[Please read the OP before voting. Special voting rules apply.]
The replication initiative (the push to replicate the majority of scientific studies) is reasonably likely to do more harm than good. Most of the points raised by Jason Mitchell in The Emptiness of Failed Replications are correct.
Zvi, I agree with you that the CDC’s reasoning was pretty sketchy, but I think their actual recommendation is correct while everyone else (e.g. the UK) is wrong. I think the order should be something like:
Nursing homes → HCWs → 80+ → frontline essential workers → …
(Possibly switching the order of HCWs and 80+.)
The public analyses saying that we should start with the elderly are these two papers:
Notably, both papers don’t even consider vaccinating essential workers as a potential intervention. The only option categories are by age, comorbidities, and whether you’re a healthcare worker. The first paper only considers age and concludes unsurprisingly that if your only option is to order by age, you should start with the oldest. In the second paper, which includes HCWs as a category (modeling them as having higher susceptibility but not higher risk of transmitting to others), HCWs jump up on the queue to right after the 80+ age group (!!!). Since the only factor being considered is susceptibility, presumably many types of essential workers would also have a higher susceptibility and fall into the same group.
If we apply the Zvi cynical lens here, we can ask why these papers perform an analysis that suggests prioritizing healthcare workers but don’t bother to point out that the same analysis applies to 10% of the population (hint: there is less than 10% available vaccines and the authors are in the healthcare profession).
The actual problem with the original CDC recommendations was that essential workers is so broad a category that it encompasses lots of people who aren’t actually at increased risk (because their jobs don’t require much contact). The new recommendations revised this to focus on frontline essential workers, a more-focused category that is about half of all essential workers. This is a huge improvement but I think even the original recommendations are better than the UK approach of prioritizing only based on age.
Remember, we should focus on results. If the CDC is right while everyone else is wrong, even if the stated reasoning is bad, pressuring them to conform to everyone else’s worse approach is even worse.
I believe that most people hoping to do independent academic research vastly underestimate both the amount of prior work done in their field of interest, and the advantages of working with other very smart and knowledgeable people. Note that it isn’t just about working with other people, but with other very smart people. That is, there is a difference between “working at a university / research institute” and “working at a top university / research institute”. (For instance, if you want to do AI research in the U.S., you probably want to be at MIT, Princeton, Carnegie Mellon, Stanford, CalTech, or UC Berkeley. I don’t know about other countries.)
Unfortunately, my general impression is that most people on LessWrong are mostly unaware of the progress made in statistical machine learning (presumably the brand of AI that most LWers care about) and cognitive science in the last 20 years (I mention these two fields because I assume they are the most popular on LW, and also because I know the most about them). And I’m not talking about impressive-looking results that dodge around the real issues, I’m talking about fundamental progress towards resolving the key problems in artificial intelligence. Anyone planning to do AI research should probably at least understand these first, and what the remaining obstacles are.
You aren’t going to understand this without doing a lot of reading, and by the time you’ve done that reading, you’ll probably have identified a research group whose work clearly reflects your personal research goals. At this point it seems like the obvious next step is to apply to work with that group as a graduate student / post doc. This circumvents the problem of having to work on research you aren’t interested in. As for other annoyances, while teaching can potentially be a time-sink, the rest of “wasted” time seems to be about publishing your work; I really find it hard to justify not publishing your work, since (a) other people need to know about it, and (b) writing up your results formally oftentimes leads to a noticeably deeper understanding than otherwise. Of course, you can waste time trying to make your results look better than they are, but this certainly isn’t a requirement and has obvious ethical issues.
EDIT: There is the eventual problem that senior professors spend more and more of their time on administrative work / providing guidance to their lab, rather than doing research themselves. But this isn’t going to be an issue until you get tenure, which is, if you do a post-doc, something like 10-15 years out from starting graduate school.
Have you run this by a trusted bio expert? When I did this test (picking a bio person who I know personally, who I think of as open-minded and fairly smart), they thought that this vaccine is pretty unlikely to be effective and that the risks in this article may be understated (e.g. food grade is lower-quality than lab grade, and it’s not obvious that inhaling food is completely safe). I don’t know enough biology to evaluate their argument, beyond my respect for them.
I’d be curious if the author, or others who are considering trying this, have applied this test.
My (fairly uninformed) estimates would be: − 10% chance that the vaccine works in the abstract − 4% chance that it works for a given LW user − 3% chance that a given LW user has an adverse reaction −12% chance at least 1 LW user has an adverse reaction
Of course, from a selfish perspective, I am happy for others to try this. In the 10% of cases where it works I will be glad to have that information. I’m more worried that some might substantially overestimate the benefit and underestimate the risks, however.
Hmm, important as in “important to discuss”, or “important to hear about”?
My best guess based on talking to a smart open-minded biologist is that this vaccine probably doesn’t work, and that the author understates the risks involved. I’m interpreting the decision to frontpage as saying that you think I’m wrong with reasonably high confidence, but I’m not sure if I should interpret it that way.
For many smart people, academia is one of the highest-value careers they could pursue.
I think you are mistaken about the relative efficiency / inefficiency of scientific research. I believe that research is comparably efficient to much of industry, and that many of the things that look like inefficiencies are actually trading off small local gains for large global gains. I’ve come to this conclusion as the result of years of doing scientific research, where almost every promising idea I’ve come up with (including some that I thought quite clever) had already been explored by someone else. In fact, the typical case for when I was able to
make progress was when solving the problem required a combination of tools, each of which individually was relatively rare in the field.
For instance, my paper on stochastic verification required: (i) familiarity of sum-of-squares programming; (ii) the application of supermartingale
techniques from statistics; and (iii) the ability to produce relatively non-trivial convex relaxations of a difficult optimization
problem. In robotics, most people are familiar with convex optimization, and at least some are familiar with sum-of-squares programming and supermartingales. In fact, at least one other person had already published a fairly similar paper but had not formulated the problem in a way that was useful for systems of practical levels of complexity, probably because they had (i) and (ii) but lacked (iii).
Of course, it is true that your point #3 (“new tools open up promising new angles of attack”) often does lead to low-hanging fruit. In machine learning, often when a new tool is discovered there will be a sizable contingent of the community who devotes resources to exploring possible uses of that tool. However, I think the time-scale for this is shorter than you might imagine. I would guess that, in machine learning at least, most of the ``easy″ applications of a new tool have been exhausted within five years of its introduction. This is not to say that new applications don’t trickle in, but
they tend to either be esoteric or else require adapting the tool in some non-trivial way.
My impression is that MIRI believes that most of what drives what they perceive to be low-hanging fruit comes from #4 (“progress is only valuable to those with unusual views”). I think this is probably true, but not, as you probably believe, due to differing views about the intelligent explosion; I believe instead that MIRI’s differing views come from a misunderstanding of the context of their research.
For instance, MIRI repeatedly brings up Paul’s probabilistic metamathematics as an important piece of research progress produced by MIRI. I’ve mostly hedged when people ask me about this, because I do think
it is an interesting result, but I feel by now that it has been oversold by MIRI. One could argue that the result is relevant to one or more of logic, philosophy, or AI, but I’ll focus on the last one because it is presumably what MIRI cares about most. The standard argument in favor of this line of research is that Lob’s theorem implies that purely logical systems will have difficulty reasoning about their future behavior, and that therefore it is worth asking whether a probabilistic language of thought can overcome this obstacle. But I believe this is due to a failure to think sufficiently concretely about the problem. To give two examples: first, humans seem perfectly capable of making predictions about their future behavior despite this issue, and I have yet to see an argument for why AIs should be different. Secondly, we manage to routinely prove facts about the behavior of programs (this is the field of program analysis) despite the fact that in theory this should be “undecidable”. This is because the undecidability issues don’t really crop up in practice. If MIRI wanted to make the case that they will crop up for a human-level AI in important ways, they should at least mention and respond to the large literature pointing in the opposite direction, and explain why those tools or their successors would be insufficient.
A full response to the misguidedness of the Lob obstacle as a problem of interest probably necessitates its own post, but hopefully this serves to at least partially illustrate my point. Another example would be decision theory of modal agents. I also won’t take the time to treat this in detail, but will simply note that this work studies a form of decision theory that MIRI itself invented, and that no one else uses or studies. It should therefore perhaps be unsurprising that it is relatively easy to prove novel results about it. It would be the equivalent of an activity I did to amuse myself in high school, which was to invent new mathematical objects and then study their properties. But I think it would be mistaken to point to this as an example of inefficiency in research. (And yes, I do think the idea of formalizing decision theory in terms of code is an interesting one. I just don’t think you get to point to this as an example of why research is inefficient.)
I’m not sure this is the best description of my objections to this post, but I need to start work now and it seems best to avoid keeping this around as a draft for too long, so I’m going to go ahead and post it and wait for everyone to tear it to shreds.
I’m glad to see the large number amount of sincere discussion here, and thanks to Luke and Pei for doing this.
Although most people are not guilty of this, I would like to personally plea that people keep references towards Pei civil; insulting him or belittling his ideas without taking the time to genuinely respond to them (linking to a sequence post doesn’t count) will make future people less likely to want to hold such discussions, which will be bad for the community, whichever side of the argument you are on.
My small non-meta contribution to this thread: I suspect that some of Pei’s statements that seem wrong at face value are a result of him lacking the language to state things in a way that would satisfy most LWers. Can someone try to charitably translate his arguments into such terms? In particular, his stuff about goal adaptation are somewhat similar in spirit to jtaylor’s recent posts on learning utility functions.
I think an interesting related meme is “leadership as service”. This idea certainly existed in the Boy Scouts when I was in high school, and a related idea of “management as service” exists in at least some good tech companies.
I don’t personally like the “hero” narrative that much but I am highly ambitious and willing to do things even if no one else is doing them. Nevertheless, in fact, as a result of this, I often end up in what might seem like “sidekick roles”. I’ve oftentimes taken on logistical tasks even though it’s easy to argue that my comparative advantage is elsewhere. Why? Because if I don’t, then some important thing won’t get done, and that’s all that matters. This is what I think Eliezer means when he refers to “heroic responsibility”, and I think you among all people I’ve met exemplify this the most. So that’s one interesting observation.
Another observation in my personal experience is that it’s extremely difficult to take a “hero” role in more than one thing at once, simply because it’s too time-consuming. I have several causes that I contribute my time to, but in many cases my ability to do so is limited by someone else willing to take the lead and spearhead the project. But again interestingly, that person often ends up performing “sidekick-like” tasks while I am more free to focus on creating value directly. I think this again inverts the narrative: who do we call the “hero” and who do we call the “sidekick”? One person is taking the lead but only creating value indirectly by allowing a group of other people to be more productive. The other people are the ones that are directly creating value, but are working within a pre-existing system. I don’t know if you think that this relates to the ideas in your post or not, but I thought it was another interesting observation either way.
Zvi, I still think that your model of vaccination ordering is wrong, and that the best read of the data is that frontline essential workers should be very highly prioritized from a DALY / deaths averted perspective. I left this comment on the last thread that explains my reasoning in detail, looking at both of the published papers I’ve seen that model vaccine ordering: link. I’d be happy to elaborate on it but I haven’t yet seen anyone provide any disagreement.
More minor, but regarding rehab facilities, from a bureaucratic perspective they are “congregate living facilities” and in the same category as retirement homes. I don’t think New York is doing anything exceptional by having them high on the list, for instance California is doing the same thing if I understand correctly. We can of course argue over whether it’s good for them to be high on the list; I personally think of them as 20-person group houses and so feel reasonably good prioritizing them highly, though I’m not confident in that conclusion.
How do we know that calibration training will improve our calibration on long-term predictions? It would seem that we necessarily have little evidence about the efficacy of short-term calibration training on calibrating long-term predictions.