Red-teaming contest: demographics and power structures in EA

TheOtherHannah31 Aug 2022 4:36 UTC
110 points
14 comments33 min readEA link

Summary

In this post we[1] look at two common high-level criticisms of the EA movement, and evaluate them against specific examples from the global health and development cause area.

These high-level criticisms are [Section 1]:

  1. EA’s narrow demographic is bad for the movement, and

  2. EA has power structures that mirror (neo)colonialism, paternalism and institutional racism.

We evaluate these through three case studies [described in Section 2], and we identify three issues that we believe are restricting the EA movement’s effectiveness and growth [Section 3]. These identified issues are:

3.1 Demographic homogeneity (specifically, low cultural/​geographic diversity) is limiting the EA community’s ability to design and deliver effective, innovative solutions. [confidence: relatively strong, size of impact: medium in non-longtermist cause areas]

3.2 Decision-makers in some EA orgs do not have enough information or robust processes for addressing their blindspots, and this decreases the accuracy of their models. [confidence: medium, size of impact: relatively small]

3.3 Power structures within EA organisations have parallels to colonialism and institutional racism, and this is damaging the reputation and credibility of the EA movement. [confidence: medium, size of impact: difficult to determine, and debatable]

We believe these issues are limiting EA’s ability to become a truly global movement that inspires innovative and effective solutions for the world’s most pressing problems [Section 4].

Finally, we make recommendations for the EA community and some EA orgs to adopt in the near-term future [Section 5]. Broadly, our recommendations seek to:

  • Increase the sharing of EA tools with people in different geographic areas, without prescribing the end goal for those we share them with.

  • Increase conversations within EA about colonialism and institutional racism, with the aim that community members can better identify and avoid biases and power structures that perpetuate them.

Epistemic status:

We have considered the high-level criticisms from a variety of angles, and this post includes only the issues we think are the most persuasive. None of us are experts in the fields investigated, so we anticipate that there will be some mistakes or blindspots of our own contained in this critique. If you think we’ve completely missed work that EA is doing in these areas, perhaps this signals that the EA culture we’ve been exposed to doesn’t have enough mention of these aspects of EA.

Section One: Introducing the original high-level criticisms

1.1 EA’s narrow demographic is bad for the movement

1.2 EA has power structures that mirror (neo)colonialism, paternalism and institutional racism.

We will give a quick overview of these common high-level criticisms and state what we hope to add to the discussion.

1.1 EA’s narrow demographic is bad for the movement

The demographic of people who respond to the EA survey has stayed relatively consistent over recent years. Respondents are over 70% each of White, male and young, and live in the same set of five high-income, western countries (the US, the UK, Germany, Australia, and Canada) that were most common in previous years[2].

People commonly comment on the importance of demographic diversity, and how to increase it. The Centre for Effective Altruism (CEA) has a statement on why diversity, equity and belonging are important in an EA context, and the steps they are taking to increase these. It’s really good and you should check it out. This discussion is missing:

  • concrete examples that demonstrate a theory of change from “increased diversity” to “better ability to improve the lives of others”, and

  • Systems-change (outside of HR) that the EA movement and EA orgs could adopt to increase diversity.

This post seeks to add to the discussion by exploring:

  • concrete examples of the EA community’s blindspots in relation to demographic factors,

  • how these blindspots influence our approach to doing good, and

  • possible consequences on our collective ability to do good (3A, 3B)

Our recommendations give examples of systems and processes that can be adopted to pull the biggest ‘levers’ we see in our theory of change. We’ve focused mainly on geographic diversity, as we think this has clear levers to pull and geographic blindspots have the clearest direct influence on our approach to doing good. This isn’t to say that other forms of demographic diversity aren’t important—they’re just out of scope!

1.2 EA has power structures that mirror (neo)colonialism, paternalism and institutional racism.

“Neo-colonialism[3], institutional racism, power-imbalance, something-something” criticisms of EA have been made by public intellectuals, tweeters, orgs within EA, people within EA, and non-EAs that I talk to about EA.

Despite this, there aren’t many posts on the EA forum about the relationship between EA, neo-colonialism, institutional racism, and power-sharing. When neocolonialism (1, 2, 3) and institutional racism (1, 2, 3) are discussed, they are looking outwards at other orgs, rather than inwards at EA orgs (exception here on power imbalance in international aid, here on insider-positioning in global health and development, and here on institutional structures in EA). All of the exceptions took intentional searching to find; these aren’t common discussions in our experience in EA.

Criticisms that “neocolonialism and institutional racism are present within EA and this is bad” often end with ¯\_(ツ)_/​¯ because they are:

  • very general (“power imbalance is bad”)

  • over-reliant on a moral stance (“agency above all else”) without clearly explaining effect on impact, or

  • too uninformed to be conclusive (“I don’t know enough examples to make a nuanced judgement on this, so I can’t accurately update based on examples brought up for or against this argument”).

Additionally, these criticisms can be claimed in a way that places the burden of proof on EA, rather than pointing out specific examples or problems that allow a nuanced discussion.

This post will explore:

  • What “neocolonialism”, “institutional racism”, and “power-imbalances” mean in relation to the international aid sector,

  • Whether these are present or absent within power structures of EA orgs,

  • How this impacts our approach to doing good, and

  • How this approach impacts the outcomes achieved by the EA community (sections 3.1 and 3.2) and the reputation and credibility of the movement (in section 3.3).

Section Two: Case studies

We use three case studies to explore these criticisms in the context of specific decisions or interventions. These case studies are illustrative rather than comprehensive. Decision-makers in all three case studies had access to additional information or priorities that we aren’t privy to; we understand that it’s easier to critique from the outside without the full picture.

2A. GiveWell and IDinsight’s beneficiary preference study

We use this case study to demonstrate that decision-makers in EA have undervalued evidence that would illuminate their blindspots and improve their decision-making from a utilitarian perspective. We think the consequence of this on GiveWell’s absolute impact is relatively small given GiveWell’s particular target audience. We also demonstrate that critiques of the international aid sector are relevant to EA organisations: GiveWell structures and processes are designed in ways that perpetuate power imbalances and limit beneficiary autonomy. This discussion (Section 3.3) is more of an ideological discussion based on common leftist critique, and we don’t directly evaluate the consequence on absolute impact.

In 2019 Givewell funded IDinsight to study the preferences of beneficiaries for Givewell’s top charities. The results would be used to inform their approach to moral weights. The following section of the IDinsight report explains ‘moral weights’:

International development leaders frequently make complex resource allocation decisions that require weighing trade-offs between different types of good outcomes. For example, given limited resources, which should be prioritised: a program that increases household income or one that saves lives?

When comparing diverse charities, GiveWell makes these decisions transparent by asking staff members to provide their ‘moral weights.’ These judgments are based on philosophical reasoning, intuition and data on beneficiary lives, and extrapolation of preferences from studies of less relevant populations. Prior to this study, there was a clear lack of data on how potential beneficiaries of such interventions trade-off between different outcomes. This study represents a step to fill this gap for strategic international development decision-making.

You can play around with moral weights here, and see how different weights would affect the top recommended charity. GiveWell’s moral weights determine how millions of dollars per year will be distributed via their Maximum Impact fund.

The IDinsight study measured preference of beneficiaries across four diverse regions in Ghana and Kenya. It had a number of limitations (including the intended use of the data being within the limitations of the methodology chosen). Therefore, the following results should be considered indicative rather than definitive. It found:

  • Beneficiaries place a higher value on averting a death than predicted by most extrapolations from studies in high income countries (a value 1.7 times higher than the GiveWell staff median at that time), and;

  • Beneficiaries consistently value the lives of individuals under 5 higher than individuals 5 and older, which is consistent with high-income country studies, but contrary to median GiveWell moral weights (a value 4.9 times higher than the GiveWell staff median at that time).

The study’s design places limitations on the usefulness of the results because it allows people to place infinite monetary value on a life, and therefore these numbers are likely to misrepresent beneficiary choices in real scenarios. Additionally, beneficiaries may have been giving answers that were more socially acceptable or biassed towards what they think the researcher wanted to hear. More discussion on the limitations can be found in the GiveWell’s 2020 update on moral weights.

Givewell’s page on the study results notes that despite the limitations, they “have provisionally updated [their] moral weights [...] driven by this study and other, independent arguments for putting more weight on health relative to income.” They said they would be excited to see further research that addresses these limitations (in Dec 2019) and as far as we can tell they haven’t yet funded this; we assume this is due to the Covid-19 pandemic and/​or lack of available funds, but it could also signify a lack of priority.

Givewell’s analysis finishes by saying that they “expect to move away from using the median staff member’s moral weights in [their] decision making, and to instead have a single set of moral weights.” Their 2020 update states that they asked donors for their preferences and aggregated this with the IDinsight survey results and GiveWell staff views.

2B. NZ community lockdown response, and Māori vaccination rates for Covid-19

We use this case study to demonstrate that adoption of an intervention can increase significantly when designed by community insiders. Although this case doesn’t involve EA orgs, we have a higher understanding of the cultural context so can give a more comprehensive opinion. We believe it demonstrates an aspect of intervention design that the EA community should consider more often; insider-designed, centrally-resourced. Section 2C describes an EA case study which mirrors this in a similar context (which we personally know less about).

In response to a global pandemic, New Zealand went into a highly restrictive nation-wide lockdown starting 26 March 2020. The government made funding available to marae (loosely: community hubs for Māori (indigenous NZ) tribes) to support their local communities in this initial 2020 lockdown. This was seen as a successful investment by both the communities and the government. This is summarised well in a Treasury 2020 discussion paper (p16):

Levels of trust in government are uneven in New Zealand, however, and are fragile among Māori and Pacific communities. Such communities may view contemporary issues and crises, including COVID-19, in the context of New Zealand’s colonial history, and be less trusting than others of official responses. The Waitangi Tribunal has found that the legacy of colonisation continues to affect the lived experiences of Māori today, causing low levels of trust in the Crown (The Waitangi Tribunal, 2019).

[...]

The strengths of [Māori tribes] and community organisations were highlighted in the immediate pandemic response and the ways in which communities, marae and [sub-tribes] supported the wellbeing of their [families]. Examples include the provision of food and hygiene packs and vouchers, the establishment of 0800 numbers for support and information, and the provision of social services.

For example, Kōkiri Marae called around the local community and found that many families were struggling to get enough food during the lockdown. They understood the nuanced barriers to accessing existing services e.g. being seen at a foodbank reduced your perceived mana (loosely: status, prestige—a very important Māori cultural value). Therefore, Kōkiri Marae designed a service to deliver food parcels to doors. They called all community members in their phone books to check in, learn about others in need, and they had doctors, nurses, and social workers available via phone (rather than online). Community members reported high levels of satisfaction at the Kōkiri Marae services, whilst food banks and other services at the time were unreliable and had multiple barriers to access such as transport, timing, and the cultural barrier described above.

In contrast, the vaccination rollout in 2021 was Government-led. In December that year, the Waitangi Tribunal (a permanent commission on inquiry into Crown (i.e. NZ government) actions relating to Māori) released an assessment of the Crown’s Māori vaccination strategy and rollout. They reported that “it was Cabinet, through its early poor decision-making on the age-based vaccine rollout, that made Māori less protected against COVID-19.” For example, in October 2021 only 49% of Māori had received two doses of the vaccine compared with 72% of the entire eligible population.

Two months earlier, the low rates of first-dose vaccination for Māori had actually led the Government to change its approach to a power-sharing model. The Waitangi Tribunal report (above) describes the success of this approach:

As resources and funding for Māori primary health and social service providers increased measurably from August 2021 onwards, [Māori] were able to achieve impressive results. Between 6 October and 9 December, Māori health providers ‘vaccinated over 152,000 Māori, a 54.7% increase, twice the national increase of 27.1%’. This indicates that if more Māori had been eligible earlier, and more funding and resource provided earlier, there would likely not have been the lag in vaccination rates that we see now. In fact, largely due to the efforts of Māori providers, Māori vaccination rates are now accelerating at such a pace that the 20 percent inequity is closing fast.

Some of the interventions that were designed include campervans that were converted into mobile vaccination clinics, “Super Saturday” mass vaccination events with a community vibe, and anti-misinformation campaigns through social media and face-to-face efforts.

We believe that Māori communities were successful in designing and leading the response because they had strong contextual knowledge: local connections, high trust from community members, and contextual understanding of community culture, processes and needs. This allowed them to design bespoke and high-impact initiatives with high adoption rates when backed by enough financial and other resourcing.

In the Tribunal report, the Government reported that it realised it would need to rely heavily on Māori health and social service providers. However, their decision-making did not reflect this for months; there were several attempts by Māori to reach out and design solutions earlier, but the Government didn’t engage with these. Finally, the report notes that “While these [interventions] may be novel and innovative from the Government’s perspective, they are ‘normal’ models for Māori.”

2C. Suvita

This case study is an example of a similar intervention to 2B, this time through an EA organisation. Similarly, it demonstrates that people with an insider-understanding of the beneficiary community can be better placed to design effective interventions than people outside of that community. This means EA’s current demographics could lead to delays in finding the most effective interventions.

Suvita is a product of the Charity Entrepreneurship incubation programme. They deliver a bespoke, highly effective intervention to increase child vaccinations in India. Suvita is based in India due to the founders’ professional and personal connections, as well as the scale of the problem; approximately half of the world’s undervaccinated children live in India. The supply-chain in India is more robust than most countries with low child vaccination rates; Suvita founder Varsha explains on the 80,000 Hours podcast:

“India has an extensive supply of vaccination and it has an excellent health system that’s rolled out to deliver these vaccines. We also know they have had successes in the past with vaccines such as polio, where there was a huge government and community and other supporter effort, where they managed to eradicate the polio from the whole country. So I think we are fairly confident, the challenge is very much on the demand side and there isn’t anything else that the government could be doing on the supply side to make this better.”

Suvita partners with high-leverage community members (“gossips”) to spread vaccination information to parents, alongside text message reminders. New Incentives gives cash payments for child vaccinations, and currently has an estimated cost-effectiveness of $5,000 per life saved. Suvita’s estimate was $1,700 to save a child’s life (based on outdated figures which need to be recalculated). Assuming this is in the right ballpark (i.e. somewhere below $8,000), this is likely to be comparable to New Incentives, and could develop into a highly effective, scalable NGO.

Varsha explains some advantages she had in designing this intervention based on her position as a relative “insider” to the community:

Varsha Venugopal: Yes, I have family in India and I was born and I grew up there.

Rob Wiblin: All right, yeah. Was that a substantial advantage as a co-founder? Do you think maybe… Would Fiona have found it more difficult without someone who was more familiar with operating in India?

Varsha Venugopal: Possibly, yeah. So India is of course a big country and my middle class experience may not be the same as people in other regions and different income levels. But having said that, because I grew up there, I’d like to think I can tap into my personal and professional networks. Maybe I also understand some of the local context and politics better, and I’d like to think I can also relate better to my team and the end users in different ways.

Varsha Venugopal: I’m also a mother, which maybe gives me a personal understanding of the challenges that parents face and thought processes they may use to make decisions. I think something I mentioned earlier, being aware of one’s strengths and limitations and actively seeking a co-founder to complement them could be valuable for any startup. So Fiona, she’s much younger and she’s closer to our team’s median age, I think, and she manages our team on the ground.

Varsha Venugopal: So I think the fact that we bring complementary skills and experience allows us to run the organisation in a way that we may not be able to do by ourselves.

Cost-effectiveness, in this case, arises from a design informed by

  • nuanced understanding of the community culture, processes and priorities

  • EA tools and methodologies, e.g. a randomised control trial during their research phase of the charity incubation programme.

Section 3: Identified issues

We will explain three issues we identified, drawing from the case studies in section 2.

3.1 Demographic homogeneity (specifically, low cultural/​geographic diversity) is limiting the EA community’s ability to design and deliver effective, innovative solutions. [confidence: relatively strong, impact: medium in non-longtermist cause areas]

3.2 Decision-makers in some EA orgs do not have enough information or robust processes for addressing their blindspots, and this decreases the accuracy of their models. [confidence: medium, impact: probably small]

3.3 Power structures within EA organisations have parallels to colonialism and institutional racism, and this is damaging the reputation and credibility of the EA movement. [confidence: medium, impact: difficult to determine, and debatable.]

3.1 Demographic homogeneity (specifically, low cultural/​geographic diversity) is limiting the EA community’s ability to design and deliver effective, innovative solutions.

There’s a lot of different ways to talk about demographics. For section 3.1 we are talking about a fuzzy ‘cultural/​geographic’ demographic that combines:

  • Where you’ve lived,

  • Which communities you’ve been a part of, and

  • Which worldviews and circumstances you can accurately relate to.

Our examples in this post are all from global health and development. However, there is a strong argument that this applies to animal welfare too.

We believe that low geographic diversity in EA and low outreach limits our ability to design and deliver effective, innovative solutions to the world’s most pressing problems. The high-level theory of change on this is roughly:

Highly effective interventions can be identified as having disproportionate impact per dollar when delivered successfully. In some circumstances, these highly effective intentions have low adoption rates by the community. In addition to the last-mile problem of physically delivery, low adoption could be because of:

  • high activation energy or ongoing energy required to receive/​install/​use the intervention

  • low trust of the organisation delivering the intervention

  • cultural misalignment

  • intervention is uncomfortable e.g. vaccinations hurt, pills taste bad

  • low knowledge about the intervention and its benefits

  • beneficiaries have other priorities.

In cases with low adoption rates, it’s probably easier for a person inside (or close to) the community to understand (1) why the intervention isn’t being adopted and (2) what would need to change to make the intervention more widely adopted. They also have higher leverage within the community themselves due to higher trust and connections. Therefore, in these cases, interventions designed by people with relatively high insider-positioning may increase adoption rates much more effectively than those with outsider-positioning.

Most people within EA are outsiders to beneficiary communities (of Global health and development initiatives, and animal welfare initiatives). If a wider cultural/​geographic spread of people were resourced with EA tools, connections, and funding, there would be higher ability to design and deliver innovative solutions which are cost-effective in these low-adoption cases.

In both vaccination case studies, we can see that vaccination rates increased when insiders to the community were resourced to design and deliver the intervention.

  • In 2B, Māori community leaders had unique knowledge of the barriers to existing services, and had high trust (where trust in the NZ government is notably low in the context of colonisation). They used a nuanced understanding of the levers that could be pulled to successfully increase adoption rates in both the initial lockdown and for vaccination rates.

  • In 2C, Varsha’s background gave insight into barriers that parents were facing in these contexts—while her background didn’t match entirely, having lived in India and being a parent gave her an understanding of barriers that parents were facing, and political levers. It was also informed by a randomised control trial in 2018 that suggested that aid was adopted more successfully in Indian villages when ‘gossips’ were given the information to spread, and the scale of the undervaccination problem in India.

Suvita is an example where geographic diversity within the EA movement has opened up new avenues for design and delivery of an effective intervention within a nuanced situation. It seems the success came from someone with *at least some* insider knowledge being resourced with EA tools, connections, and funding.

To be clear, for this post, we are not making the claim that:

  • All interventions impacting a community need to be designed in partnership

  • Only interventions that are designed by community members will be successful.

Rather, we are saying that: in circumstances where a highly effective intervention has low adoption rates in communities, we think greater insider knowledge is likely to increase adoption of the intervention. Insider perspectives could be included through:

  • People with insider knowledge identifying as an effective altruist through involvement in the EA community

  • People within the effective altruist community partnering with governments, communities or individuals to enable them through EA tools, connections, and funding.

Karen Levy talks about partnership with governments to design and deliver effective interventions on this episode of the 80,000 Hours podcast.

Our first conclusion on this issue is that EA could be more impactful through innovation that’s unlocked by greater cultural/​geographic diversity.

Our second conclusion is that we should discuss partnership approaches more in the EA community; not just with people working on these cause areas. While writing this post, we were surprised to find that EA does have examples of EA orgs partnering with insider communities in a way that addresses the issue we’ve identified. However, most discussions we’re exposed online, in books and in our local community on global health and development stopped at “GiveWell vs. GiveDirectly.” Without intentionally seeking additional examples, we thought that these two orgs were “the EA position” and felt that this position was ignoring wider nuances and opportunities. After seeking these examples, we are more able to participate in nuanced discussion about different approaches tried and endorsed by the EA community.

3.2 Decision-makers do not have enough information or robust processes for addressing their blindspots.

Given our relatively homogenous demographics as a movement, do we have good systems to notice and address the blindspots that arise?

For this question, we mainly look at GiveWell’s moral weights (Case study 2A). We argue that the IDinsight’s 2019 study highlighted that GiveWell’s ‘map’ of beneficiary preferences is not as accurate as it could be. As the study itself had major limitations, this suggests that a more robust study is needed to ensure GiveWell’s decisions are well-informed. Our main conclusion here is that GiveWell is aware of this blindspot and moving in the right direction, but doesn’t seem to be prioritising this as highly as we believe they should.

What is the blindspot?

GiveWell doesn’t have a good understanding of the actual preferences of the beneficiaries impacted by the interventions GiveWell evaluates.

Their moral weights were previously calculated through a staff average. This didn’t include any weighting on preferences of the beneficiaries themselves; preferences of higher-income countries were used as a proxy because no research was available on preferences of the most relevant countries.

IDinsight’s 2019 study suggests that beneficiaries place 1.7x the value on averting death compared to the GiveWell staff median at the time, and 4.9x the value on lives of individuals under five years old (in comparison to lives of individuals over five years old). There are limitations on the validity of these results due to study design, but it still indicates that there’s an unexpected difference between GiveWell’s ‘map’ and the ‘territory’ in this case.

Why this matters:

Givewell’s model makes recommendations that follow the staff members’ model of what should be most valuable to beneficiaries, rather than actual value to beneficiaries. From a consequentialist utilitarian perspective, if there is a large gap between the modelled values and the actual values, the intervention is falling short of the ideal outcome: to deliver interventions that serve beneficiaries’ actual needs, values and preferences.

When comparing, say, a deworming vs vitamin-A supplements, the value placed on saving the life of an under-5 year old compared to saving the life of an older person changes which charity to recommend: Vitamin-A supplements primarily target younger children, deworming primarily targets older children.

The study indicated that beneficiaries in low-income countries have different preferences than higher-income countries. GiveWell’s working model of beneficiary preferences (initially based on similar studies in wealthier countries) was therefore informed by GiveWell’s blindspots and inaccurate assumptions of beneficiaries’ preferred outcomes.

What system does GiveWell have in place to address this blindspot, and is it working?

In 2019, GiveWell ran a sensitivity analysis and concluded that changes to moral weights based on the study would not change their planned allocations in the near term[4]. However, their planned allocations only partially included the beneficiary preferences due to the study limitations. It seems to us that more robust data has a reasonable chance of having a non-negligible impact on how funds are allocated, given the 4.9x difference between beneficiary preferences and GiveWell’s prior weight on the value of an under-5 year old compared to an over-5 year old.

In 2020, they reduced the weight they placed on the GiveWell the staff median, weighting different sources as follows to calculate their overall moral weights:

  • 60% weight on donor responses

  • 10% weight on James Snowden’s 2018 weights (as a proxy for 2018 GiveWell staff)

  • 30% weight on YLLs (both as a commonly-used metric itself and as a proxy for the IDinsight survey)

They do have a system in place to address this blindspot. The system, unfortunately, doesn’t have enough information to accurately address the blindspot. To reduce the staff blindspot, they’ve included a high weighting on donors’ expressed moral weights… which presumably has similar, or even more, blindspots about beneficiary utility functions.

GiveWell said in 2019 that they would be excited to see more robust research in this space, but in the 2.5 years since this statement, they haven’t funded further research. In the above-linked documents, I haven’t been able to see any commitment or projections about future work on this other than “we would be excited to see further research that addresses these limitations.”

An uncharitable conclusion would be that GiveWell received the information in 2019, said “The study has big limitations, it doesn’t change much; this isn’t important to us but another study would be interesting if someone else wants to fund it.” We think it’s more likely they haven’t funded another study due to the covid-19 pandemic and/​or funding constraints. However, I raise this uncharitable conclusion to show why I think it’s important for GiveWell to be transparent about their intentions for funding further research.

Is this a priority for GiveWell? What’s their timeframe? What changes would they make if they had more robust data, and what would the limits of that be?

More transparency would allow the EA community and people outside the community to further discuss whether GiveWell’s systems for addressing this blindspot are robust. A public commitment to funding another study within a certain time frame, or given a certain level of validity, would make GiveWell and the EA community less open to critique under issue 3.3.

3.3 Power structures within EA organisations have parallels to colonialism and institutional racism, which damages the reputation and credibility of the EA movement.

  1. What are the concerns about colonialism and institutional racism in the international aid sector?

  2. How valid are these concerns in relation to EA?

  3. How does EA talk about or respond to these concerns?

We will discuss all three questions in this section. Although this section focuses solely on the risk to our reputation and credibility, we aren’t meaning to say that these issues don’t limit our actual impact. The sections above have talked more about tangible impacts that may result from the power structures discussed in this section.

3.3.1. What are the concerns about colonialism and institutional racism in the international aid sector?

Most ideas from this section are based around the UK’s International Development Committee (IDC)’s Racism in the Aid sector report, which highlights reports of institutional racism in the international aid sector. We’ll focus on three key take-aways that are relevant to EA:

  • Colonial legacy

  • Imbalanced power structures

  • Paternalism

This section doesn’t claim that people within EA haven’t noticed the skulls of colonialism and (more broadly) institutional racism. It does claim that EA lacks a culture of talking about how colonisation and institutional racism interacts with our actions; therefore, we aren’t able to ensure, as a community, that we aren’t perpetuating harms done by these power structures. This section does claim that the lack of discussion in the EA community comes across as negligent, and presents a risk to our reputation and credibility.

Colonial legacy

While the EA community hasn’t directly exploited countries in a neocolonialist fashion[3], most EA global health and development orgs serve people living in a cultural context of prior colonisation, and most aid organisations have power structures designed from colonialist assumptions.

Here’s their summary on the colonial legacy in the context of aid (pages 8-9), emphasis added for areas particularly relevant to EA organisations):

Many evidence submissions we received to this inquiry described the structure of the aid sector as a legacy of the colonial era, which continues to replicate imperial power imbalances. Dr Lata Narayanaswamy, Associate Professor in the Politics of Global Development at the University of Leeds explained how numerous colonial officers went on to shape the multilateral ‘Bretton Woods’ institutions including the World Bank and IMF as well as some UN agencies at the forefront of the aid sector. Best practice for systems and approaches continue to be directed by donors and NGOs from White majority, high income countries such as the UK.

We also received evidence that the UK’s involvement in the slave trade and wealth extraction from around the British Empire contributed significantly to the economic development of our nation, at the direct expense of others. In turn, this has contributed to many of the conditions that necessitate the aid sector to exist today. Furthermore, we heard how colonialism left a legacy of racial hierarchies, having used illegitimate beliefs about communities that are Black, Indigenous and People of Colour to justify slavery and forced displacement. This led to perceptions that these communities were incapable of self-governance, and is reflected in the discrimination they face today. The aid sector needs to have difficult conversations about how power imbalances, racial injustice and poverty came about and how it can help to address these underlying factors.

In New Zealand, colonisation has been a driving factor for the decrease or loss of Māori language, culture, land, knowledge, and restricted Māori ability to self-govern, among other things. In many (though not all) cases, missionaries or the state created the conditions for these losses under the belief that it would be good for Māori people. The common rebuttal that “colonisation was good on balance” doesn’t consider a counterfactual where Māori retained self-governance and gained access to Western goods and ideas through co-governance, trade and/​or support from overseas institutions to use evidence to inform decisions.

Colonial legacy influences collective assumptions about who has the default right to make decisions, and therefore impacts peoples’ decisions about how to structure power when designing new institutions.

Imbalance of power

The IDC identified another expression of racism in the aid sector as imbalances in power: organisations place high trust in decisions made by international aid organisations, and low trust or value in the voice of beneficiaries (page 12):

Across the global aid sector, racism manifests in decisions around whose expertise we value. Evidence to our inquiry suggested that institutions in high income countries like the UK assume they have the knowledge and best practice to assist people in low and middle-income countries. Due to a belief that these institutions represent the ‘gold standard’, local partners are often required to adapt to their way of working.

You might argue that multi-national organisations do have the wider context to be able to better perceive how to deliver the most good to beneficiaries, and are better placed to decide which interventions are most effective. However, I think this is misleading. Multi-national organisations uniquely have the efficiency of scale that allows them to purchase and distribute more resources per dollar. The power structures that people have designed into these organisations give donors and head-office workers the academic tools and information; this doesn’t have to be the case.

Paternalism

A third form of institutional racism often criticised in the international aid sector is paternalism, defined in Baker (2015) as the belief that recipients of aid are unable to develop without the assistance of white, Western providers. In discussing his results, Baker gives a short literature review of paternalism in relation to race of recipients (page 14, edited for brevity):

The prevailing mindset among donor-country politicians and aid bureaucrats is that “poor people should be enticed to do what we . . . think is good for them” (Banerjee & Duflo, 2011, 9). Similarly, Michael Barnett claims that “paternalism is the form of power most familiar to humanitarians” (Barnett 2011, 233). Vast amounts of resources are expended to have an enormous bureaucratic infrastructure [...] dictate how funds are channelled from developed-world taxpayers to developing-world recipients (Gibson et al. 2005).

Moreover, this may be linked to race. For example, developing countries with black majorities are less likely than those with nonblack majorities to receive their development assistance in the form of country programmable aid (CPA), which is “the portion of aid on which recipient countries have, or could have, a significant say” (Benn, Rogerson, and Steensen 2010, 1; see also Kharas 2008). From 2000 to 2011, countries with black majorities received 63 percent of their bilateral aid as CPA, whereas the rate among all other countries was 79 percent. In other words, black-majority countries are more likely to receive official development assistance in a form that gives them little control over how to use it[5].

By assuming that we know what’s best for people in the developing world, we continue the trend of colonialism; we view beneficiaries not as agents who have a role to play in the development of their own communities, but as helpless people who need to be given the best interventions by smarter people in other countries.

The summation of these above issues is that people are concerned when power is structured in ways that limit the agency of beneficiary communities or governments, or make decisions on their behalf.

Let’s take a look at how an EA approach can pattern match for the issues described above.

3.3.2. How valid are the concerns? GiveWell’s approach to moral weights

GiveWell intends to move towards a single set of averaged moral weights (henceforth ‘SSMW’) to inform their future recommendations. “We expect that choosing this set will involve determining what moral weights would be implied by a variety of approaches or worldviews and then taking a weighted average of those views.” This shows a blindspot in GiveWell’s understanding of cultural diversity. There is likely to be high global variance in which kinds of interventions are valued more. You may be able to find usable trends within communities, but personally think that a SSMW would serve most people poorly. GiveWell recognises that physical, social, cultural and spiritual contexts will give different communities a different set of needs, values and preferences. However, their response is to average out and standardise, rather than design a solution that accounts for the variation.

Furthermore, GiveWell’s belief that they would be the appropriate body to oversee the creation of a SSMW has echoes of colonial governments and missionaries asserting they know what is right for indigenous peoples around the globe.

In general, EA tends to:

  • treat the question of “who gets to decide good” as irrelevant,

  • place higher trust in our own conclusions than others, and

  • design systems that, by default, mean we need to make decisions on behalf of beneficiaries.

These are seen again in GiveWell’s response to the beneficiary preferences study, namely that they updated based on the limited data, and haven’t yet sought better data. Although GiveWell’s sensitivity analysis noted it wouldn’t change their near-future allocations, the study’s “major” limitations mean that a more reliable study would likely give a different weighting, and this may change their near-future allocations in unexpected ways. Rather than prioritising another, more robust study, GiveWell’s behaviour could imply that better data is closer to a ‘nice to have’ than ‘urgent priority’. As they don’t articulate concrete plans to fund future studies, it’s quite hard to understand where they sit on the spectrum. (Note: this is a particularly uncharitable reading, see final paragraphs of section 3.2 for more context).

Finally, when I look on Givewell’s Our Mistakes page, “Failing to consider beneficiary preferences in moral weights” … isn’t listed as a mistake. Given that they updated their moral weights due to the results of this study and other research at the time, failing to note this as a mistake seems like an oversight. It indicates to us that Givewell’s staff don’t consider this as an indicator of their potential blindspots as decision-makers, and suggests that GiveWell’s staff don’t have a culture of critical and transparent reflection on their position of power in the aid sector.

Defining ‘good’ or choosing which intervention is ‘best’ for beneficiaries doesn’t inherently sit best at the international org level. Here’s the actual types of leverage that are inherent to international orgs:

  • Immense purchasing power and increased cost-effectiveness through economies of scale.

  • Use of international metrics (e.g. QALYs) to consider counterfactuals and the marginal impact of additional funding towards an existing charity

  • Access to powerful tools of analysis and technical expertise for evaluation.

  • Access to decision-makers, networks and connections for other aid delivery, allowing last-mile delivery to be more efficient. (though, debatably, the local government may have access to equivalent leverage).

This doesn’t necessitate the international org to be the one defining ‘good’, although it does include evaluation using a definition of good (the QALY). A new org could be structured such that the above international leverage is used to support beneficiary communities to access data and methodologies that inform their decisions about what’s best for their own community.

Here are some alternative approaches to GiveWell’s proposal of creating a single set of averaged moral weights. For example:

  • The Life You Can Save places value on the expressed needs, values and preferences of beneficiary communities in their charity evaluation framework.

  • A meta-charity could look at coordinating the expression of beneficiary preferences. International aid organisations then become suppliers to customer (beneficiary) demands. Beneficiary communities are empowered to define what they want, and international aid organisations use their buying power and efficiencies of scale to serve these needs.

  • A cost-benefit analysis that is designed to inform beneficiary community choice, rather than donor choice. An example of this is the cost-utility analysis that Pharmac uses to decide which medicine New Zealand’s Government should fund. This involves consideration of international metrics, such as QALYs, alongside suitability for the New Zealand context.

  • Partnership programmes that supply EA methodologies, connections and resources to local leaders. For example a programme similar to Charity Entrepreneurship’s incubation programme, with a specific focus on supporting leaders in beneficiary communities to design effective and scalable interventions.

The above approaches flip the power dynamic and reduce the similarities between international aid and colonialism. They place a higher value on beneficiary agency, so that people with power and connections are encouraged to share information, resources, and decision-making power with beneficiaries.

A model that I would like to hear more about in EA discussions is the Copenhagen Consensus, which EA surprisingly doesn’t talk about much. They engage with citizens, NGOs, decision-makers, sector experts and businesses to identify priority areas and suggest cost-effective options. The decision-making power remains with the decision-makers within the country. It seems that Karen Levy’s work in Fit for Purpose might be similar to this, and I would be interested to hear more about these programmes and evaluations of their success.

3.3.3. How does EA talk about or respond to these concerns?

This concern is often quickly raised when I discuss EA with colleagues or friends; who defines ‘good’ and what gives them the licence to do so? The common response seems to be “People in EA think very hard about the decisions they make, including the question of what is ‘good’? We use clever maths and science to figure this out, which tackles one common problem in the aid sector that ‘good advertising gets the most funding’.”

This response doesn’t display a critical awareness of the broader (cultural) harms caused through colonisation, nor does it articulate how we avoid recreating these harms. Talking about this more, and specifically articulating how we are avoiding these harms, is something recommended in the comments of noticed the skulls, but I haven’t yet seen this articulated for global health and development, animal welfare, or longtermism.

I expect that some people in the community—especially those working in relevant orgs—have thought about this a lot, and have detailed nuanced views as to which types of aid can create this harm, and how they weigh the importance of, for example, cultural loss vs. the loss of a life. GiveWell even wrote about this in 2012; I wish this post was more visible or more widely discussed in EA circles. However, these discussions aren’t part of our community discourse, and members of the community (including me!) find it hard to convincingly endorse EA when these criticisms are raised. Lack of discussion on this threatens the credibility of EA as a movement who pride themselves on arriving at nuanced, considered opinions, and being transparent about what we value.

Following the Black Lives Matter movement and the IDC’s Racism in the Aid Sector report, and as EA becomes a larger community, I expect these criticisms to become increasingly frequent over time. Given the historical precedents of white, male, Western, academic and wealthy institutions making decisions ‘for the good of’ another party that ended up being resented by that party… I think the burden of proof does fall on us as a community to negate this criticism and show how we’re protecting against unintended harms to beneficiaries. We need to show that we are different and give others confidence that our decision-making isn’t falling ill to the same biases and blindspots of bad-actors in the past.

This post has shown one way that these criticisms apply to the EA movement, and that we don’t have any robust processes to identify, articulate, and avoid the criticised power structures. If the EA community is gaining a bad reputation as a relatively homogeneous (white, male, non-religious, wealthy, etc) population that seeks to hold power rather than sharing it, and has parallels to colonialism … it isn’t enough to say we’ve “noticed these skulls.” We need to have robust, transparent discussion that:

  • takes these concerns seriously,

  • shows we have taken time to understand them as a community, and

  • results in systems and processes moving forward.

A year from now, we should be able to point to a number of actions we’ve taken as a collective community and the systems in place that address these concerns and say, “Yes, we’ve noticed the skulls. And here’s why that won’t happen again.”

Section Four: Conclusions; becoming a truly global movement

EA’s lack of diversity and parallels to colonialism and institutional racism have meaningful repercussions for EA as a movement.

  • They limit our ability to design and deliver innovative, effective interventions to important, large-scale problems. [Issue 3.1]

  • They cause blindspots that miss valuable information to inform our approach. [Issue 3.2]

  • They influence our default designs about organisational structure. [Issue 3.3]

  • When we can’t engage with people who make these critiques, it damages our reputation and credibility. [Issue 3.3]

There’s a possible future of EA that is decidedly big-tent: People from all around the world have access to tools and methodologies that help them innovate and solve meaningful problems. We develop our goals and ambitions through connection and collaboration as a global community; we are driven by data to design and deliver impactful initiatives that decrease poverty, disease and death at an unprecedented rate; an international network of people passionately working to ensure that global powers choose collaboration over competition when technology or conflict threatens to limit or erase our collective future.

This possible future of EA might never happen.

Instead, we may find ourselves talking past or against others with similar aims, who see us as overly-confident, missing important contextual information, and misdirecting millions of dollars from an ivory tower.

A lack of geographic diversity is plausibly limiting our ability to improve other peoples’ lives in cases where high-impact interventions have low adoption rates. Addressing this lack of geographic diversity in EA should be a priority, as it will help other people solve their problems more effectively, and also increase the innovation that goes into problems the EA community has already identified as highly important, neglected, and tractable.

People who hear about EA for the first time often want to explore discussions on power and beneficiary agency; intelligent, motivated, caring people who have nuanced models of the world. If EA can’t discuss these criticisms with examples and humility, this will turn some of these people away. This has, at least, been my personal experience when discussing EA with colleagues and friends.

Lack of conversation on these topics also erodes our credibility in the public and political eye; it doesn’t look good when we claim to ‘do good better’ but can’t articulate how our actions meet others’ critical frameworks, (or why we have chosen not to).

One strength of the EA movement is that community members are open to new ideas, reflective, and are responsive to criticisms. However, I see a lack of a coordinated EA ‘systems response’ to the issues above. This means that people external to the movement (and indeed, people like me) will continue to be dissatisfied that (a) the skulls have been noticed, and (b) the movement has systems in place to avoid more skulls.

A number of 80,000 Hours’ high priority cause areas will require EA to be a truly global movement. For example:

  • designing solutions that tackle global health issues and poverty will benefit from talented people who have access to EA methodologies, in combination with their insider knowledge of beneficiary communities;

  • EA-origin think tanks need positive reputation and credibility for policy-makers to adopt their recommendations in a range of cause areas;

  • reducing great power conflict over time will require support from citizens and political actors in a wide range of countries;

  • policy to address existential risk from AGI may require ideas to have support from citizens and bureaucrats in many countries.

To achieve this, we believe the EA movement needs to become more able to talk about power imbalances, our own blindspots, and how our tools can enable others to achieve their needs.

Section Five: Recommendations

This section lists our recommendations to address these issues and help EA become a truly global movement. This list is most applicable to global health and development, and animal welfare cause areas.

Broadly, our recommendations seek to:

  • Increase the sharing of EA tools with people in different geographic areas, without prescribing the end-goal for those we share them with.

  • Increase conversations within EA about colonisation, institutional racism, and our own blindspots, with the aim that community members can better identify and avoid power structures that perpetuate them.

Recommendations for GiveWell:

  1. Commit publicly to a timeline for increasing your understanding of beneficiary preferences. This will give transparency on the priority that you place on better understanding beneficiary preferences, and any reasons why this has been deprioritised or delayed. It will also stimulate discussion from the EA community about the value and limitations on beneficiary preferences, which could result in new ideas about how to measure, and how to apply knowledge about beneficiary preferences. We also, obviously, recommend that this is treated as a priority.

  2. Add a discussion on beneficiary preferences to your “Our Mistakes” page. Noting any assumptions that were contradicted by the IDinsight study will contribute to transparency about your process, your prior blindspots and support visitors to the website to understand donor preferences better.

  3. Reconsider the purpose and use-case of a single set of moral weights. Consult widely before proceeding—are there preferences, needs or values that would reasonably differ between different beneficiary communities? Are there any strategic opportunities to build these into the model in ways that could support NGOs to deliver more targeted aid in the future? Given the cost and work required to set up a model like this, are there any opportunities for collaboration? Are there opportunities to use your position of power in the international aid sector to incentivise any incidental benefits through this work?

  4. Research options for adding ‘power-sharing’ or ‘community needs’ metrics to your charity evaluations. GiveWell has a lot of power in the international aid sector due to its position as a top charity evaluator; you set a metric of ‘best practice’ that some charities aim towards. Therefore there is a large potential for spurring wider change in the international aid sector through a change in GiveWell’s evaluations. We are recommending further research because we expect that integrating power-sharing dynamics would have a definite cost to impact in the short term, but could have a large benefit in the long run. We also expect that there’s a calculation involved about the current GiveWell target audience and whether a change would increase or decrease the funds raised and the absolute ‘good’ achieved. Therefore, we expect this would require a lot of additional research in order to create a case for adding either of these metrics.

  5. Promote The Life You Can Save at opportunistic moments on your website. We expect that GiveWell and The Life You Can Save have different target audiences. Identify parts of your website that may deter donors who are critical of the EA movement for reasons identified in section 3.3, and link to The Life You Can Save. E.g. here and here. In turn, perhaps The Life You Can Save could mention GiveWell as another charity evaluator which doesn’t include the additional criteria described here.

Recommendations for funders or grant-making organisations in EA:

  1. Increase funding towards initiatives like Charity Entrepreneurship’s incubation programme that empower people through EA methodologies; prioritise geographic spread of recipients. I can’t see much on their website about diversity, and whether they prioritise entrepreneurs in this fashion already. Ensuring a programme like this has greater geographic diversity in participants is likely to encourage innovation, reduce the EA community’s blindspots and biases, while also incubating a range of highly effective charities in different worldwide communities.

  2. Centre for Effective Altruism should increase the number of targeted grants towards areas impacted by scalable problems. For example, funding or support could be allocated in ways that encourage new organisations to set up outside of key EA hubs to increase geographical diversity. They have already taken some steps towards this with EAGx Hong Kong and EAGx Nairobi. We suggest the funding should prioritise areas that have demographic similarities to locations with the worlds most important, tractable and neglected problems. Keeping the context of colonisation in mind, there should be some thought to make this funding ‘enabling’ rather than ‘prescriptive.’ This funding has a different purpose, and should be considered separately, to GiveWell’s charity recommendations (which are targeted towards an audience that seeks specific, known, relatively immediate impact).

  3. Organisations like OpenPhilanthropy and FTX’s Future Fund should decidedly increase the number of regranters in minority–within–EA communities. FTX seeks regranters who can (among other things) leverage their diverse networks, bring in new people, and make use of local knowledge. It’s unclear how much they prioritise for people with geographic diversity as discussed in this post. I am less familiar with OpenPhilanthropy’s model, and the applicability of this recommendation to their regranter challenge.

  4. Support research on alternative models for effective aid distribution that have emphasis on empowering beneficiary communities or governments. If interest in power-sharing in international aid increases in the coming years, an EA-designed power-sharing model could have a founder effect on the international aid sector in a way that results in long-term increased use of EA tools and methodologies to evaluate impact.

  5. Consider setting up networks for local EA group leaders or group ambassadors to connect with others on the issues described in this post. These could be focused around increasing knowledge around these issues to share it further with their local group, or working groups that focus on designing solutions for local or wider implementation within the EA movement.

Recommendations for members of the EA community, and particularly group leaders:

  1. Promote The Life You Can Save alongside GiveWell; it appears to be referenced less frequently than GiveWell within EA circles and offers an equally clear introduction to effective altruism’s tools, priorities and philosophies. This may appeal to a different audience, and therefore higher emphasis may support growth of EA through giving people different options for charity evaluators to mention when they talk about EA to non-EAs. And, of course, it may increase the absolute amount of ‘good’ achieved through inspiring a larger number of people to donate to highly effective charities.

  2. Promote discussions on colonisation, power-sharing and paternalism. For example, you could run a local EA group night that looks more in depth at any of these criticisms about the international aid sector, and evaluate their relevance for animal welfare, community building, various existential risks, or global power conflicts. In particular, consider which perspectives are invited to the table to design interventions or decide what’s ‘good’.

  3. Do a stocktake of the EA orgs that you and your group members know about, and look for EA-aligned orgs you didn’t know about. For example, do members of your local EA meetup know a range of different EA orgs in global health and development work, aside from GiveWell and GiveDirectly? Does your tl;dr of EA introduce charity evaluators and EA-founded orgs with different approaches to global health and development? Have you discussed or investigated non-EA orgs that seem aligned with EA ideas, such as the Copenhagen Consensus Centre?

We found the exercise of writing this post to be both interesting and rewarding, and look forward to further discussion in the comment. As always, we reserve the right to change our minds on anything (and anticipate that comments will change our minds in some ways). I personally started out with a relatively uncharitable “red-teaming mindset” and it was nice to see end up believing that organisations like GiveWell have taken meaningful steps to address these concerns, and to learn about more EA-aligned orgs that are doing work in space that addresses these concerns.

And finally, thanks!

  • Thanks to GiveWell, for having so much of their process available online which allowed us to find enough information to use as a case study in this post. We really appreciate and admire the work that the GiveWell team has done to promote effective charity, and this critique is more of a ‘nit pick for further tweaks’ rather than trying to express fundamental disagreement with their approach.

  • Thanks to the three others who worked on this critique with me in detail, to tease out the meaningful threads from the less persuasive threads.

  • Thanks to the Effective Altruism Wellington community for reading and providing feedback on a draft version of this critique.

  • Thanks to all those I’ve talked to in my research for this, including those who helped to shape the research such as Alex, Morena and EA Wellington peers.

  • Thanks to those who comment below!

  1. ^

    We are four members of EA Wellington (New Zealand) who have all been a part of the EA community for at least two years. Broadly, working professionals with a range of interests including education, biorisk and community building. Most of the post is written by me (TheOtherHannah), with research input and edits from three others. Where this matters, I’ve distinguished opinions through the use of “I” and “we” throughout the post.

  2. ^

    Such high-level analysis probably obscures lots of nuanced detail.

  3. ^

    Neocolonialism =the use of economic, political, cultural, or other pressures to control or influence other countries, especially former dependencies. [Oxford Dictionaries]

  4. ^

    I note that their sensitivity analysis didn’t compare benefits other than “increase consumption” vs. “save a life”. It’s plausible that their planned allocations could change if evidence was stronger for indirect/​infrastructural impacts. The flow-on effects of indirect/​infrastructural impacts are widely debated in the international aid sector, so we’ve considered this beyond the scope of this post. For more on infrastructural change, see: Page and Pande (2018).

  5. ^

    I am not an expert in international development, and it’s unclear to me whether the claim above is robust – there may be a mediating factor to the allocation of CPA.

TheOtherHannah31 Aug 2022 4:36 UTC
110 points
14 comments33 min readEA link