In diversity lies epistemic strength

by FJehn7 min read6th Feb 202126 comments

11

Diversity and inclusion
Frontpage

Summary

Diversity is important. Not only inherently, but instrumentally as well. Based on ideas from the book “Why Trust Science?” by science historian Naomi Oreskes this post argues for increased diversity as a chance to produce more reliable knowledge. The main argument is that a more diverse community has more diverse perspectives on any given topic. This leads to more assumptions being challenged, which in turn increases the likelihood of being closer to the truth. 

Edit (12/02/21): There was a lot of discussion surrounding this post. I tried to clarify my positions in this comment

Varieties of Empiricism

How are we able to make progress in the realm of ideas and science? Think for example of the idea of continental drift. Right now we are pretty sure that this theory is correct, but how did we end up with this conclusion? The most straightforward way to tackle this question is to simply look for evidence for your hypothesis. For example, you can cut out the shapes of all continents and puzzle them together to form Pangea. While this gives the hints that your theory might be correct, falsification of your hypothesis might be right around the corner. Maybe we might find some evidence that the Atlantic Ocean does not experience seafloor spreading, which would be a strong argument against continental drift. 

So, is falsification the way to find truth? Not really, as falsification only really tries to tell you what is not true and also runs into the problem of underdetermination. This means that you never test only a single hypothesis, but also all other theories your hypothesis stands on as well. To stay with our example: you can measure if the Atlantic Ocean experiences seafloor spreading with radio telescopes. But, when you are using those radio telescopes you also assume that they are working correctly and that your understanding of electromagnetism is correct. Therefore, if our experiments show that there is no seafloor spreading, this might mean that continental drift is wrong, but it might also be just the case that our radio telescope is broken or electromagnetism does not work as we think in this case. There is no way to tell for sure, as we can never rule out all the possible ways our experiment might have gone wrong or what part of the natural world we did not understand correctly. 

Science as a Team Effort

We see that it is not really possible to prove nor disprove a theory using a single “scientific method”. But maybe we are looking at the wrong part of science to understand how we make progress? The theories we looked at so far (Positivism, Falsification, Underdetermination) are focused on finding a single scientific method. However, everyone that ever engaged with science knows that there are a plethora of different methods and that the key part of science is it being a team effort. Therefore, our focus should be less on individual scientists and more on the scientific community as a whole. Or to be more specific, we should focus on the thought collectives in different fields and how they work. If we do this, we can see that science is quite different from field to field and the tools and techniques used differ widely. 

We can also see that most scientists are not really trying to prove or disprove major theories, but are more similar to puzzle solvers who try to understand a tiny part of our world better. Only if those puzzles cannot be solved in the current paradigm of thinking, major scientific shifts occur. However, during those paradigm shifts not only theories change, but also values and beliefs, which makes it hard to compare paradigms to each other. Nevertheless, it seems that the paradigm shifts of the past somehow moved us in the direction of scientific progress. Still, this leaves open the question of how we can trust science and its findings if we have neither a single scientific method nor a quantifiable way to compare our current understanding of the world to earlier paradigms? 

Feminism as a way forward

The solution to this problem might come from an unexpected direction: Feminism. Feminist philosophers like Sandra Harding or Helen Longino have been asking similar questions for decades: how can we trust science if so many of its theories are based on prejudice? How can we trust science if so many of its scientists are biased, especially in regard to gender and race? For example, the underrepresentation of women in scientific studies. The answer here is that objectivity is not something that a single person has, but that objectivity is a social achievement of a diverse community. This is based on the idea that the assumptions and perspectives we have are strongly influenced by the life that we have lived. For example, a person growing up in a slum, always struggling for money, will have different beliefs about how the world works than a person that lives a life of luxury. But many of the assumptions we have remain hidden even to ourselves if they are not challenged. 

I would guess that almost everyone who reads this text has been in a situation where you talked with another person about a topic and you made a statement that was completely uncontroversial to you and true beyond doubt. However, to your surprise, the other person disagreed and stated that the opposite of your beliefs is true. It doesn’t really matter here which side of this argument was right or what the topic was, but the point is that you would have never challenged these assumptions on your own because you were so certain you never even realized that your idea was carried by underlying assumptions. You are not alone with this, this is just how humans and thus scientists as well, operate (if you disagree with me on this you have kind of proven the point). 

Objectivity as it is understood here is a continuum between being more or less objective. Objectivity increases with the diversity of perspectives, as more perspectives in a given discussion lead to more assumptions being challenged and, thus, to an answer that is more likely to be true. As we cannot measure the diversity of perspectives of a person directly, our best proxy for it is demographic diversity, as our life shapes our assumptions. Simply considering the perspectives that you heard from other people is not enough. For example, I can imagine that growing up extremely poor is a harsh experience, but I will never be able to wrap my head around how it really feels and how it would influence my thinking if it had happened to me. This is also true for my own life as it is. While I can describe what experiences I had so far in my life, I cannot explicitly state how they have influenced my thinking and shifted my perspectives, as those changes occurred mostly subconsciously.

It is difficult to give concrete evidence for the positive effect of diversity in perspectives in science, as we have no objective measure for “good science”. However, there are large scientific organisations like the IPCC that have acknowledged that they need scientists with a diverse set of backgrounds to create reliable science that does not leave out the perspectives of often underrepresented groups. In addition, there is lots of research about gender diversity and team performance in companies and much of it points into the direction that more diverse companies and teams are also more profitable, creative and productive (For example, take a look at this meta-analysis, but beware the man of one study and rather take a look at a random sample on Google Scholar for the search “gender diversity and team performance”). As we learned earlier in this text, this is not conclusive evidence (it never is), but we should not discard it, especially as we have a mechanism, in the form of diversity of perspectives, that explains the findings. 

Conclusion

The ideas presented here are certainly important for science, but they are also applicable to every other group that wants to make sure they are closer to the truth. This highlights the importance of voices that raise awareness on the issue of diversity in the EA community and propose efforts to make our movement more welcoming. The closer we come to the ideal of having diversity of perspectives, the more certain we can be that the consensus we reach is a good map of reality.

 

 

Thanks to Florian Zeidler, Jan Wittenbecher, Birte Spekker, Jonathan Michel, Ekaterina Ilin, Magdalena Wache and Luise Wolf for helping me improve this post with valuable feedback :)

If you want to look deeper into topics mentioned in this post, the main things to look for are (in the order they appear in the post): Positivism (Auguste Comte, Vienna Circle), Falsification (Karl Popper), Duhem-Quine Thesis/Underdetermination (Willard Van Orman Quine), Thought Collectives (Ludwig Fleck), Scientific Revolutions/Paradigm Change (Thomas Kuhn), Feminist Philosophy of Science (Sandra Harding, Helen Longino). Or simply take a look at “Why Trust Science?”, the main argument is laid out in the first ~ 150 pages. The rest is interesting case studies to prove her points and some remarks made by other scientists about how to increase trust in science.

11

25 comments, sorted by Highlighting new comments since Today at 9:05 PM
New Comment

As we cannot measure the diversity of perspectives of a person directly, our best proxy for it is demographic diversity, as our life shapes our assumptions.

This seems like quite a key premise to your argument, but you don't seem to spend much time arguing for its plausibility; indeed, it seems quite likely false. Could we not simply ask someone their views on a variety of subjects? If we know what their views are on Sports, Reincarnation, the Holy Roman Empire, Two Dogmas of Empiricism, Abortion, Pokemon, the Axiom of Choice, MIRI, and Socks with Sandles, I feel like we could have a pretty good sense of whether they add a new perspective. 

Similarly with education and career; I would in general expect more perspective diversity from a group consisting of an economist, a biologist, a nurse, a poet, a cop and a prostitute, even if they were all the same race, age and sex, than I would from a demographically diverse group of harvard-educated lawyers.

Yeah, just seems pretty obvious that if we care about intellectual and cognitive diversity, we can just measure that directly, by getting people with different educational and professional backgrounds. 

It seems that demographic diversity is one of the worst proxies of cognitive and intellectual diversity that I can think of, with it's primary benefit just being that it is really easy to enforce (i.e. it's very easy to tell whether a group is demographically diverse, whereas it might take a conversation with a group of people to figure out the diversity of their cognitive styles and philosophical assumptions).

You seem to assume that diversity of perspectives is easy to measure, because you only link it to the professional background of a person. However, I would argue that while profession is important, so is how I grew up and what experiences I had in my life due to sex, gender, race and other markers. Those things you cannot easily measure directly, but they improve discussions, as they lead to more assumptions being challenged. 

Sure, but in the above post you claim that demographic diversity is the best way to measure diversity of perspectives, which is a much stronger claim. I am not saying demographic diversity is completely irrelevant, I am just saying that it seems far from the best measure of cognitive diversity that we have.

As an addendum: First, suppose you compare a group of random people from the same demographic to a random groups of people from different demographics. Next suppose you compare a group of random lawyers to a group of random laywers of different demogaphics. I would suggest that in the second case the increase in diversity from adding demographic diversity would be significantly reduced as the bar to becoming a lawyer would filter out a lot of diversity of experiences from the first case. For example, a greater proportion of African Americans experience poverty than the general population, but the difference among those who become laywers would be much less.

Simply asking someone about their beliefs works if you have something conrete to ask for and know that kind of perspective you want to include. However, how would you know which questions to ask for? Aren't the questions you are asking not based on your own perspectives? What this post aims for is highlighting the importance of perspectives you cannot easily predict. For example, if you would you are doing a Hamming Circle you might have a hunch beforehand which people you would like to include, but during the circle the best feedback and help comes from a person and perspective you never even would have considered to be important. 

And to your second point: Why not both? My post aims to highlight the importance of diverse perspectives. Therefore, I would assume that I would get the most valuable consensus from a group consisting of economist, a biologist, a nurse, a poet, a cop and a prostitute, which are also diverse on race, age and sex. 

Of course you can ask people about their views and then use that to establish the intellectual diversity you want, and that can be a helpful tool. The problem, however, is that whoever creates the questionaire is likely biased towards their own beliefs and they might not even know what to ask to surface a specific subject that is mostly relevant for people of a certain demographic background. They may do this on purpose or subconsciously, even in good faith.

TL;DR Underrepresented groups are underrepresented because they self-select into EA circles at lower rates due to divergent values. Concerted initiatives to increase diversity are strongly biased towards the values of the people running the initiative. Communities are exclusionary by necessity.

I hope as one of the rare developing world EAs, my perspective is useful here. This post is a nice thought, but like many nice thoughts relating to diversity I believe it collapses on deeper consideration of how it would be implemented. There is a reason EA mainly consists of WEIRD (Western, Educated, Industrialized, Rich, Democratic) people, and underrepresented groups are not underrepresented due to some conspiracy to exclude them. It's because EA is based on a set of assumptions derived from WEIRD patterns of thought that are not common outside certain groups. For example, statements such as "All people's lives should be valued equally regardless of their age, gender, sexuality, religion etc." or "Empirical reasoning is the best path towards truth" would receive very different levels of agreement at Germany  universities and Bangladeshi villages. Be careful what you wish for! Those who claim to seek diversity are also the most likely to be horrified when they find it.  As a result most diversity initiatives in WEIRD circles tend to only select members of underrepresented groups from an elite, liberally educated minority within those groups who are rarely representative of the values of the groups as a whole, and much more representative of the values of the diversity seekers. So I more or less agree with other commenters on this post - demographic diversity is a fairly poor proxy for intellectual diversity, especially  when it is brought in through a concerted effort to increase diversity, rather than in an organic manner that reflects the changing norms of different groups and the communities they are joining. 

Communities are exclusionary by necessity : the questions you ask to determine entry are biased by the perspectives of existing members because they have to be! There is no reason for the community to exist if it is perfectly representative of all opinions in society at large - to paraphrase Syndrome, if you have everyone's values, you have no values. Would I be happy to meet more EAs who shared my background? Of course! Would I be happy if this happened at the cost of losing the values that make EA such a powerful lens with which to view the world? Hell no! Diversity is good at the margin, but it is not an intrinsic good. There is a balancing point that is best for the overall health of the community - it's important to try to rigorously determine what that is before making calls for increasing diversity.

One concrete way to do that would be to carry out two very large surveys - one of EAs, and one of randomly sampled people. They would measure agreement with certain values that are generally considered axiomatic to the EA way of thought. We could then statistically determine whether certain values cluster with certain identities, and the degree to which those values/identities are represented within EA. Would be a lot of work and money, but I believe it's the only rigorous and safe way to run a diversity initiative!

Aditya, thank you for your perspective! I think you have raised some excellent points, especially that movements are by definition exclusive as they would just be the "mainstream" otherwise, and that EA values align with a WEIRD background. I also agree that many people think of diversity as a box to tick and they don't actually want to engage with a set of truly diverse people. And honestly, I think everyone (including myself) should meditate on their motivations for increasing diversity (e.g. wanting more women in EA to extend one's dating pool might be a natural instinct but it's the worst possible driver for a gender diversity initiative).

On the other hand, over 50% of WEIRD people are women (academic participation and success of women now outnumbers men), probably 10% are LGBTQ, a significant number are People of Color and an also significant number vote for conservative parties. If these people don't show up in the EA community, I think it's valid to ask ourselves "Why?". Do we drive them away with sexist and racist behavior? If they are not privileged enough to show up, should we think of raising  their privileges as an important cause area? If they don't share our values, can we be confident that we have the right values or should we open ourselves up to their perspective and reevaluate our values afterwards? It doesn't mean we have to accept diluting our values, but it would be irrational not to consider the possibility.

I like your idea of surveying people to determine the main differences in values between EA-aligned people and the rest of societies. My only concern, which I've voiced in other comments under this post, is that such a survey might already be biased. Which means, we'd already need a somewhat diverse (both intellectual and demographic) panel to get started.

Hi Lukas, thanks for your kind words! I agree with you on most points - my analysis was mostly regarding ethnic identities rather than those relating to gender, sexuality or political orientation, since the former is what I have experience with. I do of course think that EA would benefit from more equal representation of these groups, and my intuition is that it's more tractable as well (relative to religious or ethnic minorities). I'm not entirely sure why women are underrepresented in EA or even if that is indeed the case. If it is, I'd imagine it's related to the same reason they're underrepresented in fields like computer engineering, which EA draws much of its membership from - some complex combination of innate differences, social conditioning and discrimination. I'm not sure how you would fix this, but what I'm trying to do is cautioning against shortcuts - the negative effects of deliberately  trying to increase diversity may outweigh the benefits, since any selection mechanism would likely introduce more bias than already exists in the community at large. Decentralized, organic change seems to be the way to go, but that takes time. Curious to hear OP (or other feminist identifying people's) take on this.

To address your point on opening up to different perspectives, while I think EA's openness to change is one of its greatest strengths, there is such a thing as being too open to different perspectives; as you've recognized, that would eventually just make us the mainstream. You do need to have some degree of confidence in your values. Along the same lines, the survey will of course be biased, because it's supposed to be! It would not be an effective selection mechanism if it did not already represent the existing values of the community, whether those values originate from the current demographic makeup of EA or are unrelated to it. Creating social norms that are truly unbiased seems like an infinite regress problem, so my preferred approach would be evolutionary - let different local EA chapters adopt different social norms, then promote the norms that are most successful in achieving the broader aims of the EA community.

FWIW, I don’t think your argument goes through for ethnic diversity either; EA is much whiter than its WEIRD base. I agree aiming to match the ethnic diversity of the world would be a mistake.

(Disclaimer: Not white)

Are you saying ethnic minorities in the West are less likely to be WEIRD and hence underrepresented in EA, or that ethnic minorities who are WEIRD are underrepresented in EA? The former wouldn't surprise me at all, given the significant disparities in income and educational opportunity between ethnic minorities in the West. The latter would surprise me, but I'm not sure how you would go about proving it, since it would require you to already have an estimate of demographics of true WEIRDs, and I'm not sure how you'd go about collecting that. Unless the assumption that any educated person from a Western Developed country is WEIRD, which I would disagree with.

I'm claiming the latter, yes. I do agree it's hard to prove, but I place high subjective credence (~88%) on it. Put simply, if I can directly observe factors that would tend to lower the representation of WEIRD ethnic minorities, I don't necessarily need to have an estmate of the percentages of WEIRD people who are ethnic minorities, or even of the percentage of people in EA who are from ethnic minorities. I only need to think that the factors are meaningful enough to lead to meaningful differences in representation, and not being offset by comparably-meaningful factors in the other direction. Some of these factors are innocuous, some less so. 

But if you're interested in public attempts to take the direct comparison route, which I fully acknowledge would be stronger evidence if done well, you might find this post of relevance. (Note I'm not necessarily advocating for the concrete suggestions in the post, mostly linking for the counts at the start.)

I agree with something in this direction, though not with everything as stated.
Some ways I see research affected by demographics:

  • Research of all kinds has been shaped by the viewpoints of people with more status, money, etc.
  • Sometimes this leads to serious slants in our understanding of the world - for example, the fact that so much psychology research has been done on WEIRD undergraduates means that our understanding of human psychology is badly skewed. Medical research that’s primarily carried out on one demographic group may not be generalizable to other populations, meaning worse health outcomes for less-studied groups.
  • In some other cases, findings are less susceptible to bias coming from the demographics of the researchers or research subjects. I can imagine how some aspects of climate science are swayed by demographics: do we look at how people of different geographies, ages, and genders are affected by food and water shortages, for example? But it's hard for me to see how some other aspects, like how polar ice melts, are all that susceptible to demographic bias.

Aside from what or whom is being studied, there is also the question of who is taken seriously in the research community itself.

  • When there are unnecessary slants in what’s seen as professional or credible, we systematically devalue the contributions of groups who don’t fit that mold. Example: someone with an accent perceived as lower-status may be taken less seriously. Or a workplace may treat straight hairstyles as professional but textured hair as less so.
  • Structural problems. Examples: a lack of all-gender or single-person bathrooms in a workplace or conference venue disproportionately affects trans people who have reason to fear harassment in bathrooms. A bad parental leave policy disproportionately affects staff who give birth.
  • Reducing people to their demographic, even if it feels positive. Example: after my first child was born, I went to an EA talk with my baby. Two male friends of mine were there. One talked excitedly in front of the group about the baby and my new status as a parent. The other didn’t mention the baby and only talked to me about EA-related topics, including asking me about an area and pointing out that I knew more than he did about this topic. (He did come over at the end of the gathering to visit with the baby.) While both friends were completely well-intentioned, the first made me fear “Oh no, am I just going to be a mommy now? Is that how they’ll see me?” The second was intentionally trying to avoid that, and he made me feel I was still valued for intellectual contributions.

When you have slants like these that affect who gets published, who gets funded, who gets hired, who can focus better on their work, etc, you lose good contributions. And that’s a loss to the field as well as to those individuals. (And all of this is aside from any knowing/intentional discrimination.)

I think these are problems we should be on guard against. I’m always happy to see people in EA thinking about how to avoid them in workplaces, funding, local groups, etc.

I do agree with other commenters that different backgrounds and demographics will not by themselves necessarily be good for epistemics or truth-seeking.

Thanks, this was an interesting write-up. I have one, well, let’s call it a concern or maybe caveat. You write:

Objectivity as it is understood here is a continuum between being more or less objective. Objectivity increases with the diversity of perspectives, as more perspectives in a given discussion lead to more assumptions being challenged and, thus, to an answer that is more likely to be true.

I think this relies on all perspective-havers having some shared norms that enable them to find truth collectively. Philosophy, for example, which while not a science benefits enormously from diverse viewpoints, has norms of logic, reasoning & charity that are essential to finding truth. More generally, my impression is that groups & teams function better when they have some shared values, goals & norms. So that’s the caveat that I would add – that there still need to be shared norms, at least truth-seeking norms.

This stood out to me, too. The situation that came to my mind was an extreme one, but maybe it illustrates the importance of having some basic shared beliefs about how to seek truth:

I used to live upstairs from a neighbor who was from a culture that considered witchcraft a real and serious problem. When she heard noises that she considered strange, she came to my door and told me to stop practicing witchcraft against her. She was not interested in hearing my objections that I was not practicing witchcraft, because she was very sure that I was a witch. I was very sure that I was not (but of course that’s what a witch would say!)

Let’s just say the sharing of our different perspectives did not lead to any epistemic breakthroughs.

That is true. To participate in any discussion you must know something about the topic at hand. Still, I don't think this is at odds with my post. To stay with your example in philosophy, my post does not intend to argue that basically everyone in the world should partake in philosophy discussions, but merely that the philosophy community should make sure that important perspectives are not overlooked, by a diverse set of people. Your idea of having to have a claim of expertise to meaningfully contribute to a discussion is also highlighted in "Why trust Science?" and I probably should have it highlighted more in my post. We should acknowledge that knowledge about a topic can also stem from lived experience. In addition, even if I have little knowledge about a topic I might be able to challenge an assumption that was overlooked by a homogenous group.

It seems like my post created more of a buzz than I anticipated. Many people seem to get the message from it: “we should only care about demographic diversity and nothing else”. I’m sorry that my wording was apparently so vague, as this is not really what I meant. 

To create a fruitful discussion you not only need diversity, but also at least some value alignment and some knowledge about the topic that is being discussed. Given that some value alignment and some knowledge about the topic are present, diversity of perspectives is a powerful way to make sure that the knowledge and insight gained from such a discussion increases. It allows the participants in the discussion to detect each other’s blind spots and challenge their assumptions. And this is where demographic diversity comes into play. I think that you cannot easily (or even at all) measure your subconscious assumptions and biases, but I think that those assumptions and biases originate from the experiences that you have in your life. Those experiences are strongly shaped by demographic markers like age, gender, race, etc.. Therefore, we should make sure to have enough people from different subgroups to not miss out on perspectives that would challenge erroneous assumptions in our thinking. 

Another post from the EA Forum which might have done a better job at highlighting a similar idea is “EA Diversity: Unpacking Pandora's Box”, as it unpacks the different facets of diversity explicitly. Unfortunately, I only came across it after I had published my post. 

"As we cannot measure the diversity of perspectives of a person directly, our best proxy for it is demographic diversity"

Demographic diversity is a useful proxy and may add something additional even if we did have diversity of general philosophy. However, we can measure diversity of perspectives directly, ie. by running surveys like Heterodox Academy has.

"The answer here is that objectivity is not something that a single person has, but that objectivity is a social achievement of a diverse community"

Feminism offers some valuable lens, but I feel it often leads to a hyperfocus on the underprivileged. Suppose we're discussing raising taxes on the rich, it might be useful to have a rich guy in a room. They might share some useful perspectives like, "It won't change the behaviour of my friends one bit. Most of us won't even notice. Our accountants handled our taxes, so we have no idea how much we're paying" or "If the California tax law passes, I'm headed to Texas". They might lie, but that's true of everyone. They might be biased, but the poor are likely to be biased as well.

I'm not claiming this is equally important as representing the perspectives of the poor, just that we shouldn't be hyperfocused.

I would appreciate if you could add a reference to the methods the Heterodox Academy is using to measure diversity of perspective s directly. 

Thank you for your comment. Could you tell me which part of my post led you to the conclusion that we should leave out perspectives of privileged people? I saw my argument as "include as many perspectives as you can to challenge more assumptions". Or are you making a general comment on your view of feminism?

It was a general comment how this lens is often applied in practise, even though this isn't the only possible way for it to be applied.

I'm not aware of any details about the Heterodox Academy or their methods, but according to Wikipedia, they were founded under the premise that conservative viewpoints are underrepresented in scientific discourse. That is be a valid concern, but it also means that they might have an inherent bias in their methodology because they also want to represent the underrepresented, not establish "objectivity" in itself.

The example you've given with having rich and poor people have a say in tax policy is a great example that we need diverse representation and I agree with it (which is also why I disagree with leftist slogans like "eat the rich"). However, I think it's fair to say that if a member of a more privileged group voices their worries about having to much focus on the interests of the underprivileged groups, it sounds more like someone afraid to lose their privilege disguising it by a call for balance (either intentional or subconsciously).

"They were founded under the premise that conservative viewpoints are underrepresented in scientific discourse" - that's definitely a possibility, although I suspect that for research into underrepresented groups in general almost all research will have been conducted by people withn strong pre-existing beliefs about whether or not such a group is underrepresented.

I think there's value in considering people's possible psychological motivations, but I find it more helpful to consider these for all parties. In such a conversation, the rich could very well be afraid of losing their privilege and the poor could very well be jealous or resentful.

Yes, fully agreed. It's a common cognitive bias to consider only the biases and motives of the others and not your own. I didn't want to call out the Heterodox Academy in particular, the same could be said about practically any group or movement. It just emphasizes my point that it's hard to guarantee inclusion of diverse thoughts through questionaires.