The back-and-forth (here and elsewhere) between Kaj & pjeby was an unusually good, rich, productive discussion, and it would be cool if the book could capture some of that. Not sure how feasible that is, given the sprawling nature of the discussion.
Unnamed
Karma: 6,546
This post seems to me to be misunderstanding a major piece of Paul’s “sluggish updating” post, and clashing with Paul’s post in ways that aren’t explicit.
The core of Paul’s post, as I understood it, is that incentive landscapes often reward people for changing their stated views too gradually in response to new arguments/evidence, and Paul thinks he has often observed this behavioral pattern which he called “sluggish updating.” Paul illustrated this incentive landscape through a story involving Alice and Bob, where Bob is thinking through his optimal strategy, since that’s a convenient way to describe incentive landscapes. But that kind of intentional strategic thinking isn’t how the incentives typically manifest themselves in behavior, in Paul’s view (e.g., “I expect this to result in unconscious bias rather than conscious misrepresentation. I suspect this incentive significantly distorts the beliefs of many reasonable people on important questions”). This post by Zvi misunderstands this as Paul describing the processes that go on inside the heads of actual Bobs. This loses track of the important distinction (which is the subject of multiple other LW Review nominees) between the rewards that shape an agent’s behavior and the agent’s intentions. It also sweeps much of the disagreement between Paul & Zvi’s posts under the rug.
A few related ways the views in the two posts clash:This post by Zvi focuses on dishonesty, while Paul suggests that unconsciously distorted beliefs are the typical case. This could be because Zvi disagrees with Paul and thinks that dishonesty is the typical case. Or it could be that Zvi is using the word “dishonest” broadly—he mostly agrees with Paul about what happens in people’s heads, but applies the “dishonesty” frame in places where Paul wouldn’t. Or maybe Zvi is just choosing to focus on the dishonest subset of cases. Or some combination of these.
Zvi focuses on cases where Bob is going to the extreme in following these incentives, optimizing heavily for it and propagating it into his thinking. “This is a world where all one cares about is how one is evaluated, and lying and deceiving others is free as long as you’re not caught.” “Bob’s optimal strategy is full anti-epistemology.” Paul seems especially interested in cases where pretty reasonable people (with some pretty good features in their epistemics, motivations, and incentives) still sometimes succumb to these incentives for sluggishness. Again, it’s unclear how much of this is due to Zvi & Paul having different beliefs about the typical case and how much is about choosing to focus on different subsets of cases (or which cases to treats as central for model-building).Paul’s post is written from a perspective of ‘Good epistemics don’t happen by default’, where thinking well as an individual involves noticing places where your mental processes haven’t been aimed towards accurate beliefs and trying to do better, and social epistemics are an extension of that at the group level. Zvi’s post is written from a perspective of ‘catching cheaters’, where good social epistemics is about noticing ways that people are something-like-lying to you, and trying to stop that from happening.
Zvi treats Bob as an adversary. Paul treats him as a potential ally (or as a state that you or I or anyone could find oneself in), and mentions “gaining awareness” of the sluggishness as one way for an individual to counter it.
Related to all of this, the terminology clashes (as I mentioned in a comment). I’d like to say a simple sentence like “Paul sees [?sluggishness?] as mainly due to [?unconscious processes?], Zvi as mainly due to [?dishonest update reporting?]” but I’m not sure what terms go in the blanks.
The “fire Bob” recommendation depends a lot on how you’re looking at the problem space / which part of the problem space you’re looking at. If it’s just a recommendation for a narrow set of cases then I think it wouldn’t apply to most of the cases that Paul was talking about in his “Observations in the wild”, but if it’s meant to apply more widely then that could get messy in ways that interact with the clashes I’ve described.
The other proposed solutions seem less central to these two posts, and to the clash between Paul & Zvi’s perspectives.
I think there is something interesting in the contrast between Paul & Zvi’s perspectives, but this post didn’t work as a way to shine light on that contrast. It focuses on a different part of the problem space, while bringing in bits from Paul’s post in ways that make it seem like it’s engaging with Paul’s perspective more than it actually does and make it confusing to look at both perspectives side by side.
Sounds like the thing that is typically called “regret aversion”.
Crunching some numbers in a copy of the spreadsheet… Zvi’s predictions are better than the naive model of assuming next week’s numbers will be the same as this week’s numbers.
Biggest improvement over the null model for predicting deaths (mean squared error is 47% as big), smallest improvement for positive test % (MSE 80% as big), in between for number of tests (MSE 67% as big).
Although if I instead look at the predicted weekly change and compare it to the actual change that week, all three sets of predictions are roughly equally accurate with correlations (predicted change vs. actual change) between .52 and .58.
When I read this bit:
Only 37% of all distributed doses have been given
I wondered how that would look translated into a delay. The number of doses given through January 13 equals the number of doses that had been distributed __ days earlier.
I see from the graph titled “The US COVID-19 Vaccine Shortfall” (and introduced with “We can start with how it’s gone so far:”) that the answer is about 17 days.
This seems like a more natural framing—it matches the process of why many doses haven’t been given yet, and it seems likely to be more stable as we project the curves forward over many weeks (and less dependent on the shape of the ‘doses distributed’ curve).
So now I’m wondering if the delay (now 17 days) is likely to get smaller over time, or larger, or stay about the same.
Seems like the terminology is still not settled well.
There’s a general thing which can be divided into two more specific things.
General Thing: The information points to 50%, the incentive landscape points to 70%, Bob says “70%”.
Specific Thing 1: The information points to 50%, the incentive landscape points to 70%, Bob believes 50% and says “70%”.
Specific Thing 2: The information points to 50%, the incentive landscape points to 70%, Bob believes and says “70%”.
There are three Things and just two names, so the terminology is at least incomplete.
“Dishonest update reporting” sounds like the name of Specific Thing 1.
In Paul’s post “sluggish updating” referred to the General Thing, but Dagon’s argument here is that “sluggish updating” should only refer to Specific Thing 2. So there’s ambiguity.
It seems most important to have a good name for the General Thing. And that’s maybe the one that’s nameless? Perhaps “sluggish update reporting”, which can happen either because the updating is sluggish or because the reporting is sluggish/dishonest. Or “sluggish social updating”? Or something related to lightness? Or maybe “sluggish updating” is ok despite Dagon’s concerns (e.g. a meteorologist updating their forecast could refer to changes that they make to the forecast that they present to the world).
Here is a thing I wrote 10 years ago assessing an N-back study. That’s a easy-for-me-to-remember example, where I also remember that the writeup comes pretty close to reflecting how I was thinking through things as I was looking at the paper.
Well, the sun being the only object in our solar system that emits light is evidence for it being at the center.
It seems likely that there’s something special about whichever body is in the center of the solar system. A lot of astronomers thought the Earth was special for being made of rock & water, and that this was related to the Earth being at the center, but they just conjectured that Mars & Venus & the other planets were made of something else. Whereas Kepler had much more direct observations about the sun’s unique luminosity.
Aristarchus had a heliocentric model of the solar system in Ancient Greece, apparently motivated in large part by the fact that the sun was the largest object in the solar system.
In hindsight, we know that both luminosity and size relative to neighbors are both highly correlated with being at the center of a solar system, with Aristarchus’s size thing having a tighter causal relationship with centrality.
Those public health official examples seem unrelated to tip #59 (“Those who generate anxiety in you and promise that they have the solution are grifters.”).
I took hermanc1 to be pointing to how, in Feb-Mar 2020, the people who were saying scary sounding stuff (like using the word “pandemic”) and proposing things to do about it were the ones who had insights and were telling it straight. Meanwhile many other people were calling those people out for “fearmongering” or spinning things to downplay the risk in order to prevent panic.
There are grifters who try to generate anxiety so they can sell you something. And also the world contains problems, and noticing problems can induce anxiety, and searching for & sharing (partial) solutions to problems is good. Maybe a sophisticated way of following tip #59 can distinguish between those, but the naive way of doing it can run into trouble and fail to see the smoke.
Back in March, there was a lot of concern that uncontrolled spread would overwhelm the medical system and some hope that delay would improve the standard of care. Do we have good estimates now of those two effects? They could influence IFR estimates by a fair amount.
Also, my understanding is that the number of infections could’ve shot way past herd immunity levels. Herd immunity is just the point at which the number of active infections starts declining rather than increasing, and if there are lots of active infections at that time then they can spread to much of the remaining people before dwindling.
I’ve had similar thoughts; the working title that I jotted down at some point is “Two Aspects of Morality: Do-Gooding and Coordination.” A quick summary of those thoughts:
Do-gooding is about seeing some worlds as better than others, and steering towards the better ones. Consequentialism, basically. A widely held view is that what makes some worlds better than others is how good they are for the beings in those worlds, and so people often contrast do-gooding with selfishness because do-gooding requires recognizing that the world is full of moral patients.
Coordination is about recognizing that the world is full of other agents, who are trying to steer towards (at least somewhat) different worlds. It’s about finding ways to arrange the efforts of many agents so that they add up to more than the sum of their parts, rather than less. In other words, try for: many agents combine their efforts to get to worlds that are better (according to each agent) than the world that that agent would have reached without working together. And try to avoid: agents stepping on each other’s toes, devoting lots of their efforts to undoing what other agents have done, or otherwise undermining each other’s efforts. Related: game theory, Moloch, decision theory, contractualism.
These both seem like aspects of morality because:
“moral emotions”, “moral intuitions”, and other places where people use words like “moral” arise from both sorts of situations
both aspects involve some deep structure related to being an agent in the world, neither seems like just messy implementation details for the other
a person who is trying to cultivate virtues or become a more effective agent will work on both
Related to section I: Dunning, Meyerowitz,& Holzberg (1989) Ambiguity and self-evaluation: The role of idiosyncratic trait definitions in self-serving assessments of ability. From the abstract:
When people are asked to compare their abilities to those of their peers, they predominantly provide self-serving assessments that appear objectively indefensible. This article proposes that such assessments occur because the meaning of most characteristics is ambiguous, which allows people to use self-serving trait definitions when providing self-evaluations. Studies 1 and 2 revealed that people provide self-serving assessments to the extent that the trait is ambiguous, that is, to the extent that it can describe a wide variety of behaviors.
It seems clear that we want politicians to honestly talk about what they’re intending to do with the policies that they’re actively trying to change (especially if they have a reasonable chance of enacting new policies before the next election). That’s how voters can know what they’re getting.
It’s less obvious how this should apply to their views on things which aren’t going to be enacted into policy. Three lines of thinking that point in the direction of maybe it’s good for politicians to keep quiet about (many of) their unpopular views:
It can be hard for listeners to tell how likely the policy is to be enacted, or how actively the politician will try to make it happen. I guess it’s hard to fit into 5 words? e.g. I saw a list of politicians’ “broken promises” on one of the fact checking sites, which was full of examples where the politician said they were in favor of something and then it didn’t get enacted, and the fact checkers deemed that sufficient to count it as a broken promise. This can lead to voters putting too little weight on the things that they’re actually electing the politician to do, e.g. local politics seems less functional if local politicians focus on talking about their views on national issues that they have no control over.
Another issue is that it’s cheap talk. The incentive structure / feedback loops seem terrible for politicians talking about things unrelated to the policies they’re enacting or blocking. Might be more functional to have a political system where politicians mostly talk about things that are more closely related to their actions, so that their words have meaning that voters can see.
Also, you can think of politicians’ speech as attempted persuasion. You could think of voters as picking a person to go around advocating for the voters’ hard-to-enact views (as well as to implement policies for the voters’ feasible-to-enact views). So it seems like it could be reasonable for voters to say “I think X is bad, so I’m not going to vote for you if you go around advocating for X”, and for a politician who personally favors X but doesn’t talk about it to be successfully representing those voters.
You can think of growth mindset as a deidentification, basically identical to that example of Anna the student except done by Anna about herself rather than by her teacher. “Yet” is a wedge that gets you to separate your concept of you-so-far from your concept of future you. “I’m bad at X” sneaks in an equivocation to imply “now and always.”
I notice that many of these examples involve something like vice signalling—the person is destroying value in order to demonstrate that they have a quality which I (and most LWers) consider to be undesirable. It seems bad for the middle manager, politician, and start-up founder to aim for the shallow thing that they’re prioritizing. And then they take the extra step of destroying something I do value in order to accentuate that. It’s a combination that feels real icky.
The romantic dinner and the handmade gift examples don’t have that feature. And those two cases feel more ambiguous—I can imagine versions of these where it seems good that the person is doing these things, and versions where it seems bad. I can picture a friend telling me “I took my partner out for their birthday to a restaurant that I don’t really care for, but they just adore” and it being a heartwarming story, where it seems like something good is happening for my friend and their relationship.
Katja’s recent post on Opposite Attractions points to one thing that seems good about taking your spouse to a restaurant that only they love—your spouse’s life is full of things that you both like, and perhaps starved of certain styles of things that they like and you don’t, and they could be getting something out of drawing from that latter category even if there’s some sense in which they don’t like it any more than a thing in the “youboth like it” category. And there’s something good about them getting some of those things within the relationship, of having the ground that the relationship covers not be limited to the intersection of “the things you like” and “the things your spouse likes”—your relationship mostly takes advantage of that part of the territory but sometimes it’s good to explore other parts of it together. And I could imagine you bringing an attitude to the meal where you’re tuned in to your spouse’s experience, trying to take pleasure in how much they enjoy the meal, rather than being focused on your own food. And (this is the part where paying a cost to resolve motive ambiguity comes in directly) going to a restaurant that they love and you don’t like seems like it can help set the context for this kind of thing—putting the information in common knowledge between you two that this is a special occasion, and what sort of special occasion it’s trying to be. It seems harder to hit some of these notes in a context where both people love the food.
(There are also versions the one-sided romantic dinner which seem worse, and good relationships where this version doesn’t fit or isn’t necessary.)
Can you tell your spellchecker that they’re not words?
If you’re just doing this occasionally without recordkeeping, then it seems convenient to have the game result in “winners” rather than a more fine-grained score. But it could be fine to sometimes have multiple winners, or zero winners. Here’s a simple protocol that does that:
The person who asks the question also defines what counts as “winning”. e.g. “What’s the value of such-and-such? Can anybody get it within 10%?” Then everyone guesses simultaneously, and all the people whose guesses are within 10% of the true value are “winners”.
(“Simultaneous” guessing can mean that first everyone comes up with their guess in their head, and then they take turns saying them out loud while on the honor system to not change their guess.)
Slightly more complicated, the asker could propose 2 standards of winning. “When did X happen? Grand prize if you guess the exact year, honorable mention if you get it within 5 years.” Then if anyone guesses the exact year they’re the big winner(s) and the people who get it within 5 years get the lesser glow of “honorable mention”. And if no one guesses the exact year then the people who get it within 5 years feel more like winners.
If you continue farther in this direction you could get to one of Ericf’s proposals. I think my version has lower barriers to entry, while Ericf’s version could work better among people who use it regularly.
Note that Trump got around 63M votes in 2016, and around 71M in 2020, whereas Democrats got 66M and 75M respectively.
The 2020 results are 81M-74M with some votes still left to count. 75M-71M might have been the margin a few weeks ago when there were still a bunch more not-yet-counted votes.
It seems like the core thing that this post is doing is treating the concept of “rule” as fundamental.
If you have a general rule plus some exceptions, then obviously that “general rule” isn’t the real process that is determining the results. And noticing that (obvious once you look at it) fact can be a useful insight/reframing.
The core claim that this post is putting forward, IMO, is that you should think of that “real process” as being a rule, and aim to give it the virtues of good rules such as being simple, explicit, stable, and legitimate (having legible justifications).
An alternative approach is to step outside of the “rules” framework and get in touch with what the rule is for—what preferences/values/strategy/patterns/structures/relationships/etc. it serves. Once you’re in touch with that purpose, then you can think about both the current case, and what will become of the “general rule”, in that light. This could end up with an explicitly reformulated rule, or not.
It seems like treating the “real process” as a rule is more fitting in some cases than others, a better fit for some people’s style of thinking than for other people’s, and also something that a person could choose to aim for more or less.
I think I’d find it easier to think through this topic if there was a long, diverse list of brief examples.