I agree that carefully-vetted institutional solutions are probably where one would like to end up.

I agree that the EMH-consistent version of this still suggests the collective EA portfolio should be more leveraged, and more managing correlations across donors now and over time, and that there is a large factor literature in support of these factors (although in general academic finance suffers from datamining/backtesting and EMH dissipation of real factors that become known).

Re the text you quoted, I just mean that if EAs damage their portfolios (e.g. by taking large amounts of leverage and not properly monitoring it so that leverage ratios explode, taking out the portfolio) that's fewer EA dollars donated (aside from the reputational effects), and I would want to do more to ensure readers don't go half-cocked and blow up their portfolios without really knowing what they are doing.

I appreciate many important points in this essay about the additional considerations for altruistic investing, including taking more risk for return than normal because of lower philanthropic risk aversion, attending to correlations with other donors (current and future), and the variations in diminishing returns curve for different causes and interventions popular in effective altruism. I think there are very large gains that could be attained by effective altruists better making use of these considerations.

But at the same time I am quite disconcerted by the strong forward-looking EMH violating claims about massively outsized returns to the specific investment strategies, despite the limited disclaimers (and the factor literature). Concern for relative performance only goes so far as an explanation for predicting such strong inefficiencies going forward: the analysis would seem to predict that, e.g. very wealthy individuals investing on their own accounts will pile into such strategies if their advantage is discernible. I would be much less willing to share the article because of the inclusion of those elements.

I would also add some of the special anti-risk considerations for altruists and financial writing directed at them, e.g.

• Investment blowups that cause personal financial difficulties attributed to effective altruism having fallout on others
• Investment writings taken as advice damaging especially valuable assets that would otherwise be used for altruism (just the flip side of the value of improvements)
• Relative underperformance (especially blame-drawing) of the broad EA portfolio contributing to weaker reputation for effective altruism, interacting negatively with the very valuable asset (much of the EA portfolio) of entry of new altruists

On net, it still looks like EA should be taking much more risk than is commonly recommended for individual retirement investments, and I'd like to see active development of this sort of thinking, but want to emphasize the importance of caution and rigor in doing so.

Interactive Brokers allows much higher leverage for accounts with portfolio margin enabled, e.g. greater than 6:1. That requires options trading permissions, in turn requiring some combination of options experience and an online (easy) test.

I would be more worried about people blowing up their life savings with ill-considered extreme leverage strategies and the broader fallout of that.

Agreed (see this post for an argument along these lines), but it would require much higher adoption and so merits the critique relative to alternatives where the donations can be used more effectively.

I have reposted the comment as a top-level post.

My sense of what is happening regarding discussions of EA and systemic change is:

• Actual EA is able to do assessments of systemic change interventions including electoral politics and policy change, and has done so a number of times
• Empirical data on the impact of votes, the effectiveness of lobbying and campaign spending work out without any problems of fancy decision theory or increasing marginal returns
• E.g. Andrew Gelman's data on US Presidential elections shows that given polling and forecasting uncertainty a marginal vote in a swing state average something like a 1 in 10 million chance of swinging an election over multiple elections (and one can save to make campaign contributions
• 80,000 Hours has a page (there have been a number of other such posts and discussion, note that 'worth voting' and 'worth buying a vote through campaign spending or GOTV' are two quite different thresholds) discussing this data and approaches to valuing differences in political outcomes between candidates; these suggest that a swing state vote might be worth tens of thousands of dollars of income to rich country citizens
• But if one thinks that charities like AMF do 100x or more good per dollar by saving the lives of the global poor so cheaply, then these are compatible with a vote being worth only a few hundred dollars
• If one thinks that some other interventions, such as gene drives for malaria eradication, animal advocacy, or existential risk interventions are much more cost-effective than AMF, that would lower the value further except insofar as one could identify strong variation in more highly-valued effects
• Experimental data on the effects of campaign contributions suggest a cost of a few hundred dollars per marginal vote (see, e.g. Gerber's work on GOTV experiments)
• Prediction markets and polling models give a good basis for assessing the chance of billions of dollars of campaign funds swinging an election
• If there are increasing returns to scale from large-scale spending, small donors can convert their funds into a small chance of huge funds, e.g. using a lottery, or more efficiently (more than 1/1,000 chance of more than 1000x funds) through making longshot bets in financial markets, so IMR are never a bar to action (also see the donor lottery)
• The main thing needed to improve precision for such estimation of electoral politics spending is carefully cataloging and valuing different channels of impact (cost per vote and electoral impact per vote are well-understood)
• More broadly there are also likely higher returns than campaign spending in some areas such as think tanks, lobbying, and grassroots movement-building; ballot initiative campaign spending is one example that seems like it may have better returns than spending on candidates (and EAs have supported several ballot initiatives financially, such as restoration of voting rights to convicts in Florida, cage bans, and increased foreign spending)
• A recent blog post by the Open Philanthropy Project describes their cost-effectiveness estimates from policy search in human-oriented US domestic policy, including criminal justice reform, housing reform, and others
• It states that thus far even ex ante estimates of effect there seem to have only rarely outperformed GiveWell style charities
• However it says: "One hypothesis we’re interested in exploring is the idea of combining multiple sources of leverage for philanthropic impact (e.g., advocacy, scientific research, helping the global poor) to get more humanitarian impact per dollar (for instance via advocacy around scientific research funding or policies, or scientific research around global health interventions, or policy around global health and development). Additionally, on the advocacy side, we’re interested in exploring opportunities outside the U.S.; we initially focused on U.S. policy for epistemic rather than moral reasons, and expect most of the most promising opportunities to be elsewhere. "
• Let's Fund's fundraising for climate policy work similarly made an attempt to estimate the impacts of their proposed intervention in this sort of fashion; without endorsing all the details of their analysis, I think it is an example of EA methodologies being quite capable of modeling systemic interventions
• Animal advocates in EA have obviously pursued corporate campaigns and ballot initiatives which look like systemic change to me, including quantitative estimates of the impact of the changes and the effects of the campaigns
• The great majority of critics of EA invoking systemic change fail to present the simple sort of quantitative analysis given above for the interventions they claim, and frequently when such analysis is done the intervention does not look competitive by EA lights
• A common reason for this is EAs taking into account the welfare of foreigners, nonhuman animals and future generations; critics may propose to get leverage by working through the political system but give up on leverage from concern for neglected beneficiaries, and in other cases the competition is interventions that get leverage from advocacy or science combined with a focus on neglected beneficiaries
• Sometimes systemic change critiques come from a Marxist perspective that assumes Marxist revolution will produce a utopia, whereas empirically such revolution has been responsible for impoverishing billions of people, mass killing, the Cold War, (with risk of nuclear war) and increased tensions between China and democracies, creating large object-level disagreements with many EAs who want to accurately forecast the results of political action
• Nonetheless, my view is that historical data do show that the most efficient political/advocacy spending, particularly aiming at candidates and issues selected with an eye to global poverty or the long term, does have higher returns than GiveWell top charities (even ignoring nonhumans and future generations or future technologies); one can connect the systemic change critique as a position in intramural debates among EAs about the degree to which one should focus on highly linear, giving as consumption, type interventions
• E.g. I would rather see $1000 go to something like the Center for Global Development, Target Malaria's gene drive effort, or the Swiss effective foreign aid ballot initiative than the Against Malaria Foundation
• I do think it is true that well-targeted electoral politics spending has higher returns than AMF, because of the impacts of elections on things such as science, foreign aid, great power war, AI policy, etc, provided that one actually directs one's efforts based on the neglected considerations
• EAs who are willing to consider riskier and less linear interventions are mostly already pursuing fairly dramatic systemic change, in areas with budgets that are small relative to political spending (unlike foreign aid),
• Global catastrophic risks work is focused on research and advocacy to shift the direction of society as a whole on critical issues, and the collapse of human civilization or its replacement by an undesirable successor would certainly be a systemic change
• As mentioned previously, short-term animal EA work is overwhelmingly focused on systemic changes, through changing norms and laws, or producing technologies that would replace and eliminate the factory farming system
• A number of EA global poverty focused donors do give to organizations like CGD, meta interventions to grow the EA movement (which can eventually be cashed in for larger systemic change), and groups like GiveWell or the Poverty Action Lab
• Although there is a relative gap in longtermist and high-risk global poverty work compared to other cause areas, that does make sense in terms of ceiling effects, arguments for the importance of trajectory changes from a longtermist perspective, and the role of GiveWell as a respected charity evaluator providing a service lacking for other areas
• Issue-specific focus in advocacy makes sense for these areas given the view that they are much more important than the average issue and with very low spending
• As funding expands in focused EA priority issues, eventually diminishing returns there will equalize with returns for broader political spending, and activity in the latter area could increase enormously: since broad political impact per dollar is flatter over a large range political spending should either be a very small or very large portion of EA activity
• Essentially, the cost per vote achieved through things like campaign spending is currently set by the broader political culture and has the capacity to absorb billions of dollars at similar cost-effectiveness to the current level, so it should either be the case that EA funds very little of it or enormous amounts of it
• There is a complication in that close elections or other opportunities can vary the effectiveness of political spending over time, which would suggest saving most funds for those
• The considerations are similar to GiveDirectly: since cash transfers could absorb all EA funds many times over at similar cost-effectiveness (with continued rapid scaling), it should take in either very little or almost all EA funding; in a forced choice it should either be the case that most funding goes to cash transfers, whereas for other interventions with diminishing returns on the relevant scale as mixed portfolio will yield more impact
• For now areas like animal advocacy and AI safety with budgets of only tens of millions of dollars are very small relative to political spending, and the impact of the focused work (including relevant movement building) makes more of a difference to those areas than a typical difference between political candidates; but if billions of dollars were being spent in those areas it would seem that political activity could be a competitive use (e.g. supporting pro-animal candidates for office)

My read is that Millenarian religious cults have often existed in nontrivial numbers, particularly but as you say the idea of systematic, let alone accelerating, progress (as opposed to past golden ages or stagnation) is new and coincided with actual sustained noticeable progress. The Wikipedia page for Millenarianism lists ~all religious cults, plus belief in an AI intelligence explosion.

So the argument seems, first order, to reduce to the question of whether credence in AI growth boom (to much faster than IR rates) is caused by the same factors as religious cults rather than secular scholarly opinion, and the historical share/power of those Millenarian sentiments as a share of the population. But if one takes a narrower scope (not exceptionally important transformation of the world as a whole, but more local phenomena like the collapse of empires or how long new dynasties would last) one sees smaller distortion of relative importance for propaganda frequently (not that it was necessarily believed by outside observers).

She’s unsure whether this speeds up or slows down AI development; her credence is imprecise, represented by the interval [0.4, 0.6]. She’s confident, let’s say, that speeding up AI development is bad.

That's an awfully (in)convenient interval to have! That is the unique position for an interval of that length with no distinguishing views about any parts of the interval, such that integrating over it gives you a probability of 0.5 and expected impact of 0.

The standard response to that is that you should weigh all these and do what is in expectation best, according to your best-guess credences. But maybe we just don’t have sufficiently fine-grained credences for this to work,

If the argument from cluelessness depends on giving that kind of special status to imprecise credences, then I just reject them for the general reason that coarsening credences leads to worse decisions and predictions (particularly if one has done basic calibration training and has some numeracy and skill at prediction). There is signal to be lost in coarsening on individual questions. And for compound questions with various premises or contributing factors making use of the signal on each of those means your views will be moved by signal.

Chapter 3 of Jeffrey Friedman's book War and Chance: Assessing Uncertainty in International Politics presents data and arguments showing large losses from coarsening credences instead of just giving a number between 0 and 1. I largely share his negative sentiments about imprecise credences, especially.

[VOI considerations around less investigated credences that are more likely to be moved by investigation are fruitful grounds to delay action to acquire or await information that one expects may be actually attained, but are not the same thing as imprecise credences.]

(In contrast, it seems you thought I was referring to AI vs some other putative great longtermist intervention. I agree that plausible longtermist rivals to AI and bio are thin on the ground.)

That was an example of the phenomenon of not searching a supposedly vast space and finding that in fact the # of top-level considerations are manageable (at least compared to thousands), based off experience with other people saying that there must be thousands of similarly plausible risks. I would likewise say that the DeepMind employee in your example doesn't face thousands upon thousands of ballpark-similar distinct considerations to assess.

I think that is basically true in practice, but I am also saying that even absent those pragmatic considerations constraining utilitarianism, I still would hold these other non-utilitarian normative views and reject things like not leaving some space for existing beings for a tiny proportional increase in resources for utility monsters.

The first words of my comment were "I don't identify as a utilitarian" (among other reasons because I reject the idea of things like feeding all existing beings to utility monsters for a trivial proportional gains to the latter, even absent all the pragmatic reasons not to; even if I thought such things more plausible it would require extreme certainty or non-pluralism to get such fanatical behavior).

I don't think a 100% utilitarian dictator with local charge of a society on Earth removes pragmatic considerations, e.g. what if they are actually a computer simulation designed to provide data about and respond to other civilizations, or the principle of their action provides evidence about what other locally dominant dictators on other planets will do including for other ideologies, or if they contact alien life?

But you could elaborate on the scenario to stipulate such things not existing in the hypothetical, and get a situation where your character would commit atrocities, and measures to prevent the situation hadn't been taken when the risk was foreseeable.

That's reason for everyone else to prevent and deter such a person or ideology from gaining the power to commit such atrocities while we can, such as in our current situation. That would go even more strongly for negative utilitarianism, since it doesn't treat any life or part of life as being intrinsically good, regardless of the being in question valuing it, and is therefore even more misaligned with the rest of the world (in valuation of the lives of everyone else, and in the lives of their descendants). And such responses give reason even for utilitarian extremists to take actions that reduce such conflicts.

Insofar as purely psychological self-binding is hard, there are still externally available actions, such as visibly refraining from pursuit of unaccountable power to harm others, and taking actions to make it more difficult to do so, such as transferring power to those with less radical ideologies, or ensuring transparency and accountability to them.