle the hard questions that most donors don’t have time to engage with: how to compare welfare across species, how to reason under deep uncertainty, how to weigh present benefits against future ones, how to aggregate competing moral views into a single allocation.
The team brings together training in philosophy, economics, statistics, cognitive science, moral psychology, and decision theory . The fund’s allocations also draw on the in-house expertise of RP’s Global Health and Development, Animal Welfare, AI departments, so the cross-cause model is informed by researchers working directly in each area, not just secondary literature. ”
So I’m interpreting that paragraph as saying that WIT work goes into the report, but not necessarily that the WIT team did all the work (in particular, the interdisciplinary research team clearly was also involved, and the second paragraph suggests other teams contributed as well).
Thanks so much @mal_graham🔸 that’s the web page appreciate that! I understand that the WIT doesn’t do all the work, but I think it was reasonable for me to assume that they were the major contributors given that they have been the team publishing the cross-cause work up to now, and were the team linked from the page.
I’m curious about how you model the time discounting more in detail:
As I read it right now, you pick six fixed multipliers for six fixed future time intervals. Am I understanding this correctly?
Under some assumptions that are common in EA that I don’t share such as the time of perils hypothesis, this would mean that, at least for some ways of computing, most utility is in the 500+ bucket.
And then the model would be very sensitive to people putting 0 vs any nonzero number for the 500+ years bucket. Is this the case?
Hi Clara,
Thanks for the good question!
Short answer: yes, you’re understanding this correctly. And, yes, the model’s allocations can be sensitive to putting any non-zero weight in for the value of impacts 500+ years out (but worldview diversification can mitigate this factor’s impact on the outcome).
The model computes a “score” for each fund based on the sum of: (impact in time period t)*(weight in time period t) over all time periods.
In our estimates of the impacts of GCR funds, we do take the approach of estimating the impact of avoiding an existential catastrophe over many generations in the future. As such, if you were to give full weight to further-out time periods, the vast majority of the expected impact of avoiding an existential catastrophe is in the long-run future. So any weight that is meaningfully above zero will make these projects have a large-in-magnitude score, all else equal, compared to putting a weight of zero. (We’re preparing a more detailed sensitivity analysis that we’ll release soon, and which will address your question in more detail.)
Nevertheless, it’s important to note that our recommended allocations are influenced by 14 worldviews, some of which assign zero weight to the far future. As such, we’re not forced to choose between putting zero weight on the 500+ year period vs. some non-trivial amount that swamps the calculations. This creates a meaningful amount of diversification within the model’s results, allowing other factors like moral weights to be influential.
Additionally, there’s an interaction between risk attitudes and the amount of weight that one puts on the far future, such that putting a non-zero amount on the 500+ year period doesn’t automatically recommend you spend 100% of your budget on GCR funds. For instance, if you’re highly risk-averse and put a weight of 1% on the 500+ year period, then you might avoid funding certain GCR funds that have a high enough chance of raising existential risk, because permanently harming the long-run future would have such a considerable impact.
If you’re interested in reading more about the worldviews we’ve used in the model, please feel free to reference this link.
Thanks again for the question!
Hey, thanks for your reply, I found it very informative.
I’m excited to read your sensitivity analysis writeup when you publish it.
Of course! We’re also happy to answer any additional methodological questions you may have in the future
Exciting! Are you eligible for gift-aid in the UK? Seems like it is based around US donors so far (reasonable triage).
Hi Toby, the CCF is not eligible for UK gift aid, I’m afraid, as it’s a US-held fund, so donations are tax-deductible in the US only.
We’re actively looking into workarounds. Theoretically, one can give to a giving vehicle eligible for UK gift aid, and that entity could then re-grant to this fund. Anyone interested can contact us at fund@rethinkpriorities.org to discuss options.
Thanks for your work on this.
Have you considered making recommendations for people who do not have definitive views about the topics above? For example, question 1 of the Donor Compass asks about how much animal welfare whould factor into funding decisions, as illustrated below. I understand each option is represented by a set of point estimates describing welfare comparisons across species, but I do not undorse any particular set. I can see reasonable best guesses ranging from “Only humans matter [in practice]” to “Animals matter, but somewhat less than humans”. So I think the priority should be decreasing uncertainty instead of acting based on a given set of best guesses.
It’s an interesting one Vasco. I prefer the current questions to uncertainty questions which I think are more intuitive for most people. I think it’s important to lean towards questions which are easier to understand, rather than the “best theoretical” questions to help differenciate. Not everyone thinks as deeply as you about these things, and I like the accessibility of the current questions.
Hi Nick. I agree it is important that the questions are easy to answer. However, I would say there are simple ways of letting users express their uncertainty. For the question about how much animal welfare should factor into funding decisions, there could be an option saying something like “I am very uncertain about which of the 4 views above I should pick”, or users could be allowed to give weights to each of the 4 views (instead of giving a weight of 1 to a single view). Then these answers could be used to define distributions for the probability of sentience, and welfare range conditional on sentience. Wider distributions would tend to result in a higher expected value of perfect information.
I like the idea of a 5th option “I don’t know”, but adding weights adds too much complexity for a public question stream I think. For the moral parliament or something deeper like that makes more sense to me
Hi Vasco,
Thanks for the question! We’ve designed the Donor Compass to be more streamlined, but we certainly appreciate and share your moral uncertainty. Our recommendations for the Cross-Cause Fund take into consideration 14 distinct worldviews, which take into account a wide range of animal moral weights (among other variables). You can investigate the assumptions for each here, along with the credences placed on each. If the range of views you place credence on significantly differ, however, you can use the Advanced Mode of the Cross-Cause fund to tailor it to your specific needs. (The definitions of each term can be found in this spreadsheet)
Thanks for the links, Laura. To clarify, I meant to ask whether you have considered making recommendations that specifically target decreasing the key uncertainties that matter for cross-cause prioritisation. For example, decreasing uncertainty about welfare comparisons across species by supporting work like RP’s moral weight project.
I think I see now, thanks for the clarification. We don’t currently include funds/interventions that specifically work on research (to reduce key uncertainties or otherwise). We do think this kind of work is important, and we aim to include more topics (which could include “meta” work and research) in future iterations of the model.
I haven’t dived into the figures/calculation, but I’m pretty skeptical about donating 46% to regular animal welfare projects given AI progress. Is the AI discount just based on x-risk or does it account for the possibility of AI radically transforming the world in such a way that the impact of current charities becomes moot?
Hi Chris,
The way it practically works is a blanket cut on the overall cost-effectiveness. That could be because there is an x-risk, or because the effect of non-AI work is reduced or eliminated. For the current inputs, a 40% discount is applied, which I think is large enough to account for all those possibilities.
But perhaps the phrase “regular animal welfare” is potentially another source of disagreement? The funds we’re donating to are aware of AI’s impact on the world and seem likely to take steps to use AI to improve their outcomes. The current allocation is based on what specific interventions the AW funds are funding but it’s also worth explicitly noting that giving money to AW funds or AW groups isn’t a commitment to pursue the same interventions that exist today indefinitely (this same reasoning also applies to GHD funds). Perhaps current interventions become moot (that’s why we are discounting them) but it’s also possible, for example, that AW work could be more effective in the future because AI makes monitoring welfare vastly cheaper or, say, boosts the efficiency of animal groups.
All that said, if you think there should be a higher discount, you can apply it, but that doesn’t substantially change the results regarding what goes to AW. Indeed, if you increase the discount to 90%, you don’t get much more to AI (and less overall to GCRs as a whole, likely because the discount also applies to biorisk and nuclear weapons). Generally, this is because there are interaction effects between risk preferences, diminishing returns, overall cost-effectiveness, and how the worldviews are aggregated. Making everything else less cost-effective doesn’t change, say, that you have to have a certain risk attitude to take longer-shot bets at success.
How was the 40% figure calculated?
There is definitely some correlation between current competence and ability to adapt to a changing world, though I suspect that there’s also a huge amount of stickiness that undermines this.
Fascinating. Is biorisk depreciating because some of those projects have a long payoff time?
My intuition is that biorisk talent will be more easily redeployable in a world undergoing an AI transition than a lot of the animal talent (more sticky).
This is definitely an approximation, not based on a rigorous underlying model. I informally personally asked several people in the cause prio and AI spaces, including some at labs, to answer the question about what type of discount seemed appropriate and, along with my best judgment, settled on 40%.
I would very much like to improve this input in the future, perhaps through formal surveys or, perhaps, a Delphi panel-type process. Though I think my weakly held intuition here is a bigger possible change isn’t the precise value (which in the current model doesn’t change the result dramatically), but distinguishing between reduced effectiveness overall and probability of an effect going decreasing, with consideration of different time periods when those effects happen in each cause area.
We included biorisk in the AI discount not just because of payoff times but because of the same type of uncertainty you raised about AW. That is, it seems possible the actions people are taking today could be mooted by changes in AI development.
It does seem possible that biorisk talent could potentially be more easily redeployable than AW. I would also add a couple more related additional considerations in what discount to use for biorisk: (i) some pathways to very negative outcomes for AI run through biorisk and (ii) it seems plausible to me that some civilizational hardening measures (i.e. more widely available PPE) seem perhaps more robust to AI uncertainty than some interventions in other cause areas because they act on AI risk itself given (i).
This is why in future versions of this model, I could imagine both nukes and biorisk having a different AI discount since there are clear interactions here between AI and these other GCRs.
I’m really excited about this project and think it arrives at a pivotal moment!
Thank you, Sarah!
Very cool, and will strongly be considering switching my donations if/where I can, potentially even at the cost of tax-effectiveness (in Australia)!
I wonder if it may be helpful to combine this with a moral parliament tool. Personally, my credence in consequentialism is far lower than 60% (which is the credence used on the linked spreadsheet), and my preferred normative theory is not even considered (Ross’ prima facie duties theory).
Hi Benton,
Thanks for the comment! To clarify, our model and recommendations do draw upon the same approaches to moral uncertainty as our moral parliament. Our Cross-Cause Fund draws upon 14 different worldviews (see here for the exact details) and uses a weighted average of recommendations across nine methods of aggregating across them.
Additionally, though our main Donor Compass tool is simplified, you can use a more Advanced Mode here to incorporate moral uncertainty across several combinations of worldviews. (I’m not familiar with Ross’ prima facie duties theory, but hopefully you could represent it adequately in the model.) Then, just like the Moral Parliament, you can specify which aggregation method you want to apply to resolve disagreements between worldviews.