Rethink Priorities

Stated Theory of Change

RP claims to reduce suffering and improve wellbeing by producing rigorous research that changes how funders, policymakers, and organizations allocate resources. Their causal chain runs: identify neglected questions --> produce rigorous research --> communicate to decision-makers --> decision-makers change allocations --> more resources flow to effective interventions. Additionally, they incubate new organizations (via Special Projects) and directly coordinate actors (via convenings like the Animal Advocacy Strategy Forum).

The theory has become more specific over time. Initially (2018) it was "produce good research and hope it matters." After self-identified failure of this approach, it evolved to "produce research specifically commissioned by identified stakeholders who have committed to considering the results." This is a meaningful improvement -- it closes the gap between research and action.

Revealed Theory of Change

RP's actions suggest a somewhat different theory than stated:

What they actually optimize for: Pleasing Open Philanthropy. OP is ~70-80% of cumulative non-SP funding and the primary commissioner of their GHD and animal welfare research. RP's research priorities demonstrably track what OP wants investigated. This is not necessarily bad -- OP funds research they consider important for their own allocation decisions -- but the "independent think tank" framing understates the degree of financial dependence.

Where they actually spend time: Animal welfare research and convening remains the intellectual heart (29% of resources, longest track record, most cited work). AI/longtermism is growing (23%) but newer and less established. GHD is increasingly a consultancy for mainstream funders, not an independent research program.

What actually generates their influence: Three things: (1) the Moral Weight Project and related work that shifts how the EA community values animal welfare, (2) the Special Projects incubator that bootstrapped Apollo Research, Epoch, and other now-significant organizations, and (3) IAPS, which provides an AI policy presence in Washington. The first is RP's clearest intellectual contribution. The second is underappreciated. The third is spinning off.

Divergence from stated theory: The stated theory emphasizes "cross-cause" analysis -- using the same rigorous framework to compare opportunities across animal welfare, GHD, AI safety, and longtermism. In practice, the teams are fairly siloed (Marcus noted "the animal welfare department and the longtermism department don't necessarily need to talk to each other that much"). The cross-cause cost-effectiveness model (CCM) exists but is one tool among many rather than the central organizing framework.

Key Assumptions

Assumption 1: Research changes large funders' behavior.

• Evidence for: RP cites specific examples -- EU animal welfare strategy influencing $1M+ in organizational budgets, OP creating a new Global Health R&D role citing RP research. Gates Foundation and Packard Foundation commissioning work suggests they find it useful.
• Evidence against: RP's 2019 impact survey found "few clear career changes or interventions started/stopped." Most impact claims are confidential or hedged. No publicly verifiable case of a major allocation shift attributable to RP research.
• Testable? Partially. If RP tracked and published the before/after allocation decisions of funders it works with, this would be testable. They plan to "present results in 2026" from a new evaluation system.
• If wrong: RP is a well-intentioned but ultimately low-impact research organization producing outputs that confirm what funders already believe.

Assumption 2: The moral weight of animals is not vastly lower than humans'.

• Evidence for: The Moral Weight Project's extensive methodology and engagement of 18 specialists. 80,000 Hours and other EA orgs treat the results as credible.
• Evidence against: The "too animal friendly" critique identifies four systematic biases that could inflate estimates by 5-500x. Neuron count (arguably the most objective proxy) is largely dismissed.
• Testable? Somewhat. The methodology is transparent and others can run it with different assumptions. A "neuron-weighted" version would produce very different results.
• If wrong: A significant chunk of RP's influence on EA resource allocation (toward animal welfare and away from other causes) may be based on inflated moral weights. This is RP's single highest-stakes intellectual claim.

Assumption 3: Cross-cause analysis is better than specialization.

• Evidence for: RP's unique position enables connections others miss (e.g., the DCM bridges animal welfare research and AI consciousness). The CURVE sequence challenged longtermist assumptions using tools from GHD and welfare economics.
• Evidence against: No external criticism of this approach exists (which could mean nobody disagrees or nobody has thought carefully about it). Cross-cause organizations risk being mediocre at everything rather than excellent at something.
• Testable? Compare RP's influence per dollar to specialized equivalents. Hard to do in practice.
• If wrong: RP would have more impact by focusing exclusively on animal welfare research (its strongest suit) or AI policy (its most timely work).

Assumption 4: Digital consciousness will become a policy-relevant issue.

• Evidence for: The DCM got a dedicated workshop at NYU's Center for Minds, Ethics & Policy. AI labs are beginning to consider these questions. If AI systems become arguably conscious, the policy implications are enormous.
• Evidence against: Current evidence is against LLM consciousness. Policy relevance depends on AI architecture evolving toward systems more plausibly conscious than current LLMs. Timeline is uncertain.
• If wrong: The DCM and AI Cognition Initiative would be interesting academic work with limited practical impact.

Strengths

Intellectual honesty. RP's willingness to publicly name its failures (PriorityWiki, the health economics series, Rethink Grants) and acknowledge its biggest early mistake (research without stakeholder engagement) is genuinely unusual in the EA ecosystem. This suggests an organization that updates on evidence rather than defending past decisions.

Incubator track record. Via Special Projects, RP helped launch Apollo Research, Epoch AI, Truthful AI, and other now-significant organizations. This is arguably as impactful as any of their research, and gets less attention.

Unique niche in animal welfare science. No comparable organization bridges rigorous welfare science, philosophical analysis of moral weight, and practical cost-effectiveness for the EA community. The Moral Weight Project, invertebrate sentience work, and now DCM represent genuinely novel intellectual contributions.

Financial resilience. Net assets of $14.8M (2023) provide substantial runway. Survived FTX collapse without major disruption. SP is self-sustaining. Revenue diversification is proceeding.

Talent pipeline. Staff who leave go to OP, Founders Pledge, CEA, 80K Hours, and elected government. This suggests RP develops people who are broadly capable rather than narrowly specialized.

Weaknesses and Risks

Extreme funder concentration. OP/CG at 70-80%+ of non-SP funding is the most obvious vulnerability. RP's research agenda is significantly shaped by what its dominant funder wants. A change in OP's priorities or leadership could devastate RP.

Impact attribution remains weak. After 8 years, RP still cannot point to a single publicly verifiable case where their research caused a major, specific allocation shift. Their 2026 planned evaluation system will be a critical test.

Spreading too thin. Six research areas (soon more with AI Strategy and AI Cognition), new initiatives (Research Subsidy Fund, Neglected Animals Talent Program), plus Special Projects and IAPS spin-off management -- all on a $7.5M core budget. The risk of doing many things adequately but nothing exceptionally is real.

Board opacity. 4 of 5 board members have no publicly discoverable backgrounds. For a $20M organization, this is a meaningful governance gap.

The consultancy trap. As GHD becomes more consulting-driven (5-10% independent research in 2025, goal of 20%), RP risks becoming a research-for-hire service that confirms what funders already think rather than challenging assumptions. The "research and hope" failure they identified was real, but the pendulum may have swung too far toward client-driven work.

Loss of Peter Wildeford. Peter's shift from co-CEO to IAPS focus, and IAPS's impending spin-off, means RP loses its most prominent AI policy voice and its most visible forecasting expert. Marcus's strengths are practical and operational rather than intellectual and public-facing.

Cross-References

Complementary to: GiveWell (RP does research GiveWell-adjacent funders use), Open Philanthropy (primary client/funder), Animal Charity Evaluators (RP provides moral weight framework ACE uses), 80,000 Hours (talent pipeline in both directions).

Competed with by: Global Priorities Institute (foundational cause prioritization, more academic), Charity Entrepreneurship (incubation, though different style), Happier Lives Institute (wellbeing measurement, narrower focus).

Incubated: Apollo Research, Epoch AI, Truthful AI, QURI, Existential Risk Alliance, Condor Camp, Global Challenges Project, Consultants for Impact, Vista Institute for AI Policy.

Gap-filling role: RP uniquely bridges animal welfare science and AI consciousness research via the DCM. No other organization connects welfare philosophy, empirical consciousness research, and AI policy in this way.

What Would Change This Assessment

• Positive update: If RP's 2026 evaluation system produces publicly verifiable evidence of research changing major allocations. If the DCM becomes a standard tool used by AI labs for consciousness assessment. If the moral weight methodology survives a serious adversarial replication.
• Negative update: If OP significantly reduces funding and RP cannot replace it. If the moral weight estimates are shown to be 100x+ inflated by researcher bias. If IAPS spin-off leaves RP without AI policy credibility. If key researchers depart citing intellectual constraints from the consultancy model.
• Biggest uncertainty: Whether RP's research actually changes behavior or merely provides sophisticated confirmation of what funders already intended to do.

Self-Critique

Limitations of this analysis:

• Most of RP's claimed highest-impact work (commissioned research for funders) is confidential and therefore unverifiable. This analysis necessarily over-weights the public evidence, which may be less representative of RP's actual influence.
• I could not access forum comments and discussion threads that might contain important criticisms or defenses not captured in top-level posts.
• The Moral Weight Project critique is the main external counterpoint. There may be stronger arguments for or against RP's methodology that I have not encountered.
• I may be underweighting RP's convening and coordination function (Animal Advocacy Strategy Forum, fish welfare roadmap), which is inherently hard to assess from documents but may be very impactful in practice.

What a thoughtful disagreer would say: "You're being too critical of OP dependency. OP funding RP to investigate questions OP cares about is not a conflict of interest -- it's how applied research works. Gates Foundation funds research it wants investigated too. The real question is whether RP's conclusions are rigorous, and the answer is clearly yes given the methodology's transparency and the caliber of external specialists engaged."

Weakest claim: That RP's research may merely confirm what funders already believe. This is speculative -- I have no evidence that RP has produced research contradicting funder expectations and been suppressed, nor evidence that they haven't. The absence of evidence is genuinely ambiguous here.

What information would most change my view: (1) Publicly verifiable cases of RP research causing a funder to significantly reallocate (not just optimize within existing strategy). (2) An adversarial replication of the Moral Weight Project with different researcher priors. (3) Board member backgrounds and conflict-of-interest disclosures.