This is a linkpost for https://globalprioritiesinstitute.org/wp-content/uploads/2019/Mogensen_Maximal_Cluelessness.pdf
Andreas Mogensen, a Senior Research Fellow at the Global Priorities Institute, has just published a draft of a paper on "Maximal Cluelessness". Abstract:
I argue that many of the priority rankings that have been proposed by effective altruists seem to be in tension with apparently reasonable assumptions about the rational pursuit of our aims in the face of uncertainty. The particular issue on which I focus arises from recognition of the overwhelming importance and inscrutability of the indirect effects of our actions, conjoined with the plausibility of a permissive decision principle governing cases of deep uncertainty, known as the maximality rule. I conclude that we lack a compelling decision theory that is consistent with a long-termist perspective and does not downplay the depth of our uncertainty while supporting orthodox effective altruist conclusions about cause prioritization.
Mogensen writes (p. 20):
These concerns are forceful, but don't seem to generalize to all intervention types aimed at improving the long-term future. If one believes that the readily available evidence is insufficient to constrain our decision making, one still can accumulate resources to be disbursed at a later time when good enough evidence emerges. Although we may at present be radically uncertain about the sign and the magnitude of most far-future interventions, the intervention of accumulating resources for future disbursal does not itself appear to be subject to such radical uncertainty.
Robin Hanson, Paul Christiano, and others have made similar points in the past.
Hanson (2014):
Christiano (2014):
By accumulating resources for the future, we give increased power to whatever decision-makers in the future we bequeath these resources. (Whether these decision-makers are us in 20 years, or our descendants in 200 years.)
In a clueless world, why do we think that increasing their power is good? What if those future decision makers make a bad decision, and the increased resources we've given them mean the impact is worse?
In other words, if we are clueless today, why will we be less clueless in the future? One might hope cluelessness decreases monotonically over time, as we learn more, but so does the probability of a large mistake.
Indefinite accumulation of resources probably also increases the chance of being targeted by resource-seeking groups with military & political power.
I'm pretty sceptical of arguments for cluelessness. Some thoughts:
I appreciate you making this point, as I think it's interesting and I hadn't come across it before. However, I don't currently find it that compelling, for the following reasons [these are sketches, not fully fleshed out arguments I expect to be able to defend in all respects]:
Very interesting comment!
I don't think this defence works, because some of your current preferences are manifestly about future events. Insisting that all these preferences are ultimately about the most immediate causal antecedent (1) misdescribes our preferences and (2) lacks a sound theoretical justification. You may think that Parfit's arguments against S provide such a justification, but this isn't so. One can accept Parfit's criticism and reject the view that what is rational for an agent is to maximize their lifetime wellbeing, accepting instead a view on which it is rational for the agent to satisfy their present desires (which, incidentally, is not Parfit's view). This in no way rules out the possibility that some of these present desires are aimed at future events. So the possibility that you may be clueless about which course of action satisfies those future oriented desires remains.
Thank you for raising this, I think I was too quick here in at least implicitly suggesting that this defence would work in all cases. I definitely agree with you that we have some desires that are about the future, and that it would misdescribe our desires to conceive all of them to be about present causal antecedents.
I think a more modest claim I might be able to defend would be something like:
The justification of everyday actions does not require an appeal to preferences with the property that, epistemically, we ought to be clueless about their content.
For example, consider the action of not wandering blindly into the road. I concede that some ways of justifying this action may involve preferences about whose contents we ought to be clueless - perhaps the preference to still be alive in 40 years is such a preference (though I don't think this is obvious, cf. "dodge the bullet" above). However, I claim there would also be preferences, sufficient for justification, that don't suffer from this cluelessness problem, even though they may be about the future - perhaps the preference to still be alive tomorrow, or to meet my friend tonight, or to give a lecture next week.
On the biting the bullet answer, that doesn't seem plausible to me. The preference we have are a product of the beliefs we have about what will make our lives better over the long-run. My preference not to smoke is entirely a product of the fact that I believe that it will increase my risk of premature death. Per proponents of cluelessness, I could argue "maybe it will make me look cool to smoke, and that will increase my chances of getting a desirable partner" or something like that. In that sense the sign of the effect of smoking on my own interests is not certain. Nevertheless, I think it is irrational to smoke. I don't think a Parfitian understanding of identity would help here because then my refusal to smoke would be altruistic - I would be helping out my future self.
The dodge the bullet answer is more plausible, and I may follow up with more later.
I think this is precisely what I'm inclined to dispute. I think I simply have a preference against premature death, and that this preference doesn't rest on any belief about my long-run wellbeing. I think my long-run wellbeing is way too weird (in the sense that I'm doing things like hyperbolic discounting anyway) and uncertain to ground such preferences.
Maybe this points to a crux here: I think on sufficiently demanding notions of rationality, I'd agree with you that considerations analogous to cluelessness threaten the claim that smoking is irrational. My impression is that perhaps the key difference between our views is that I'm less troubled by this.
I'm inclined to agree. Just to clarify though, I wasn't referring to Parfit's claims about identity, which if I remember correctly are in the second or third part of Reasons and Persons. I was referring to the first part, where he among other things discusses what he calls the "self-interest theory S" (or something like this).
I agree. More specifically, I think the argument for cluelessness is defeatable, and tentatively think that we know of defeaters in some cases. Concretely, I think that we are justified in believing in the positive expected value of (i) avoiding human extinction and (ii) acquiring resources for longtermist goals. (Though I do think that for none of these it is obvious that their expected value is positive, and that considering either to be obvious would be a serious epistemic error.)
I think you overstate your case here. I agree in principle that "the level of uncertainty will always be almost entirely dependent on the facts about the particular case," and so that whether we are clueless about any particular decision is a contingent question. However, I think that inspecting the arguments for cluelessness about, say, the effects of donations to AMF do suggest that cluelessness will be pervasive, for reasons we are in principle able to isolate. To name just one example, many actions will have small but in expectation non-zero, highly uncertain effect on the pace of technological growth; this in turn will have an in expectation non-zero, highly uncertain net effect on the risk of human extinction, which in turn ... - I believe this line of reasoning alone could be fleshed out into a decisive argument for cluelessness about a wide range of decisions.
Hi Max,
I would be curious to know whether you still basically believe this, and whether you have meanwhile became convinced of the robustness of other actions.
(personal views only) In brief, yes, I still basically believe both of these things; and no, I don't think I know of any other type or action that I'd consider 'robustly positive', at least from a strictly consequentialist perspective.
To be clear, my belief regarding (i) and (ii) is closer to "there exist actions of these types that are robustly positive", as opposed to "any action that purports to be of one these types is robustly positive". E.g., it's certainly possible to try to reduce the risk of human extinction but for that attempt to be ineffective or even counterproductive (i.e., to on net increase the risk of extinction, or to otherwise cause significant harms such that I'd consider the action impermissible), it's possible for resources that were acquired for impartial welfarist purposes to eventually be misused, etc.,
I made some nuanced updates about "acquiring resources for longtermist goals", but they are mostly things like me having become more or less excited about particular examples/substrategies, me having somewhat richer views on some pitfalls of that strategy (though I don't think I became aware of qualitatively 'new' pitfalls), etc., as opposed to sweeping updates about that whole class of actions and whether they can be robustly positive.
Thanks! I think I have converged towards a similar view.
On the latter, yes that is a good point - there are general features at play here, so I retract my previous comment. However, it still seems true that your rational credal state will always depend to a very significant extent on the particular facts.
I find the use of the long-termist point of view a bit weird as applied to the AMF example. AMF is not usually justified from a long-termist point of view, so it is not really surprising that its benefits seem less obvious when you consider it from that point of view.
I agree in principle. However, there are a few other reasons why I believe making this point is worthwhile:
As I say in another comment, I think that a few effects - such as reducing the risk of human extinction - can be rescued from cluelessness. Therefore, I'm not committed to being indifferent between literally all actions.
I do, however, think that consequentialism provides a reason for only very few actions. In particular, I do not think there is a valid argument for donating to AMF instead of the Make-a-Wish Foundation based on consequentialism alone.
This is actually one example of where I believe cluelessness has practical import. Here is a related thing I wrote a few months ago in another discussion:
"Another not super well-formed claim:
- Donating 10% of one's income to GiveWell charities, prioritizing to reduce chicken consumption over reducing beef consumption, and similar 'individual' actions by EAs that at first glance seem optimized for effectiveness are valuable almost entirely for their 'symbolic' and indirect benefits such as signalling and maintaining community norms.
- Therefore, they are analogous to things like: environmentalists refusing to fly or reducing the waste produced by their household; activists participating in a protest; party members attending weekly meetings of their party; religious people donating money for missionary purposes or building temples.
- Rash criticism of such actions in other communities that appeals to their direct short-term consequences is generally unjustified, and based on a misunderstanding of the role of such actions both within EA and in other communities. If we wanted to assess the 'effectiveness' of these other movements, the crucial question to ask (ignoring higher-level questions such as cause prioritization) about, say, an environmentalist insisting to always switch of the lights when they leave a room, would not be how much CO2 emissions are avoided; instead, the relevant questions would be things like: How does promoting a norm of switching off lights affect that community's ability to attract followers and other resources? How does promoting a norm of switching off lights affect that community's actions in high-stakes situations, in particular when there is strategic interdependence -- for example, what does it imply about the psychology and ability to make credible commitments of a Green party leader negotiating a coalition government?
- It is not at all obvious that promoting norms that are ostensibly about maximizing the effectiveness of all individual 'altruistic' decisions is an optimal or even net positive choice for maximizing a community's total impact. (Both because of and independently of cluelessness.) I think there are relatively good reasons to believe that several EA norms of that kind actually have been impact-increasing innovations, but this is a claim about a messy empirical question, not a tautology."
Thanks, Max, this is interesting.
Suppose that it is true that the value of those actions comes almost entirely from their symbolic benefits. If so, then a further question is whether those symbolic benefits are dependent on the belief that that is not the case; i.e. the belief that the value of those actions, on the contrary, largely comes from their direct and non-symbolic effects. (Analogously to how indirect benefits of a religion on well-being or community cohesion may be dependent on the false belief that the religion's metaphysical claims are true.) It could be that making it widely known that the value of those actions comes almost entirely from their symbolic benefits would undermine those benefits (maybe even turn them to harms; e.g. because knowingly doing something with low direct benefits for symbolic reasons would be seen as hypocritical). Whether that's the case depends on the social context and doesn't seem straightforward to determine.
I agree this is a non-obvious question. There is a good reason why consequentialists at least since Sidgwick have asked to what extent the correct moral theory might imply to keep its own principles secret.
Yes, though it seems to me that EAs largely think one shouldn't (cf. that Integrity is one of "the guiding principles of effective altruism" as understood by a number of organisations). (Not that you would suggest otherwise.)
A tangentially related comment. What symbolic benefits or harms our actions have will be dependent on our norms, and these norms will to at least some extent be malleable. Jason Brennan has argued that we should judge such symbolic norms by their consequences.
So, we shouldn't just take symbolic benefits into account when we prioritise what action to take, but we should also consider whether to change our symbolic norms, so that the symbolic benefits (which are a consequence of those norms) change. Brennan argues that if epistocracy produces greater direct benefits than democracy, then we should change our symbolic norms so that democracy doesn't yield greater symbolic benefits than epistocracy. Similarly, one could argue that if some effective altruist intervention produces greater direct benefits than some other effective altruist intervention (say diet change), then we should change our symbolic norms so that the latter doesn't yield greater symbolic benefits than the former.
[Edit: I realise now that the last paragraph in your above comment touches on these issues.]
Hi Max, I think the link is broken. Maybe it is this one?
I don't remember I'm afraid. I don't recall having seen the article you link to, so I doubt it was that. Maybe it was this one.
Thanks for this! - My tentative view is that cluelessness is an important issue with practical implications, and so I'm particularly interested in thoughtful arguments for opposing views.
I'll post some reactions in separate comments to facilitate discussion.
I agree that are strong arguments that credence functions should be sharp. So I don't think the case for cluelessness is a slam dunk. (Granting that, roughly speaking, considering cluelessness to be an interesting problem commits one to a view using non-sharp credence functions. I'm not in fact sure if one is thus committed.) It just seems to me that the arguments for taking cluelessness seriously as a problem are stronger. Still, I'm curious what you think the best arguments for credence functions being sharp are, or where I can read about them.
[I know I'm late to the party but...]
I'm certainly not an expert here, and I think my thinking is somewhat unclear, and my explanation of it likely will be too. But I share the sense that Knightian uncertainty can't be rational. Or more specifically, I have a sense that in these sorts of discussions, a lot of the work is being done by imprecise terms that imply a sort of crisp, black-and-white distinction between something we could call "regular" uncertainty and something we could call "extreme"/"radical"/"unquantifiable" uncertainty, without this distinction being properly made explicit or defended.
For example, in Hilary Greaves's paper on cluelessness (note: similar thoughts from me would apply to Mogensen's paper, though explained differently), she discusses cases of "simple cluelessness" and then argues they're not really a problem, because in such cases the "unforeseeable effects" cancel out in expectation, even if not in reality. E.g.,
Greaves is arguing that we can therefore focus on whether the "foreseeable effects" are positive or negative in expectation, just as our intuitions would suggest.
I agree with the conclusion, but I think the way it's juxtaposed with "complex cluelessness" (which she does suggest may be a cause for concern) highlights the sort of unwarranted (and implicit) sharp distinctions between "types" of uncertainty which I think are being made.
The three key criteria Greaves proposes for a case to involve complex cluelessness are:
I think all of that actually applies to the old lady case, just very speculatively. One reason to think CC1 is that the old lady and/or anyone witnessing your kind act and/or anyone who's told about it could see altruism, kindness, community spirit, etc. as more of the norm than they previously did, and be inspired to act similarly themselves. When they act similarly themselves, this further spreads that norm. We could tell a story about how that ripples out further and further and creates huge amount of additional value over time.
Importantly, there isn't a "precise counterpart, precisely as plausible as the original", for this story. That'd have to be something like people seeing this act therefore thinking unkindness, bullying, etc. are more the norm that they previously thought they were, which is clearly less plausible.
One reason to think CC2 for the old lady case could jump off from that story; maybe your actions sparks ripples of kindness, altruism, etc., which leads to more people donating to GiveWell type charities, which (perhaps) leads to increased population (via reduced mortality), which (perhaps) leads to increased x-risk (e.g., via climate change or more rapid technological development), which eventually causes huge amounts of disvalue.
I'd argue you again can't tell a precise counterpart story that's precisely as plausible as this, and that's for reasons very similar to those covered in both Greaves and Mogensen's paper - there are separate lines of evidence and argument for GiveWell type charities leading to increased population vs them leading to decreased population, and for increased population increasing vs decreasing x-risk. (And again, it seems less plausible that witnessing your good deed would make people less likely to donate to GiveWell charities than more likely - or at least, a decrease would occur via different mechanisms than an increase, and therefore not be a "precise counterpart" story.)
I think both of these "stories" I've told are extremely unlikely, and for practical purposes aren't worth bearing in mind. But they do seem to me to meet the criteria in CC1 and CC2. And it doesn't seem to me there's a fundamental difference between their plausibility and worthiness-of-attention and that of the possibility for donations to AMF to increase vs decrease x-risk (or other claimed cases of [complex] cluelessness). I think it's just a difference of degree - perhaps a large difference in degree, but degree nonetheless. I can't see how we could draw some clear line somewhere to decide what uncertainties can just be dealt with in normal ways and which uncertainties make us count as clueless and thus unable to use regular expected value reasoning.
(And as for CC3, I'd say it's at least slightly "unclear" how to weigh up these reasons against one another.)
I think my thoughts here are essentially a criticism of the idea of a sharp, fundamental distinction between "risk" and "Knightian uncertainty"*, rather than of Greaves and Mogensen's papers as a whole. That is, if we did accept as a premise that distinction, I think that most of what Greaves and Mogensen say seems like it may well follow. (I also found their papers interesting, and acknowledge that I'm very much not an expert here and all of this could be unfounded, so for all these reasons this isn't necessarily a case of me viewing their papers negatively.)
*I do think that those can be useful concepts for heuristic-type, practical purposes. I think we should probably act differently when are credences are massively less well-founded than usual, and probably be suspicious of traditional expected value reasoning then. But I think that's because of various flaws with how humans think (e.g., overconfidence in inside-view predictions), not because different rules fundamentally should apply to fundamentally different types of uncertainty.
re: your lady example: as far as I know, the recent papers e.g. here provide the following example: (1) either you help the old lady on a Monday or on a Tuesday (you must and can do exactly one of the two options). In this case, your examples for CC1 and CC2 don't hold. One might argue that the previous example was maybe just a mistake and I find it very hard to come up with CC1 and CC2 for (1) if (supposedly) you don't know anything about Mondays or Tuesdays.
Interesting. I wonder if the switch to that example was because they had a similar thought to mine, or read that comment.
But I think I can make a similar point with the Monday vs Tuesday example. I also predict I could make a similar point with respect to any example I'm given.
This is because I do know things about Mondays and Tuesdays in general, as well as about other variables. If the papers argue we're meant to artificially suppose we know literally nothing at all about a given variable, that seems weird or question-begging, and irrelevant to actual decision-making. (Note that I haven't read the recent paper you link to.)
I could probably come up with several stories for the Monday vs Tuesday example, but my first thought is to make it connect to my prior stories so it can reuse most of the reasoning from there, and to do that via social media. Above, I wrote:
This says people tend to use social media more on Tuesday and Wednesday than on Monday. I think we therefore have some reason to believe that, if I help an old lady cross the road on Tuesday rather than Monday, it's slightly more likely that someone will post about that on social media, and/or use social media in a slightly more altruistic, kind, community-spirit-y way than they otherwise would've. (Because me doing this on Tuesday means they're slightly more likely to be on social media while this kind deed is relatively fresh in their minds.) This could then further spread those norms (compared to how much they'd be spread if we helped on Monday), and we could tell a story about how that ripples out further etc.
Above, I also wrote:
I would now say exactly the same thing is true for the Monday vs Tuesday example, given my above argument for why the norms might be spread more if we help on Tuesday rather than Monday.
(We could also probably come up with stories related to amounts of traffic on Monday vs Tuesday - e.g., the old lady may be likelier to die if un-helped on one day, or more people may be delayed. Or related to people tending to be a little happier or sadder or Monday. Or related to what we ourselves predict we'll do with our time on Monday or Tuesday, which we probably would know about. Or many other things)
As before:
Sorry, I don't have the time to comment in-depth. However, I think if one agrees with cluelessness, then you don't offer an objection. You might even extend their worries by saying that "almost everything has "asymmetric uncertainty"". I would be interested in your extension of your last sentence. " They are extremely unlikely and thus not worth bearing mind". Why is this true?
When I said "I think both of these "stories" I've told are extremely unlikely, and for practical purposes aren't worth bearing in mind", the bolded bit meant that I think a person will tend to better achieve their goals (including altruistic ones) if they don't devote explicit attention to such (extremely unlikely) "stories" when making decisions. The reason is essentially that one could generate huge numbers of such stories for basically every decision. If one tried to explicitly think through and weigh up all such stories in all such decision situations, one would probably become paralysed.
So I think the expected value of making decisions before and without thinking through such stories is higher than the expected value of trying to think through such stories before making decisions.
In other words, the value of information one would be expected to get from spending extra time thinking through such stories is probably usually lower than the opportunity cost of gaining that information (e.g., what one could've done with that time otherwise).
Disclaimer: Written on low sleep, and again reporting only independent impressions (i.e., what I'd believe before updating on the fact that various smart people don't share my views on this). I also shared related thoughts in this comment thread.
I agree that one way someone could respond to my points is indeed by saying that everything/almost everything involves complex cluelessness, rather than that complex cluelessness isn't a useful concept.
But if Greaves introduces complex cluelessness by juxtaposing it with simple cluelessness, yet the examples of simple cluelessness actually meet their definition of complex cluelessness (which I think I've shown), I think this provides reason to pause and re-evaluate the claims.
And then I think we might notice that Greaves suggests a sharp distinction between simple and complex cluelessness. And also that she (if I recall correctly) arguably suggests homogeneity within each type of cluelessness - i.e., suggesting all cases of simple cluelessness can be dealt with by just ignoring the possible flow-through effects that seem symmetrical, while we should search for a type of approach to handle all cases of complex cluelessness. (But this latter point is probably debatable.)
And we might also notice that the term "cluelessness" seems to suggest we know literally nothing about how to compare the outcomes. Whereas I've argued that in all cases we'll have some information relevant to that, and the various bits of information will vary in their importance and degree of uncertainty.
So altogether, it would just seem more natural to me to say:
I do think there are many important questions to be investigated with regards to how best to make decisions under conditions of extreme uncertainty, and that this becomes especially relevant for people who want to have a positive impact on the long-term future. But it doesn't seem to me that the idea of complex cluelessness is necessary or useful in posing or investigating those questions.
Also, when reading Greaves and Mogensen's papers, I was reminded of the ideas of cluster thinking (also here) and model combination. I could be drawing faulty analogies, but it seemed like those ideas could be ways to capture, in a form that can actually be readily worked with, the following idea (from Greaves; the same basic concept is also used in Mogensen):
That is, we can consider each probability function in the agent's representor as one model, and then either qualitatively use Holden's idea of cluster thinking, or get a weighted combination of those models. Then we'd actually have an answer, rather than just indifference.
This seems like potentially "the best of both worlds"; i.e., a way to capture both of the following intuitively appealing ideas:
Hello,
Here is a good paper on this - https://www.princeton.edu/~adame/papers/sharp/elga-subjective-probabilities-should-be-sharp.pdf
Can't the sequence proposal be fixed by conditioning on the past and only considering future sequences of actions? Committing to rejecting both bets A and B is rationally impermissible if you will be offered both since it's worse than accepting both, but after your decision on A, regardless of whether you accepted or rejected, then it could be that both accepting and rejecting B are permissible at the same time. The fact that something was my past action shouldn't matter or prevent me from completing some particular sequence of actions that includes past actions, only my future prospects and future actions matter.
I think this makes sense for sharp probabilities, too: suppose you assign some sharp probability p≤25 to H being true, and have already rejected A, even though it had positive expected value (so this decision was irrational at the time). Then, since the expected value of B is ≤0, it's permissible to reject B, and even required if the inequality is strict. You may be rationally required to complete a sequence of actions which was irrational before you started.
You can also apply Mogensen's maximality rule to sequences. Given some set of plausible probability distributions, if one sequence of actions θ is better in expectation than another ϕ under at least one distribution, and not worse under any other distribution, then θ≻ϕ. If neither strict inequality holds between the two options and these are the only two options, then both are permissible. (We sacrifice the independence of irrelevant alternatives, since a third option could dominate one but not the other, only ruling out the dominated one.)
How important do you think non-sharp credence functions are to arguments for cluelessness being important? If you generally reject Knightian uncertainty and quantify all possibilities with probabilities, how diminished is the case for problematic cluelessness?
(Or am I just misunderstanding the words here?)
My belief that cluelessness is important is fairly independent of any specific philosophical/technical account of cluelessness. In particular, I don't think me changing my mind on whether credence functions have to be sharp would significantly change my views on the importance of cluelessness.
In this comment I've explained in more detail what I think about the relationship between the basic idea and specific philosophical theories trying to describe it.
(FWIW, I don't feel like I have a well-informed view on whether credence functions have to be sharp. If anything, I have a weak intuition that it's a bit more likely than not that I'd conclude they have to be if I spent more time looking into the question.)
Thanks, looking forward to reading this. Here's an archived version.
Cluelessness deserves more attention in EA, especially from the longtermist contingent.
Haven't read the full paper, but I'm recording some brief thoughts on cluelessness here for my own records. In a clueless world, the value of having an active EA-style movement that is at least partly longtermist may come from:
I'm sure these ideas aren't original, and (as with anything I write), I'd be glad to see links to places they've been expressed in a better way.