Reflections on Effective Altruism

In last year or so I’ve started to come across Effective Altruism. First, it was a book in a coliving space, then it was hiring Rachel [name changed] a smart young researcher out of Oxford for a project on open medicines who turned out to be very into “EA”.

It was a fascinating encounter. As I discussed the ideas with her and read more about them I had an uncanny feeling of deja vu. It looked like I had been an “Effective Altruist” before its time”. For example, in 2002 I’d started the “What is To be Done” project with some collaborators to try and rank the optimal things we could do to improve the world. In 2004, I’d become a “make the world a better place” utilitarian economist writing papers on social cost benefit analysis and wondering about whether I should join a hedge fund as the best way to generate money for good projects.

And this deja vu extended to the limitations I had discovered with this kind of utilitarian, rationalist, metric-ized approach. The drunkard and the lamp-post problem that a focus on measurement excludes many of the most important issues. The reality of politics where rational correctness and evidence was only a (very) small part of getting something done. Buddhism mwhich taught me that utility might be a very misleading way to think about human psychology and wellbeing. Perhaps Effective Altruism will have to discover those limitations for themselves – my conversations so far with Rachel suggest that. Nevertheless, it’s worth writing down my own experiences and the lessons I’ve learnt from them in the hope they prove useful.

In a nutshell

1. Effective altruism suffers from a drunkard and lamp-post problem

Effective altruism suffers from drunkard and lamp-post problem. The story goes that one night you encounter man somewhat the worse the wear for drinking looking under a lamp-post. After a few moments watching him you ask what he is doing. He explains that he has lost his keys. You offer to help and start scouring the ground. After several minutes you don’t find anything and ask him where exactly did he lost the keys. He points back up the street. “Well why on earth are we looking here you ask?". Well there’s no light up there says the man.

So the drunkard and the lamp-post problem is that in addressing some issue we search where is light is not where the issue actually lies.

How does this relate to effective altruism? Many (most) interesting and important social issues are not amendable to randomized controlled trials or measurement in general. Most significantly, , major shifts in policy or societal organization. If you prioritize assessability then you will end up biasing towards causes that are assessable rather than most significant.

2. Over-optimistic belief that knowledge leads to action (evidence leads to policy) leading to over-emphasis on knowledge and reason

Even where there is evidence, Effective Altruism may suffer from the naive assumption that insight leads to impact, or that knowing what should be done leads to it being done (that politics is simple and logical). However, most of the time this is not so. Politics is complex and about power which has little to do with logic. This belief will lead to over-weighting on research and knowledge collection versus the hard work of doing the implementation.

3. Dis-engagement from politics and policy

Worse, this leads EA’s to disregard or disengage from politics. Because politics is “irrational” and complex is seen as a reason to steer clear of it, rather than all the more reason to engage actively and coordinately with it.

This is a second-order drunkard and lamp-post problem: precisely because of its “irrationality” and frustrating collective active problems of politics there is a distaste for politics and political (i.e. collective) solutions and a preference for technological and market-based solutions.

However, many – and most of the most important – social issues are not amenable to technological and market-based solutions. Perhaps even worse, Effective Altruism not only fails to take on the real solutions but it exacerbates the existing and problematic tendency to techno-solutionism and market-fetishisation.

4. Naive utilitarianism and neglect of true psychological mechanisms of human wellbeing and societal flourishing

Effective altruism appears to draw a great deal from rational, pragmatic utilitarianism and hence from economics, which is most sophisticated articulation of that worldview. As I have written about before (and below) there are a variety of issues with that worldview.

The one I want to emphasize here is the idea implicit in most utilitarian-type thinking that wants and preferences are given and the main way to get happier is to have more good stuff (and less bad stuff) – and it is, at least in economics, mostly “stuff” i.e. material goods: cars, sofas, houses, food etc.

The odd thing about this is that almost every major wisdom tradition on the planet, this not what is taught. In fact, often it is the opposite – you want less material things.

What if the major way to well-being is not “more stuff” but rather transforming our wants, our cravings, our “selves”? This is exactly what Buddhism teaches – one of the most explicit, “scientific”, empirically grounded wisdom traditions out there. Buddhism would imply a major inversion of classic Econ 101 approach that the problem we face is how do our best with allocations and production to satisfy unlimited wants with scarce resources. Whilst economics – and utilitarianism in general – focuses on the allocation and production question, Buddhism and wisdom traditions focus on the “transformation” question: how can we transform ourselves so that we joyful and liberated with what we already have (of course, this does not preclude interest in production and technology etc but it implies a very different emphasis).¹

And once we take this idea seriously, it has wider implications for approaches to social change, to “doing good better”. Specifically, if views and values have such impact on our personal wellbeing, then shifting those views and values should be a priority – including shifting them at the deepest level like the Buddhist teaching of non-self or codependent arising. And this shift could be crucial not only for our personal well-being but also for our ability to act together collectively to addressing the burning issues of our time – to address climate change, or handle nuclear weapons. This suggests a major role for culture and cultural evolution in any societal improvement program.

Examples

In one of my early discussions, Rachel told me that a major thing she was wrestling with waswhether to pursue her PhD offer in history or dedicate herself to effective altruism and existential risk. Her own passion was clearly her research and the PhD but she felt a lot of pressure to prioritize effective altruism – both from her own conscience and implicitly fromfrom others in the her effective altruist heavy friend network. Was her desire to study history “selfish” informed by her own personal enjoyment of this research area and not sufficiently focused on improving the lives of others? For her it seemed clear that effective altruism was the the better way if she wanted to do good in the world.

For me, it was not nearly so clear. A PhD in early nineteenth century women’s movements could seem irrelevant or “useless” – this is the classic scientism critique of the humanities. But was it?

After all, it seemed to me increasingly clear that the most important personal and social change – and improvement – came more deep shifts in our being and culture – in our underlying views and values. This was based on my training as a Buddhist, my experience as a researcher, policy-maker and activist as well as my recent work at Art / Earth / Tech. If so, the work of a historian, whose work gradually feeds into public discourse and culture, through their writings, their students etc, may be just as important as a scientist whose work helps cure malaria – though much less visible and much harder to measure.

This point applied more broadly. I had asked Rachel why climate change seemed to be pretty low in effective altruist priorities. She had explained that there the view was that climate change was already well resourced and EA should focus more on other more “neglected” areas. This is essentially a classic marginal return type idea from economics and finance: we should spend our resources on the area with the highest marginal return from the next dollar invested and that marginal returns decrease as we spend more (the first dollar spent improving my field is much more valuable than the millionth because you’ve already got most of the benefits).

But how on earth did you compute these comparative marginal returns in the case of our most significant social issues e.g. what was the return from an extra dollar on climate change vs AI risk? Whilst I wasn’t averse to using reason in these areas I was also increasingly aware of its limitations – and pitfalls. Furthermore, even if we did compute them how important were they – we already could do quick back of the envelope questions like this yet they weren’t that impactful.

Overall, my concern is that Effective Altruists will turn “rationality into a religion” and that is unlikely to go well. Rationality and reason are good servants and poor masters.

My journey

What is To Be Done

In 2002, when I was twenty-two, I started a project with two friends called “What is to be done?” You can still find a partial archive copy at http://witbd.rufuspollock.com.

The title reflected our optimistic rationality rather than any influence of Lenin’s famous work.² The project’s purpose was to determine “what is to be done” to improve our society, systematically, comprehensively and rationally.

We planned to start by analysing important policy and research areas ranging from immigration to nanotechnology. In each area, we would distill the state of the field and use this to develop policy recommendations. These conclusions would serve as the basis for future action – “what is to be done” in that area.

Finally, and this is where the project got truly grandiose, We wanted to survey all the different analyses and synthesise an overall prioritisation and set of recommendations based on some kind of comparative cost/benefit analysis.³ That way we’d know where we and others should spend our efforts: should we focus on climate change or nuclear proliferation, on nanotech or gene editing?

Properly done, this would yield an ultimate “what is to be done” list, providing a rational, analytically grounded course of action both personally and societally for where we should put our money and efforts..

Personally, this would be great because as an “analytical altruist” (my term at the time) I wanted to know where to dedicate my effort. Societally and collectively, because now we’d have a robust, built from the ground up, agreed set of priorities – no more endless debate down the pub or with friends about what was important, we could all just turn to “WhatIsToBeDone.org” and look up the answer.

After a small hiccup with the domain (whatistobedone.org was taken but witbd.org was available) the project got started.

Spirits were high and we succeeded in producing a number of decent essays on various topics ranging from the WTO to immigration.

But as with all grandiose projects, reality started to intrude. As time worn, collaborators lost interest (or got jobs) and I, as the driving force, started to worry about some key assumptions.

For example, we had a good essay on nano-technology written by a PhD student at Cambridge. And we had another written on women’s rights. Both ended with a good set of concrete recommendations (a requirement for a What is to be done essay). But how was I suppose to compare these set of recommendations and prioritise them. Should we spend more on primary school education in India or more on nanotech research in Cambridge?

And how was I to handle disagreement. For example, as I read deeper in any given area one often discovered significant differences of opinion amongst experts, differences that were hard to evaluate and resolve as an outsider.

Fortunately, I was a dedicated believer in the power of facts[^thefactz] and rational analysis. Rather than being daunted I resolved to dig further.

[^thefactz] my original c. 1999 personal website was named thefactz.org

Ater all, I knew two things: first that research and analysis did make a difference. My own initial foray into the WTO had shown that once one got beyond the polemics there were both good frameworks and a good set of more “factual” information. Put together these yielded conclusions. Of course, as a good rationalist I knew things are almost never certain, only more or less probable but nevertheless what I had done in the WTO area had been progress.

Second, any rationalist politics must implicitly depend on a solution to the commensurability problem: we must, somehow, be able to compare apples and oranges. Otherwise, how could we rationally compare the apples of more nanotech to the oranges of more primary schools in India when deciding the national budget.

Enter Economics

Thus, I began to look around for answers to these questions, answers that were coherent, systematic and powerful – in a word, rational. And I found economics.

There it was nicely set out: a nice tripartite division between preferences, the facts (parameters) and a model that would turn facts into outcomes rankable via those preferences. It even addressed those thorny issues like preference aggregation – how do we go from my and your separate preferences to joint preferences over outcomes that affect both of us?⁴

Thus, in 2004, having completed an undergrad and masters in mathematics, I started a masters and then PhD in Economics at the University of Cambridge.

Spurred by ongoing debates with my father over the merits of an intervention in Iraq and the probability of Saddam having weapons of mass destruction, I even dreamt of applying my new found tools to such issues. Maybe we could build a model of the (future) Iraq war, replete with probabilities and weightings. Even if it did not resolve debates it would at least clarify our disagreement. Was it that I and my father simply weighted US (or UK) lives differently to Iraqi ones, or did we differ in our probabilities of the existence of chemical weapons, or did we anticipate different chances of successful outcomes (and what did he mean by success)?

Such dreams occupied me.

But I was starting to have my concerns too. I’d graduated as a mathematician with a strong interest in dynamical systems – more popularly known as “chaos theory”. I understood sensitive dependence on initial conditions. I loved history and was all too aware of the extraordinary contingency of human affairs. Moreover, as I started to study my economics I found some concrete, and delightly analytical, limits on my dreams.

Just as Godel’s theorem used the rigours of logic to expose its limits, so Arrow’s impossibility theorem illustrated the difficulties of aggregating preferences in a coherent way. As I delved deeper into issues ranging from free trade to environmentalism I encountered the difficulties of trade-offs, the intruding of the irreducible complexities of politics: yes, free trade was better than protectionism but only if winners compensated losers promptly and fairly. Economists had a magical charm to stave off descent into the quagmire of tradeoffs called Pareto optimality but it just demonstrated the problem: few, if any, real-life situations conformed to the rigours of Pareto optimality in which the lot of every participant must be improved or unchanged.⁵

I was not giving up, but i was rapidly becoming a pragmatic rationalist economist just as most of us are pragmatic democrats: the worst except for all the rest. At least, I told myself, economics had an explicit framework, plus rigour enough for anyone (even if that rigour seemed a bit forced and the maths a bit jejeune).

But I was still having doubts. Take for example, one of my favourite areas: social cost-benefit analysis.⁶ The idea of social cost-benefit analysis is to bring some rigour to comparing policy choices: do we build more schools or hospitals, should we spend $1bn on expanding that road. This sounded great. I even wrote a major policy paper for the UK government on open data using these techniques.

However, as I dug in it was apparent that the framework only helped address the least important parts of the problem (discount factors and some aggregation) and even there only under restrictive assumptions. The real heart of most cost-benefit was assessing the costs and benefits and here you were on your own.⁷ Worse, even the stuff that was useful in the theory could be made to blow up with a few (realistic) tweaks to the basic assumptions thereby rendering the whole approach of limited value.⁸ After all, we might as well just use our own common sense and judgment rather than bringing in the fancy economics framework if the rigour does not buy us anything – or worse, adds to the confusion.

The realities of politics

In addition, I started to encounter political reality. Between 2006 and 2008 I had been part of teams that had written two expert papers for the UK government: one on the politically contentious matter of copyright term extension, the second on the question of how government should price the data they have such as maps and the company registry.⁹

On copyright our advice was clear: the most basic accounting logic showed that a retrospective copyright term extension was a terrible idea as it involved a significant transfer of wealth from citizens to a few rich performers and a small group of multinational entertainment conglomerates and resulted in an overall net loss to UK citizens. Informed by this, the official Gowers review strongly recommended that copyright in sound recordings should not be extended. Further support was provided by a formal letter from a group of world-famous economists including a Nobel prize winner (Ken Arrow).

At first, the government in the form of Gordon Brown, then head of the Treasury, accepted the advice and established a clear UK position against extension. With the UK against, a term extension was dead in the water (the UK has the largest recording industry in Europe). But then some months later the fight heated up politically and one night the relevant minister (David Lamy) did a sudden and totally unexplained volte face. There was no consultation and no explanation for this change was ever provided (the best I got was a wry comment off the record from a senior civil servant to the effect that the head of the BPI, the UK record industry lobbying association, had the direct personal phone number of the minister – an indication of the power and access the industry had).

My naive belief in evidence based policy took a further blow. I was further disillusioned by the European Commission’s recommendation for term extension which was based on an evidence assessment so shoddy and inadequate that it demonstrated straight up intellectual dishonesty in pursuit of its clearly pre-ordained political commitment to finding a given answer – itself largely determined by political pressure and lobbying (combined with the selection effect that only those with a certain mindset become Commissioners in the relevant department).

Copyright was politically contentious. But what about the pricing for government data. Surely this was simpler. Once again i was part of a team of Cambridge economists and legal experts who drafted a report. Again its recommendations were unambiguous: the government should price most of its data at marginal cost (i.e. zero for digital data), especially the key datasets we were looking at such as maps. Report was duly submitted and … nothing happened.

The basic reason was obvious: there was some powerful internal government interests against this change, not least the government agencies that had been set up with mini-monpolies over these datasets (such as the Ordnance Survey for maps). In addition, the Treasury (who had commissioned the report) were themselves conflicted as getting the benefits of opening up data required some immediate investment from them. Whilst the analysis implied this was a great investment to make (for every pound in you would get something like 10 pounds back) thee was a strong short-term concern against spending in this area. Thus, once again something that was a slam-dunk in terms of rational analysis died for “political” reasons. (As a footnote, the opening up of public data would be rapidly resurrected by the Conservative coalition in 2010 and much data was released though much of the most important data such as maps is still not open as of 2017).

This experience, removed what little was left of my illusions. All my elaborate (or simple) cost-benefit analysis had been for naught.

Importantly, this didn’t make me despair of policy or politics. The lesson I drew was not to give up on policy-making – which was clearly hugely important. Rather, it taught me to be much realistic about the role that “rational analysis” played in the process. Politicians regarded economists as some kind of mix of voodoo priests and hack journalists, treating them with an unstable mixture of incomprehending semi-respect and lackey-disdain.¹⁰

To the extent economics was used, it was usually as just another weapon in some larger political battle. And, furthermore, a weapon that was seldom significant and certainly not decisive. And this wasn’t something about economics: rational debate, reasoned analysis, real evidence. All of these seemed to have much less significance in policy debates than I might have imagined.

Thus, even when evidence and analysis were relevant they seemed to have relatively little weight.

---

1. I wrote a whole post on this point nearly a decade ago entitled Buddhist Economics ↩︎

2. When we started the project we were not even aware of Lenin’s “What is to be done?". Once the project got online we noticed it thanks to the proximity our site and results for Lenin in online searches! ↩︎

3. Fortunately, at this early point in the project we were all sufficiently in-expert to be pretty confident this wouldn’t be too hard to do – surely someone out there had thought about this before! ↩︎

4. Though as we shall see below, it isn’t as if economics and choice theory have a nice answer. Arrow’s impossibility theorem and similar tell us that aggreation is hard. Though that doesn’t trouble the social cost benefit analysts too much. ↩︎

5. A lot of important economic theorems like the Fundamental Theorems of Welfare Economics use Pareto optimality but they “cheat” in permitting reallocation (at least in “initial endowments”). In this sense, Pareto optimality is reduced to utility maximization and utility maximization is a much weaker requirement. ↩︎

6. See my overview here for an introduction https://rufuspollock.com/papers/scba.pdf ↩︎

7. This makes sense: there is no general answer to how to calculate the cost or benefits of a given project, especially most interesting real world ones e.g. the invasion or Iraq, or funding mathematics research. For me it was the realisation of misplaced hopes. Like the moment when you wake up from an infatuation and realise the other person is human with all their natural, human flaws. It makes sense but you are still disappointed. And economic’s tendency to coat basic ideas in high sounding language and a bit of algebra contributes to these misplaced expectations: coming from hard maths backgorund, seeing all that algebra made me expect a bit more substance. ↩︎

8. Suppose, for example, we are working on climate changee. A basic question would be should we spend one dollar today on preventing or mitigating climate change to gain some benefit X dollars in, say, twenty years. To decide this we obviously need to know X but we also need to know how to compare dollars today with dollars in twenty years. This latter issue is called discounting and is the part where social cost benefit theory has something useful to say. To work out this discount we want to ask what the value of a dollar is in twenty years compared today. That in turn relates a lot to how well off we are in the future. If we are rich then an extra dollar is not worth much. But if we are poor it is worth a lot. Thus, one’s assumption on the path of future growth has a big impact on the discount rate because it determines, in turn, (the probability) of how rich or poor we are.

       It turns out that a small tweak to the standard normal model of the growth rate can "blow-up" future utility (expected future utility is infinite) thereby leading to an undefined discount rate. With infinite future expected utility we should be willing to spend practically anything today because its future value (in expectation) is infinite. This clearly makes a mockery of the analysis and renders the whole framework of dubious value.
           
       I first came across this neat point in Harvard environmental economist Marty Weitzman's Sept 2007 JEL review of the Stern report https://scholar.harvard.edu/files/weitzman/files/review_of_stern_review_jel.45.3.pdf. "The bombshell fact that EMU [Expected Marginal Utility] = + infinity (as soon as we admit that we don’t know the underlying stochastic structure, and therefore parameters must be estimated) changes the rules of the game." In short, replacing the normal distribution for growth rate with a fatter-tailed t-distribution results in an infinite value for expected future utility.
   

   ↩︎

9. And from 2003-2005 I’d already been working in policy opposing the software patents directive (successfully – one of the few successes we had and mainly because we had small business support and most software big business is in the US). ↩︎

10. It reminded of the Lyndon Johnson quip: “Making a speech on economics is a lot like pissing down your leg. It may seem hot to you,but it never does to anyone else.” Robert A. Caro, Master of the Senate: The Years of Lyndon Johnson (2012) ↩︎