Envisioning the future of effective altruism

Being an EA is a weighty undertaking; I imagine the title of Effective Altruist being something that is conveyed on someone after years of fruitful service like a knighthood, not something someone calls themselves by the time they’ve ordered Doing Good Better off the Internet. I do not call myself an Effective Altruist; effective altruism is one goal I strive for, but not the only one.

This prepubescent movement has, like all good social phenomena, experienced its fair share of growing pains. There are lots of ways the movement could easily go astray and only a few ways it could truly succeed.

What might the world look like, with Effective Altruism in it? Let me sketch out three visions of the future.


In the first, EA has grown big, with members in the seven or more digits. EA leaders shout new thoughts in blog posts and conference talks like through a megaphone, desperately trying to transfer the intricate and complex knowledge they’ve developed about their respective problem-spaces into the heads of people vastly different in experience, culture and geography from them. It’s still hugely an internet phenomenon, and many hardcore EAs have never met another one in real life.

The pockets of EA physical communities collaborate wildly, but there’s always a lag between the ideology of the capital and the ideology of the colonies.

In remote outposts like new university groups sometimes it takes years for people to hear that a previously promising career track has been devalued, or that a new and promising cause area has sprung up. There are lots of shouting matches about Earning to Give.

In many of the more remote places the groups recycle the same ten intro ideas over and over again, with a revolving door of fresh acolytes learning the mantras ‘how can we use reason and evidence to do the most good?’ and something less polite along the lines of ‘shut up and multiply’ and instantly becoming insufferable in their local communities. Local leaders find the rhetoric very hit or miss amongst newcomers; precise Facebook ad targeting reveals a tight demographic – aged 20-32, mostly Caucasian males, educated at elite universities, living in countries with high trust in the government; atheism, low neuroticism, high openness, high grades in maths.

The drone of the mantras drowns out the questioning, and in the colonies people don’t question, they just leave. People wring their hands about the too-small movement numbers but the churn is just as bad. Once the honeymoon period is over and the utilitarianism starts biting newcomers in the ass, they coalesce into depressive social circles when no one feels like they are effective enough to be worth anything; there’s an exit trail littered with the broken souls of those who couldn’t stand the self-flagellation.

In the big cities of the empire, the mood is different. EAs are highly specialised, and fly frequently to efficiently-run conferences to continually explain their niche to other niches. They’ve caught the tech bros’ reputation for simplifying complex societal issues into highly distilled models that tend to break things when implemented; that one test where they got rid of all the mosquitoes in Sierra Leone still makes Effective Altruism PR people shiver. All the while the mostly white, mostly male, mostly utilitarian and tech-fluent population in the capital spends half their time publicly boasting of their effectiveness and privately dying inside of the same feeling of worthlessness that traps the EAs in the outer colonial network.

They’re getting attacked by the diversity brigade relentlessly on social media. They really can’t, for the life of them, work out why people from poor and marginalised communities don’t want to adopt the EA mindset. Who wouldn’t want to save the world effectively with them?

In the second future, the EA movement has become influential in some sorts of public policy positions. EA doctrine became much more concerned with the need for surveillance in order to prevent unfriendly AGI being developed. They haven’t yet either made or prevented a real AGI, but they’ve harnessed the less-general AIs that exist – to implement government policy for the British and Canadian governments that optimises the happiness potential of peoples’ newsfeeds and does some minor law enforcement.

More and more EAs are standing for government on a platform that some sardonically refer to as ‘Well, we’re effective, and we’re altruistic, so we’re obviously better than those ineffective slobs you’ve currently got’. The Elitism platform. They’ve got some good policies, and they want to get rid of factory farming. They’re probably a good influence?

Ten years later and there’s been a surprisingly brutal famine in poorer parts of Canada when they banned factory farming in one go and diverted a bunch of taxes to AI safety research by firing a boatload of community service workers and healthcare administrators. Nobody’s really sure though if it was the EAs’ fault, because in the meantime they’ve gotten really good at PR, and because clearing the name of Effective Altruism in these incidents is obviously better for the world overall…

I mean. just think how worse off we’d be if the EAs weren’t able to continue their good work because regular people thought some accidents were their fault.

A few years later, the deadly influenza strain in an EA biosecurity lab jumps the fence and kills 20 million people in the US and southeast Asia. Several arrests are made and some EA figureheads are hauled in front of the UN to explain themselves. They patiently explain existential risk mitigation frameworks to the simpleminded, irrational UN Assembly and assure the legislators that the chances of such an event occurring were less than 0.01% in their risk mitigation models.

Google hires a taskforce staffed entirely by EAs and throws millions at them on the condition they make a good AGI in 18 months. A handful of retired senior FBI people are hired to construct incontrovertible evidence that this is the most effective cause in existence based on the Effective Altruist literature to wave in front of the hopelessly anxious EA AI researchers. The EAs pore over the final memo but can’t find the flaws in the model, so they throw themselves into the work in earnest.

A few attempts at public arguments are hashed out online but no utilitarian objections can be found to the memo.

No one in the inner circle of EA questions the project after that.

In the third future the EA movement doesn’t exist. The idea of ‘doing the most good with the resources we have available’ is pretty common knowledge now – it’s not just a cultural meme, but embedded in the dominant political and social institutions. Givewell-like evaluators are a fundamental part of every major national policymaking institute and development agency.

There are a handful of small but well-funded international taskforces that monitor and protect against every kind of existential risk imaginable; somehow someone has made news that ‘We didn’t get hit by an asteroid this year,’ both palatable and entertaining within the news cycle and it’s a regular and welcome part of the media conversation. Major progress in slowing down the rate of development of dangerous technologies has meant that the national x-risk institutions are able to continually react and adapt and the best researchers are optimistic about humanity’s chance of survival.

We’ve stopped asking kids what they want to be when they grow up, and instead ask gentle questions about their moral values; many people pick causes to spend their entire lives on, regardless of income, and institutional programs pick up the slack to fund their work whenever the market doesn’t.

Nobody calls themselves an effective altruist; but there are historical records of the movement existing. Most of it just became the way we collectively thought about the world, and seamlessly melted into the fabric of the biggest and most widespread institutions. There’s even EA art now, even though there’s no real movement anymore; people feel genuinely grateful for the singular focus these types of questions have given to humanity and their own personal lives, and feel compelled to express that in a way that allows lots of people to share in and experience their gratitude.

Altruism, the concern and compassion for all life, present and future, is a key part of what stabilised society after the rocky 2020’s, and the ‘effectiveness’ part of EA grew a lot more nuanced and incorporated a lot more counterintuitive insights before it blossomed into a core feature of business, government and civic life. The nations look slightly different for geopolitical reasons now but the one-world, one-mission viewpoints espoused and developed by the EA movement made some lasting impact on how humans thought of themselves over the longer term.

This EA-flavoured world has learned to see suffering as its enemy, its only enemy; and with work and compassionate collaboration one that can be muted, if not destroyed.

Effective Altruism, as a movement, has its roots in positivist science, orthodox economics, and institutions glowing with privilege like Oxford with a tendency to espouse beliefs that aim at universalism. But as a social movement, it is also subject to pressures to be politicised, to grow as large as possible, and to accommodate diverse and often incompatible aims. It has also become a haven for weird ideas, about radical ideas like eradicating the malaria mosquito using gene drives and using artificial intelligence to maximise utility. The movement also has a tendency towards self-fulfilling insularity; there’s a tautological assumption that the Effective Altruism organisations working on cause X must be the most effective players in the field of cause X by virtue of being EA organisations, and that no other organisations or actors are making important progress or have valuable knowledge. I think this is myopic and likely to make EA as a movement much less effective in the long run.

Defining what is ‘altruistic’ is a group activity, and the EA movement needs to continually re-engage with this process, and encourage engagement with it at every level and in every location and not only in the inner circles of top EA organisations. Similarly, understanding how the needs, desires, biases and worldviews of the initial founders of the movement have grown out into the limitations, focus areas, and ideological frameworks that now define the movement is necessary if we want people who don’t look or think like us to be a part of it.

I think there’s an important project to be pursued in spreading the principles of effective altruism into the non-EA institutions where they can help, and inspiring people already in do-gooder roles to adopt this lens in their work. That’s a vastly different project than either growing or intensifying the Effective Altruism movement, and, I think, a more sustainable one – more of a percolation than an expansion. But it requires separating the EA identity from the movement from the values and principles that people within the movement uphold, and being willing to adapt and remix them to different contexts, in collaboration with the people who are already there.


Fluid mode via Holochain: putting theory into practice

I’ll be talking about Holochain, a crypto project I’m particularly interested in, and how it relates to how we make and share meaning in society.

This is an attempt to synthesise two illustrations I’ve previously made that point at a theoretical cluster some of us are calling metarationality, metasystematicity or fluid mode with what I think is currently one of the clearest attempts to make that theoretical cluster into a reality. I’ll try to be clear with the tech, but it does help if you understand a bit about how blockchains work and what makes them useful for decentralisation.

There is a lot of theory and implementation detail about Holochain that I’m leaving out – if you are interested I recommend you read this super clear introduction by my friend Moritz, check out more of Holochain’s online material or take a stab at their technical but well thought-out whitepaper.

Hierarchy versus grammar

When I’m ‘thinking in systems’ I’m thinking of the system as both a set of things – nodes, objects or agents – and the rules for interaction between the things. We can use this model to think about technical systems (computer networks), social systems (communities and institutions), and belief systems (ideologies and paradigms). Both the loosely-defined metarationality community and the Holochain team are rethinking how these human-made systems are designed, and what it even means for a system to be ‘designed’ at all.

I wrote in an earlier essay about existing as a human within two different types of systems – formal and playful – using the metaphors of ballet and contact dance. Within the context of being a ballet student or ballet dancer, there is a set of clear aims and a linear, planned process for training your body and mind to work together to achieve those aims. The aim is an abstract direction, a platonic ideal – the ‘perfect ballet dancer’ does not exist, but the abstract idea is largely the same for all dancers. In contact dance there is no abstract ideal. There are qualities of beauty, expression, aliveness that you might be aiming towards, but each dancer’s conception of that ideal is different. What the system provides is a grammar; rules for building interactions between dancers. There are no ‘mistakes’ in contact dance, in the sense of things that are not predicted by a formal system of movement. There are, definition, mistakes in ballet. In ballet, the purpose of the system exists outside of the movement – the goal is to be a beautiful or excellent dancer, or to perform for an audience. In contact dance the purpose is inherent in the movement; it’s just to dance.

A lot of the systems we exist within in modern society act more like formal systems, which is excellent for coordinating large groups of people or objects but not as good for maximising the flourishing of each person or organism within the system. Moreover, each formal societal system thinks it is the best, and must beat out all the other systems for the prize of becoming ‘how society works’. I’d argue that we’re in the process of noticing (slowly from the perspective of humans, quickly from the perspective of history) that no one system can actually be ‘how society works’ and that we need a new way of thinking about how these formal systems influence and relate to how society is organised.

What is the current paradigm?

In another earlier essay I discussed David Chapman’s clear but simple model that maps Robert Kegan’s stages of adult development to the organisation of society.

Stage 3 – Pre-modern, relationship-driven, pre-theoretical stage, where coordination of groups of people happens by appealing to relationships and the emotions that go along with them. (e.g. tribalism, many corporations, extended families). Group size is limited by peoples’ ability to keep track of relationships.
Stage 4 – Modern stage, theoretical stage, where coordination of groups of people happens by the evolution of a single shared belief system or set of values, and appealing to that set of values. E.g. Christianity, capitalism, democracy. Group size is often limited by the surveillance/control ability of the leaders, and their ability to suppress competing narratives.
Stage 5 – Metamodern or fluid stage, after there’s a realisation that no one system can adequately explain all of reality (*cough* the death of the metanarrative). A stable version of this metamodern stage allows people and groups to share parts of belief systems through shared grammars, and tools that help one system interact with another.

Chapman’s argument is that the postmodern movement of the last 50 yrs or so marks a transition between the modern stage and whatever this stage 5 is, after a loss of faith in belief systems we used to think were infallible and at the cusp of a larger transition to a new mode of being, and that right now there’s an opportunity to help society build a new way of relating to meaning that doesn’t rely on 100% shared data and rules.

The founders of Holochain, with a wealth of knowledge in systems thinking, mechanism and currency design, seem to be steeped already in this Stage 5 thinking. They’ve long ago somehow already drunk the Kool-Aid. They’ve argued that the fundamental shift that needs to happen in order to bring this kind of meta-system about is from designing in terms of data to designing in terms of agents. Holochains are agent-centric, which contrasts with blockchains, which are data-centric.

Hang on, since when are blockchains ‘modern’? Aren’t they meant to be all about decentralisation and not letting any centralised actor take control?

Well, Holochain argues that blockchain-based systems are just as formal and restrictive as their centralised counterparts – just in different ways. What blockchain people mean when they talk about consensus is that everybody considers the same piece of information to be true. The aim of a data-centric approach is to create one single shared data reality or truth amongst all nodes in the system (in this case, computers connecting to the network). What this means is:

‘Blockchains don’t record a universal ordering of events – they manufacture a single authoritative ordering of events – by stringing together a tiny fragment of local vantage points into one global record that has passed validation rules.’

Holochain whitepaper

In a blockchain, every single node must follow the same set of rules about what can be added to the chain, and must hold an identical copy of the whole chain. In practice, as we’ve already seen with Bitcoin and Ethereum, this means that nodes rapidly become huge, expensive and unwieldy, and require specialised equipment to run. This limits participation to a few big actors (e.g. a handful of mining companies taking advantage of cheap Chinese electricity in Inner Mongolia). Running a Bitcoin node now takes 125GB of disk space, and reports say it’s difficult, if not impossible, to run a full Ethereum node on a consumer computer. This problem will only get worse as these blockchains become more popular. In smaller blockchains it is also possible to just spend a bunch of money (with a 51% attack) in order to have complete control over the chain’s data reality.

In a data-centric system, disagreeing with the consensus is costly – you have to hard fork the entire blockchain (and lose the advantages of the network effects), or you are just evicted from a given space (in the case of centralised systems like Facebook and Twitter). Because there is only one reality, who controls that reality becomes incredibly important.

What is Holochain’s suggested alternative?

‘The term consensus, as normally used, implies deliberation with regard to differences and work on crafting a perspective that holds for all parties, rather than simply selecting one party’s dataset at random.’ 
Holochain whitepaper

According to the Holochain founders, the problem with blockchain mechanism design is the need for for ‘absolute confidence’ in order to create a trustless system.

We might be able to agree that a certain ordering of bits was placed somewhere at a certain time, but disagree on the meaning at the level of real-world identity, ownership, agreements or collective decision-making. Agents can only be sure that they see a certain piece of data and that it matches a certain model they already had. Essentially, meaning-making is a continually contested, partial and evolving activity.

The aim of an agent-centric approach is to allow nodes to share independently evolving data realities or truths. In a Holochain each node has their own personal chain, which has its own rules about what can be added to it, and what the threshold is for trusting a given data source (including other nodes). Every agent keeps track of their representation of that portion of reality that is of importance to them – given they need to manage trade-offs between the desire for trust and the cost of validation for particular circumstances. Each app asks that users use a particular set of rules in order to contribute to its shared reality, and different apps have different needs for security and cost-effectiveness. E.g. checking whether an email is spam or not and validating that a large financial transaction was sent require different levels of trust and different levels of willingness to spend resources on preventing a trust violation.

How is this metarational?

In a Holochain-based system, nodes do not need to agree to one singular set of rules and create one unified system. They only need to agree on enough rules between individual nodes and groups of nodes to establish the level of trust needed for each application they use together. As pointed out in the Holochain whitepaper, a given group of nodes could agree on a set of rules identical to those of a blockchain like Bitcoin or Ethereum – they think of a Holochain node’s parameters as being on a spectrum of highly trusting to trustless where on the ‘trustless’ end you could theoretically just replicate a blockchain within the Holochain paradigm.

On the practical, technical side, this means that from the outset Holochain avoids the scaling problems that blockchains like Bitcoin (that require validation by all nodes) are currently experiencing. On the philosophical side, this means the system does not demand a single, unified ‘truth’. You and I could run Holochain nodes on our phones that each validate the other’s data using the cryptographic hash functions that were iriginally developed as part of blockchain systems like Bitcoin. We wouldn’t need to pay to store data on the Holochains, if we were storing the data ourselves. Moreover, no two nodes would hold the same set of data, but they would share parts with each other in order to form consensus for particular applications.

I would argue that the fully realised vision of Holochain is a playful system. Each node is interacting with other nodes to build its own individual set of truths, and clusters of rules that multiple nodes follow in order to coordinate constitute ‘grammars’ or ways of playing together – rather than complete top-down recipes for defining consensus on reality. You could imagine a Holochain app being more like a grammar or a defined way for people to interact with each other in a way that is truly peer-to-peer, rather than a central ‘location’ where people connect (which is the current metaphor for apps, social networks and ‘platforms’). A truly decentralised holochain Twitter, for example, might allow you to choose your thresholds for parameters about message length, spamminess, or who can connect with you, and you wouldn’t be reliant on a central party to host your data, or protect the platform on your behalf. You could imagine it becoming easier for communities to split when differences emerge without losing network benefits – they’d just change their validation thresholds (how strict they are on who they accept) and let their membranes breathe differently – as opposed to centralised networks like FB with strictly defined, opaque membranes.

There is still a lot more to think about, and a lot more for me personally to learn about Holochain, fluid mode and agent-centric models more broadly. I don’t know how well the Holochain paradigm (and it is more of a paradigm than a single system) hold up when faced with zero-sum actors who essentially provide centralised services for simplicity and security, like Coinbase increasingly does in the regular crypto community. It doesn’t matter if the system is decentralised if you access it using a wallet whose private keys are controlled in a centralised way by a private for-profit company. Other blockchains inadvertently rely on trust in the central actors, like Ethereum’s Vitalik Buterin, in lieu of social trust in the mathematic security of the system. It remains to be seen whether Holochain as a concept could see widespread adoption without a correspondingly broad paradigm shift into positive-sum, fluid-mode thinking by a large fraction of the population. Because if it does, then adoption will be a gargantuan task, and educating app developers about the new development paradigm will require a lot of effort.

What’s next?

Personally, I’m going to continue to learn more about the Holochain paradigm, how to build apps using it, and its intellectual influences. I think the question of ‘how does an application designed for fluid mode from the outset behave in practice?’ is fascinating and a question I definitely don’t have an answer to yet. Other crypto communities like Ethereum and Bitcoin have struggled to make the user experience simple and easy, and doing so in complicated crypto systems while giving the user full control of their data and experience is hard. I also don’t really know what the developer incentives are within the system (other than ideological) – what incentivises developers to build apps that truly maximise the agentiness of their users? What happens when developer and user interests clash? I don’t know yet. Developer business models also seem tricky, and I’d like to learn more about that.

I’d also love to hear about other projects that seem to be doing this playful, fluid-mode thing in practice. One of the tricky parts is that there are so many groups with so many influences that often you have no idea that a nearby group is working on a similar thing because the language is so unfamiliar. If you know of projects like this, let me know!