Will MacAskill - Longtermism, Altruism, History, & Technology

Dwarkesh opens by asking for the high-level explanation of effective altruism’s success, and immediately turns the question reflexive: is EA itself an instance of the contingency thesis MacAskill defends in What We Owe The Future? MacAskill accepts the framing without much resistance, but he is careful about the size of the claim. He does not say EA was impossible; he says its arrival is contingent on the order of decades. That is a narrower and more defensible claim than the dramatic version, and it matters for how the rest of the conversation is read: the argument is not that history is chaotic, but that specific intellectual packages can sit dormant for a very long time and then activate quickly when the surrounding infrastructure changes.

Evidence that Effective Altruism is somewhat contingent is that similar ideas have been promoted many times during history, and not taken on.

The evidence he marshals runs from the Mohists in ancient China — who held an impartial conception of morality and acted on it strategically by providing defensive assistance to besieged cities — through the early utilitarians, to Peter Singer in the 1970s arguing that people should donate most of their income to the very poor. Singer’s case is the sharpest of the three because the doctrine and the messenger were both present, and the movement still did not materialise until GiveWell and Giving What We Can launched around 2010. Ideas, on this account, are necessary but nowhere near sufficient; the missing ingredient was organisational infrastructure capable of converting conviction into transfers.

MacAskill’s positive explanation has two parts, and only one of them is flattering to the movement. The first is technological: the internet made it possible to aggregate dispersed like-minded people at low cost, which is precisely the coordination problem that had defeated earlier iterations. The second is frankly accidental — Alex meeting Holden, MacAskill meeting Toby. He does not try to dress these up as inevitable. The honest summary is that a good idea plus a new coordination technology plus a couple of contingent personal encounters produced a movement that could plausibly have arrived a generation later, or not at all.

I think it was a good idea waiting to happen.

This is worth flagging because it cuts against a temptation common among EA’s own advocates, which is to narrate the movement as the inevitable correction of a moral blind spot. MacAskill’s version is more interesting and less comfortable: if the same arguments failed for two and a half thousand years and then succeeded in a fifteen-year window, the causal weight sits mostly in the environment, not the argument. That has an uncomfortable implication for anyone trying to replicate the success — the playbook may be far less portable than it looks.

Dwarkesh then presses the inference that most readers of the book will have wanted to press. If moral values are highly contingent, and if we are obviously biased toward the values we were raised in, then the rational prior is that our values are a median draw from the space of live possibilities — mediocre, or worse. MacAskill does not soften this. He endorses it and then makes it vivid with the Nazi counterfactual: in a world where the Nazis won, we would all be congratulating ourselves on enormous moral progress, and the content of that self-congratulation would be monstrous. The point is not shock value. The point is that the feeling of moral progress is not evidence of moral progress, because the feeling is available to any winning ideology.

I think it should make us take seriously the fact that we’re very far away from the moral truth.

The normative conclusion MacAskill draws is explicitly anti-lock-in. The lesson of contingency is not relativism but a research programme: figure out what is actually morally right, so that the long future is shaped by correct values rather than by whichever set happened to win a contingent contest. This is a substantive move, and it is where MacAskill parts company with a certain kind of conservative reader who would take the same historical evidence and conclude that we should preserve inherited institutions precisely because reason cannot be trusted to improve on them.

Dwarkesh then separates two questions that are easy to conflate. The first is whether better values than ours are possible — trivially yes. The second is whether our values are better than a random draw would predict. MacAskill’s answer to the second is a two-step procedure: hold the prior at average, then look for specific evidence of above-average performance. He offers one piece of evidence in the other direction — had the Industrial Revolution occurred in India rather than Western Europe, large-scale factory farming, which he regards as a moral atrocity, might never have arisen — before concluding that on his view we are in fact doing better than average.

The supporting ledger is concrete: abolition, the feminist movement, liberalism, democracy. These are not abstractions for MacAskill; they are things that could have been lost, and their presence is a genuine gain rather than a mere difference. Note the structure of the argument here — it is not that our values are self-evidently correct, but that a specific list of moral achievements would plausibly have failed to occur under a redraw. That is a falsifiable-ish historical claim, and it does more work than the vague assertion that moral progress is real.

Dwarkesh’s sharpest move in this section is to turn the contingency thesis against MacAskill’s own proposal. If our values are a lucky lottery draw, then a long reflection — a millennium of deliberate moral inquiry before any irreversible commitment — is not obviously a safeguard. It may simply be a machine for regression to the mean, dissolving the lucky draw back into the median. This is a genuinely awkward objection because it uses MacAskill’s own premise to attack his own institutional recommendation.

MacAskill’s reply is that moral progress is not a random walk, because culture is acted on by identifiable forces. The most important of these is the force of argument — what the best reasons actually support. His concession is that this force is weak. That concession is doing a lot of work: if reason is weak, the long reflection is a thin reed, and the case for it rests less on the power of deliberation than on the hope that weak-but-persistent pressure beats no pressure at all.

Moral progress isn’t a random walk in general.

He then defines the proposal: before taking drastic actions that would lock in a particular set of values, put society in a state where reason, empathy, debate and good-faith moral inquiry guide which values we end up with. One editorial note on the transcript: the phrase rendered as “lumbar flexion” is clearly a mis-transcription of “long reflection”, the term MacAskill uses throughout the book. The definition is otherwise unambiguous, and it is worth being precise about what it does and does not claim — it is a procedural proposal about sequencing, not a substantive proposal about which values are correct.

I think it’s a weak force, unfortunately.

The teenage analogy from the book gets inverted by Dwarkesh in a way that is philosophically pointed. Teenagers are not only plastic and open to the future; they are also reckless, dismissive of inherited wisdom, and overconfident in their own reasoning. If humanity is at its teenage stage, why not conclude that we should be Burkean longtermists — treating the accumulated judgement of the past as a check on our inside-view fears about esoteric future risks?

MacAskill’s answer is a flat rejection, and the reasoning is naturalistic rather than political. Humans are conformist by disposition: we adopt the beliefs around us even when we cannot articulate the mechanisms that make them work. In a low-change environment this is an excellent heuristic, because the accumulated practice encodes real information about a stable world. The environment we evolved in barely moved. For hundreds of thousands of years the correct strategy was to copy.

We were hunter-gatherers for hundreds of years.

The Burkean heuristic is therefore a bet on environmental stability, and MacAskill’s claim is that the bet has been called. An economy doubling roughly every twenty years, with new technologies arriving annually, is not the world the heuristic was tuned for. When the environment changes faster than the inheritance mechanism can track, deference stops being wisdom and becomes lag. Hence first principles — not because first-principles reasoning is reliable, but because the alternative is a decision procedure calibrated to a world that no longer exists.

It means that we should be trying to figure things out from first principles.

Dwarkesh’s follow-up is the right one: granted the general point, what about the margin? If EA and longtermism are already heavily first-principles in style, would the community be better served by more history? MacAskill splits the question. On historical understanding, yes — he would love EA to have more of it. But he immediately identifies a different bottleneck: the community is comparatively rich in the analytic toolkit and comparatively poor in historical knowledge, so the marginal return is not obviously in adding more analysis. The remark as transcribed — “philosophy of economics” — is almost certainly “philosophy or economics”, and either reading makes the same point about where the community’s existing strength lies.