Wiki Contributions

Comments

Re-posting a link here, on the off-chance it's of interest despite its length. ESRogs and I also had a parallel discussion on the EA Forum, which led me to write up this unjustifiably lengthy doc partly in response to that discussion and partly in response to the above comment.

Just wanted to say I really appreciate you taking the time to write up such a long, clear, and thoughtful response!

(If I have a bit of time and/or need to procrastinate anything in the near future, I may write up a few further thoughts under this comment.)

It sounds as though you're expecting anti-realists about normativity to tell you some arguments that will genuinely make you feel (close to) indifferent about whether to use Bayesianism, or whether to use induction.

Hm, this actually isn't an expectation I have. When I talk about "realists" and "anti-realists," in this post, I'm thinking of groups of people with different beliefs (rather than groups of people with different feelings). I don't think of anti-realism as having any strong link to feelings of indifference about behavior. For example: I certainly expect most anti-realist philosophers to have strong preferences against putting their hands on hot stoves (and don't see anything inconsistent in this).

But I have yet to see how that's a useful concept to introduce. I just don't get it.

I guess I don't see it as a matter of usefulness. I have this concept that a lot of other people seem to have too: the concept of the choice I "should" make or that it would be "right" for me to make. Although pretty much everyone uses these words, not everyone reports having the same concept. Nonetheless, at least I do have the concept. And, insofar as there is any such thing as the "right thing," I care a lot about doing it.

We can ask the question: "Why should people care about doing what they 'should' do?" I think the natural response to this question, though, is just sort of to evoke a tautology. People should care about doing what they should do, because they should do what they should do.

To put my "realist hat" firmly on for a second: I don't think, for example, that someone happily abusing their partner would in any way find it "useful" to believe that abuse is wrong. But I do think they should believe that abuse is wrong, and take this fact into account when deciding how to act, because abuse is wrong.

I'm unfortunately not sure, though, if I have anything much deeper or more compelling than that to say in response to the question.


Another (significantly more rambling and possibly redundant) thought on "usefulness":

One of the main things I'm trying to say in the post -- although, in hindsight, I'm unsure if I communicated it well -- is that there are a lot of debates that I personally have trouble interpretting as both non-trivial and truth-oriented if I assume that the debaters aren't employing irreducably normative concepts. A lot of debates about decision theory have this property for me.

I understand how it's possible for realists to have a substantive factual disagreement about the Newcomb scenario, for example, because they're talking about something above-and-beyond the traditional physical facts of the case (which are basically just laid out in the problem specification). But if we assume that there's nothing above-and-beyond the traditional physical facts, then I don't see what there's left for anyone to have a substantive factual disagree about.

If we want to ask "Which amount of money is the agent most likely to receive, if we condition on the information that it will one-box?", then it seems to me that pretty much everyone agrees that “one million dollars” is the answer. If we want to ask "Would the agent get more money in a counterfactual world where it instead two-boxes, but all other features of the world at that time (including the contents of the boxes) are held fixed?", then it seems to me that pretty much everyone agrees the answer is “yes.” If we want to ask “Would the agent get more money in a counterfactual world where it was born as a two-boxer, but all other features of the world at the time of its birth were held fixed?", then it seems to me that pretty much everyone agrees the answer is “no.” So I don't understand what the open question could be. People may of course have different feelings about one-boxing and about two-boxing, in the same way that people have different feelings about (e.g.) playing tennis and playing soccer, but that's not a matter of factual/substantive disagreement.

So this is sort of one way in which irreducably normative concepts can be "useful": they can, I think, allow us to make sense of and justify certain debates that many people are strongly inclined to have and certain questions that many people are strongly inclined to ask.

But the above line line of thought of course isn't, at least in any direct way, an argument for realism actually being true. Even if the line of thought is sound, then it's still entirely totally possible that these debates and questions just actually aren't non-trivial and truth-oriented. Furthermore, the line of thought could also just not be sound. It's totally possible that the debates/questions are non-trivial and truth-oriented, without evoking irreducably normative concepts, and I'm just a confused outside observer not getting what's going on. Tonally, one thing I regret about the way I wrote this post is that I think it comes across as overly skeptical of this possibility.

Belatedly, is this a fair summary of your critique?

When someone thinks about another person (e.g. to predict whether they'll submit to blackmail), the act of thinking about the other person creates a sort of 'mental simulation' that has detailed concious experiences in its own right. So you never really know whether you're a flesh-and-blood person or a 'mental simulation' based on a flesh-and-blood person.

Now, suppose you seem to find yourself in a situation where you've been blackmailed. In this context, it's reasonable to wonder whether you're actually a flesh-and-blood person who's been blackmailed -- or merely a 'mental simulation' that exists in the mind of a potential blackmailer. If you're a mental simulation, and you care about the flesh-and-blood person you're based on, then you have reason to resist blackmail. The reason is that the decision you take as a simulation will determine the blackmailer's prediction about how the flesh-and-blood person will behave. If you resist blackmail, then the blackmailer will predict the flesh-and-blood person will refuse blackmail and therefore decide not to blackmail them.

If this is roughly in the right ballpark, then I would have a couple responses:

  1. I disagree that the act of thinking about a person will tend to create a mental simulation that has detailed concious experiences in its own right. This seems like a surprising position that goes against the grain of conventional neuroscience and views on the philosophy of conciousness. As a simple illustrative case, suppose that Omega makes a prediction about Person A purely on the basis of their body language. Surely thinking "This guys looks really nervous, he's probably worried he'll be seen as the sort of guy who'll submit to blackmail -- because he is" doesn't require bringing a whole new conciousness into existance.

  2. Suppose that when a blackmailer predicts someone's behavior, they do actually create a concious mental simulation. Suppose you don't know whether you're this kind of simulation or the associated flesh-and-blood person, but you care about what happens to the flesh-and-blood person in either case. Then, depending on certain parameter values, CDT does actually say you should resist blackmail. This is because there is some chance that you will cause the flesh-and-blood person to avoid being blackmailed. So CDT gives the response you want in this case.

Overall, I don't think this line of argument really damages CDT. It seems to be based on a claim about conciousness that I think probably wrong. But even if the claim is right, all this implies is that CDT recommends a different action than one would otherwise have thought.

(If my summary is roughly in the right ballpark, then I also think it's totally reasonable for academic decision theorists to read the FDT paper to fail to know that a non-mainstream neuroscience/philosophy-of-conciousness view is being assumed and provides the main justification for FDT. The paper really doesn't directly say anything about this. It seems wrong, then, to me to suggest that Schwarz only disagrees because he lacks the ability to see his own assumptions.)

[[EDIT: Oops, rereading your comment, seems like the summary is probably not fair. I didn't process this bit:

Yes, yes, if Omega used some method other than a simulation to make his prediction, the hypothetical you wouldn't have existed and wouldn't have had a perspective--but hey, that doesn't stop me from writing from their perspective, right? After all, real people write from the perspectives of unreal people all the time; that's just called writing fiction.

But now, reading the rest of the comment in light of this point, I don't think this reduces my qualms. The suggestion seems to be that, when seem to find yourself in the box room, you should in some cases be uncertain about whether or not you exist at all. And in these cases you should one box, because, if it turns out that you don't exist, then your decision to one box will (in some sense) cause a corresponding person who does exist to get more money. You also don't personally get less money by one boxing, because you don't get any money either way, because you don't exist.

Naively, this line of thought seems sketchy. You can have uncertainty about the substrate your mind is being run on or about the features of the external world -- e.g. you can be unsure whether or not you're a simulation -- but there doesn't seem to be room for uncertainty about whether or not you exist. "Cogito, ergo sum" and all that.

There is presumably some set of metaphysical/epistemological positions under which this line of reasoning makes sense, but, again, the paper really doesn't make any of these positions explicit or argue for them directly. I mainly think it's premature to explain the paper's faillure to persuade philosophers in terms of their rigidity or inability to question assumptions.]]

Hmm, I think focusing on a simpler case might be better for getting at the crux.

Suppose Alice says: "Eating meat is the most effective way to get protein. So if you want to get protein, you should eat meat."

And then Bob, an animal welfare person, responds: "You're wrong, people shouldn't eat meat no matter how much they care about getting protein."

If Alice doesn't mean for her second sentence to be totally redundant -- or if she is able to interpret Bob's response as an intelligible (if incorrect) statement of disagreement with her second sentence -- then that suggests her second sentence actually constitutes a substantively normative claim. Her second sentence isn't just repeating the same non-normative claim as the first one.

I definitely don't think that all "If you want X, do Y" claims are best understood as normative claims. It's possible that when people make claims of this form about Bayesianism, and other commonly discussed topics, they're not really saying anything normative. But a decent chunk of statements of this form do strike me as difficult to interpret in non-normative terms.

Okay, this seems like a crux of our disagreement. This statement seems pretty much equivalent to my statement #1 in almost all practical contexts. Can you point out how you think they differ?

This stuff is definitely a bit tricky to talk about, since people can use the word "should" in different ways. I think that sometimes when people say "You should do X if you want Y" they do basically just mean to say "If you do X you will receive Y." But it doesn't seem to me like this is always the case.

A couple examples:

1. "Bayesian updating has a certain asymptoptic convergence property, in the limit of infinite experience and infinite compute. So if you want to understand the world, you should be a Bayesian."

If the first and second sentence were meant to communicate the same thing, then the second would be totally vacuous given the first. Anyone who accepted the first sentence could not intelligibly disagree with or even really consider disagreeing with the second. But I don't think that people who say things like this typically mean for the second sentence to be vacuous or typically regard disagreement as unintelligible.

Suppose, for example, that I responded to this claim by saying something like: "I disagree. Since we only have finite lives, asymptoptic convergence properties don't have direct relevance. I think we should instead use a different 'risk averse' updating rule that, for agents with finite lives, more strongly reduces the likelihood of ending up with especially inaccurate beliefs about key features of the world."

The speaker might think I'm wrong. But if the speaker thinks that what I'm saying constitutes intelligible disagreement with their claim, then it seems like this means their claim is in fact a distinct normative one.

2. (To someone with no CS background) "If you want to understand the world, you should be a Bayesian."

If this sentence were meant to communicate the same thing as the claim about asymptotic convergence, then the speaker shouldn't expect the listener to understand what they're saying (even if the speaker has already explained what it means to be a Bayesian). Most people don't naturally understand or care at all about asymptotic convergence properties.

It's important to disentangle two claims:

  1. In general, if you have the goal of understanding the world, or any other goal that relies on doing so, being Bayesian will allow you to achieve it to a greater extent than any other approach (in the limit of infinite compute).

  2. Regardless of your goals, you should be Bayesian anyway.

Believing #2 commits you to normative realism as I understand the term, but believing #1 doesn't - #1 is simply an empirical claim about what types of cognition tend to do best towards a broad class of goals. I think that many rationalists would defend #1, and few would defend #2 - if you disagree, I'd be interested in seeing examples of the latter.

I don't necessarily think that #2 is a common belief. But I do have the impression that many people would at least endorse this equally normative claim: "If you have the goal of understanding the world, you should be a Bayesian."

In general -- at least in the context of the concepts/definitions in this post -- the inclusion of an "if" clause doesn't prevent a claim from being normative. So, for example, the claim "You should go to Spain if you want to go to Spain" isn't relevantly different from the claim "You should give money to charity if you have enough money to live comfortably."


Either way, I agree with Wei that distinguishing between moral normativity and epistemic normativity is crucial for fruitful discussions on this topic.

I agree there's an important distinction, but it doesn't necessarily seem that deep to me.

For example: We can define different "epistemic utility functions" that map {agent's credences; state of the world} to real values and then discuss theories like Bayesianism in the context of "epistemic decision theory," in relatively close analogy with traditional (practical) decision theory.

It seems like some theories -- e.g. certain theories that say we should have faith in the existance of God, or theories that say that we shouldn't take into account certain traits when forming impressions of people -- might also be classified as both moral and epistemological.

I left a sub-comment under Wei's comment (above) that hopefully unpacks this suggestion a bit

I think there's a distinction (although I'm not sure if I've talked explicitly about it before). Basically there's quite possibly more to what the "right" or "reasonable" action is than "what action that someone who tends to 'win' a lot over the course of their life would take?" because the latter isn't well defined. In a multiverse the same strategy/policy would lead to 100% winning in some worlds/branches and 100% losing in other worlds/branches, so you'd need some kind of "measure" to say who wins overall. But what the right measure is seems to be (or could beLW) a normative fact that can't be determined by just looking at or thinking "who tends to 'win' a lot'.

I agree with you on this and think it's a really important point. Another (possibly redundant) way of getting at a similar concern, without evoking MW:

Due to randomness/uncertainty, an agent that tries to maximize expected "winning" won't necessarily win compared to an agent that does something else. If I spend a dollar on a lottery ticket with a one-in-a-billion chance of netting me a billion-and-one "win points," then I'm taking the choice that maximizes expected winning but I'm also almost certain to lose. So we can't treat "the action that maximizes expected winning" as synonymous with "the action taken by an agent that wins."

We can try to patch up the issue here by defining "the action that I should take" as "the action that is consistent with the VNM axioms," but in fact either action in this case is consistent with the VNM axioms. The VNM axioms don't imply that an agent must maximize the expected desirability of outcomes. They just imply that an agent must maximize the expected value of some function. It is totally consistent with the axioms, for example, to be risk averse and instead maximize the expected square root of desirability. If we try to define "the action I should take" in this way, then, as another downside, the claim "your actions should be consistent with the VNM axioms" also becomes a completely empty tautology.

So it seems very hard to make non-vacuous and potentially true claims about decision theory without evoking some additional non-reducible notion of "reasonableness," "rationality," or what an actor "should" do. Assuming that normative anti-realism is true pretty much means assuming that there is no such notion or assuming that the notion doesn't actually map onto anything in reality. And I think anti-realist views of these sort are plausible (probably for roughly the same reasons Eliezer seems to). But I think that adopting these views would also leave us with very little to say about decision theory.

If there is anything that anyone should in fact do, then I would say that meets the standards of "realism."

Does "anyone" refer to any human, or any possible being?

Sorry, I should have been clearer. I mean to say: "If there exists at least one entity, such that the entity should do something, then that meets the standards of 'realism.'"

I understand "moral realism" as a claim that there is a sequence of clever words that would convince the superintelligent spider that reducing human suffering is a good thing.

I don't think I'm aware of anyone who identifies as a "moral realist" who believes this. At least, it's not part of a normal definition of "moral realism."

The term "moral realism" is used differently by different people, but typically it's either used roughly synonymously with "normative realism" (as I've defined it in this post) or to pick out a slightly more specific position: that normative realism is true and that people should do things besides just try to fulfill their own preferences.

Load More