Posts

Operationalizing Newcomb's Problem 2019-11-11T22:52:52.835Z · score: 36 (12 votes)
Is the World Getting Better? A brief summary of recent debate 2019-02-06T17:38:43.631Z · score: 35 (14 votes)

Comments

Comment by erickball on Transportation as a Constraint · 2020-04-06T15:18:56.086Z · score: 5 (3 votes) · LW · GW

If transportation isn't a problem then transporting water/sewer/energy over long distances would presumably also be cheap, so buildings could share infrastructure without being colocated, right?

Comment by erickball on An alarm bell for the next pandemic · 2020-04-06T14:22:32.919Z · score: 6 (4 votes) · LW · GW

It would be nice if we had somebody who specifically agreed to check sites like that regularly and then let the rest of us know if there's something worth being concerned about. That way we could avoid duplication of effort.

Comment by erickball on What will happen to supply chains in the era of COVID-19? · 2020-04-02T13:17:53.932Z · score: 3 (2 votes) · LW · GW

Are those government-owned warehouses of grain like the ones Kaj mentioned in Finland? The US used to have a strategic grain reserve, although it looks like at the time it was liquidated in 2008 it had only 915,000 tons of grain, which would feed the country for... not even a week.

Requiring businesses that sell non-perishable food to store a few months' supply sounds reasonable, but I'm curious if other places have used that approach successfully. I know groceries are a low-margin business to begin with.

Comment by erickball on What will happen to supply chains in the era of COVID-19? · 2020-03-31T18:36:21.862Z · score: 15 (5 votes) · LW · GW

Pushback against just-in-time supply chains sounds appealing, but how would it actually work? Is this something we can regulate, maybe make grocery stores go through stress tests like the ones we do for big banks? Somehow I have a hard time believing practices will change.

Comment by erickball on C19 Prediction Survey Thread · 2020-03-30T18:45:37.282Z · score: 6 (4 votes) · LW · GW

May 1st prediction:

US deaths: lognormal(11.4, 0.5) mode: 69k -- this is roughly equivalent, per capita, to what Italy is at currently (March 30th). Therefore it assumes the current growth will slow considerably. 15k is not even on the radar. I expect we will hit 3000 today, and 15k around April 7th.

US confirmed cases: lognormal(15, 0.7) mode: 2 million. This is more uncertain because it largely depends on how fast we ramp up testing.

Comment by erickball on Price Gouging and Speculative Costs · 2020-03-28T01:32:50.804Z · score: 2 (2 votes) · LW · GW

Ah, okay. I guess that answers my question: you don't think it's possible to mitigate shortages by stockpiling without losing money in expectation. I wish it were--surge production is a much more tenuous solution that depends on being able to foresee a disaster shortly before it occurs, and even then it might not be possible to produce enough to help significantly.

Comment by erickball on Price Gouging and Speculative Costs · 2020-03-27T18:28:39.986Z · score: 2 (2 votes) · LW · GW
You're also in a morally worse spot - you run the risk that you are creating or amplifying the shortage, not just predicting and smoothing the consumption of rapidly-value-changing products. If you aren't solving the problem given in the post (you could make more, but not at your normal prices), you're not as clearly on the side of good.

I don't follow. The idea would be that you aren't contributing to the normal supply, you just stockpile in case of future emergency. This increases production during normal times slightly, and then when there is a pandemic you would stop purchasing and start selling. This reduces the need for manufacturers to produce more at higher marginal cost. How could it create or amplify a shortage?

It's true you'd have no profit prior to the emergency (and large costs) but in theory that shouldn't matter as long as your investors diversify.

Comment by erickball on Price Gouging and Speculative Costs · 2020-03-27T02:29:59.924Z · score: 2 (2 votes) · LW · GW

Is there a way to do this if you're not already a manufacturer? With, say, N95 masks--could we buy up a bunch of regular ones when there's no shortage, do something that costs a little money and adds a tiny bit of value (maybe print cute pictures on them for doctors who wear them around little kids) and then mark them up 900%? I mean, I don't have money to buy them with, but if stockpiling really makes economic sense then I'd think we could get investors to front it.

How do price gouging laws apply to discounted products? If we almost always offer an 80% discount, can we stop offering that during an emergency?

ETA: to be clear, masks are just an example. They would not be a good choice in the future because after this pandemic governments will actually stockpile them. We'd need to figure out what there's likely to be a major shortage of in the NEXT global emergency.

Comment by erickball on Authorities and Amateurs · 2020-03-25T20:04:43.144Z · score: 1 (1 votes) · LW · GW

It said the fraction would be a bit lower for the US because local outbreaks could be dealt with by state-level lockdowns, but I didn't see a hard number. Still, intermittent lockdowns for 18 months seems much more achievable than a continuous lockdown. The Hammer article is definitely more optimistic than the Imperial paper, though it still doesn't quite imply "return to normal".

Since "do nothing" is not a real option, what will actually happen in the US (I'm moderately confident) is some degree of lockdown for several weeks to several months depending on how effective it is. The sooner we start it the better it will work. After that we will either: 1) give up, if the lockdown was ineffective and a large fraction of the country is infected (this is "flattening"), or 2) if the lockdown succeeded in reducing the number of cases substantially, we'll move into a period of intermittent and possibly localized lockdowns interspersed with trying to test and contact trace. The fraction of time we spend in intermittent lockdowns will depend on how effective the testing and tracing is.

Comment by erickball on The Solution is Inaction · 2020-03-25T17:31:15.985Z · score: 3 (2 votes) · LW · GW

I've always thought the premium charged by landlords for the services they provide (roughly, the difference between rent and mortgage interest + taxes + insurance) is pretty small, though I don't have numbers on hand. Besides maintenance and property management stuff, they take quite a bit of risk of fluctuations in property values, damage to the building, tenants stopping payment (in which case eviction typically takes quite a while), and liability for accidents and the like.

Comment by erickball on Authorities and Amateurs · 2020-03-25T16:55:34.972Z · score: 1 (1 votes) · LW · GW

From the Imperial College paper:

For countries able to achieve it, this leaves suppression as the preferred policy option. [...] The major challenge of suppression is that this type of intensive intervention package--or something equivalently effective at reducing transmission--will need to be maintained until a vaccine becomes available (potentially 18 months or more)--given that we predict that transmission will quickly rebound if interventions are relaxed. We show that intermittent social distancing--triggered by trends in disease surveillance--may allow interventions to be relaxed temporarily in relative short time windows, but measures will need to be reintroduced if or when case numbers rebound.

So it's not really saying "lock down for 18 months or do nothing". It's saying lock down until the problem is controlled and then squash new outbreaks quickly. The Hammer and the Dance article makes this point more clearly in my opinion, especially pointing out that a temporary lockdown would give us time to build up test and treatment capacity, protective equipment supplies, etc, and implement strategies for tracing and suppressing the new cases that arise after it's lifted. However it does say "a few weeks" of strict lockdown will be enough without really supporting it well, and that seems optimistic (Bill Gates, for instance, has been saying 6-10 weeks).

Comment by erickball on The Solution is Inaction · 2020-03-25T01:03:20.733Z · score: 4 (3 votes) · LW · GW
On a tangent, I'm honestly not sure why we can't do option (1) with regard to rent, but for Chesterton's Fence reasons, I assume that landlords have to burn a certain amount of money per month in offering to the Elder Gods or else we will all die.

I don't see why you would expect landlords to have more extra cash on hand than anyone else? In the case of individuals who own one or a small number of properties, you'd expect them to have mortgages, and/or maybe living expenses if property management is their full-time job or they're retired. For larger businesses, they likely also have debt to service, plus employees and stuff.

Comment by erickball on What is the point of College? Specifically is it worth investing time to gain knowledge? · 2020-03-24T20:25:32.288Z · score: 2 (2 votes) · LW · GW

I guess I wasn't counting every little derivation or example or even formula that comes up in a class against that 90%. Those are things you see in lecture, but you don't "learn" them. The stuff you actually learn is concepts and techniques, like what you would need to answer the test questions. Even that stuff, of course, you'll mostly forget if you don't review it regularly. But... I'm not very confident that you can strip out all the stuff you're going to forget and still learn the stuff you would have remembered. I don't know of any real examples of this working. It seems like maybe the academic system of "present a whole bunch of info rapidly and then force students to study key ideas from it for homework and tests" might have become entrenched over time because it performs better than the obvious alternatives.

I would love to see evidence of something better, though. It seems like good use of spaced repetition is a non-obvious candidate to replace the lecture-homework-test system. If you haven't seen quantum.country, that's the kind of thing I'm thinking of and my initial experience with it is promising.

Comment by erickball on What is the point of College? Specifically is it worth investing time to gain knowledge? · 2020-03-24T15:10:20.606Z · score: 3 (3 votes) · LW · GW

I think 90% is an exaggeration, but even if you get a research job in the same sub-field as your graduate degree, it's true you probably will not be using the majority of your engineering course work. The problem is that you can't pick out the 20-30% you will benefit from and learn that in isolation. And I don't think judging on the basis of which courses you did or didn't like is a very good indicator (though admittedly there were some I disliked so much that I did badly and learned next to nothing). The way an engineering curriculum is structured is to let each topic build on past knowledge. Consider the class on statics that you probably took or are taking. Could you have understood bending moments and deflection of beams without first learning the concepts of derivatives and integrals? How about trigonometry? If you're like me, you've forgotten 90% of your high school trigonometry class, but if you hadn't taken it at all you'd really be struggling with the more advanced material.

I do think a lot of courses could be condensed down to a much smaller kernel of key concepts, which could be taught in far less than a semester and then retained with spaced repetition. If you have the impetus and self-discipline to design a system like that for yourself then it might be worth trying, with a couple of caveats. First, the credential of a degree is pretty important--more so in mechanical engineering than in computer science. Second, you want to choose a curriculum designed by someone who has already learned this stuff--you can't pick the topics yourself. And third, part of what you're doing when you take an engineering class is practicing application of key ideas. To return to trigonometry--we did lots and lots of proofs using trig identities. I have forgotten most of them, other than sin^2+cos^2=1. But I don't think it was all wasted, because I still derive formulas from geometric relationships sometimes, and I think it would be much more difficult and error-prone if I hadn't done a lot of that sort of thing in school. This is a small portion of my work hours but probably some of the highest-value work I do.

Here's another example from my experience. I was asked to build a simulation of a system based on some old photocopies of engineering diagrams. These included a big list of Laplace-domain controller coefficients and a diagram for a feedback system using lead and lag controllers. One of my colleagues had looked at it and didn't really understand what it meant so they ignored it. Luckily I had taken a controls class. I couldn't have defined a Laplace transform for you off the top of my head, but I could see how the controller fit into what we were modeling and figure out the gist of its purpose. My colleague could have borrowed a textbook, taken a few days to study control theory, and understood the system much better than I did. But even if that were practical in a business environment, you wouldn't know it was worth doing if you didn't have background knowledge of the topic. Did that (plus the few other times I've used control theory) make the whole course worth it? Maybe not in a naive calculation of time spent. But a big part of your value as an engineer is being able to recognize a problem and know where to look for the solution. If you have to call an expert for the solution that's often fine. But if you have to call in an expert to tell you what the problem is, well then they could have just skipped right past hiring you, couldn't they?

At one point I wrote some questions for a work-sample test to give to prospective hires at my company, with the goal of testing for general engineering competence. I don't know if they'd be helpful for demonstrating what I mean, but message me if you want to take a look at them.

Comment by erickball on What is the point of College? Specifically is it worth investing time to gain knowledge? · 2020-03-24T03:54:38.998Z · score: 3 (3 votes) · LW · GW

I agree with Dagon and johnswentworth's answers, and I would like to particularly emphasize one point. For me, as a mechanical engineer, most of the time I do not use most of the things I learned in school (and indeed I have forgotten most of them). Learning to do my job took months of studying field-specific stuff that most engineering schools don't teach. But if I had tried to learn it without first taking a lot of engineering courses, it would have taken years instead--and almost no company is willing to train someone for years. Think of education as turning yourself into a minimum viable product. It might be better in principle to intersperse work and school--either work as an engineer and then take time off to study, or do some study on the side. But you can't do either of those until you're skilled enough for someone to want to pay you.

Also, every now and then, some piece of knowledge from a course turns out to be useful, but you can't know in advance what it will be. If you haven't already studied it, then when the situation arises where you could use it you won't be able to tell; and you can't look something up unless you already know it's important.

Comment by erickball on Using smart thermometer data to estimate the number of coronavirus cases · 2020-03-24T01:22:25.913Z · score: 4 (2 votes) · LW · GW

I'm confused about the nature of the raw data here--if it's just spontaneous temperature-measurements, why would we expect it to show anything meaningful? They're showing a "% ill" which is presumably the percentage of people who took their temperature with a Kinsa thermometer recently who have a fever. But people generally only take their temperature when they think they might be sick. In a normal year, after calibration against some sort of ground-truth, I can see how you might still be able to use the data to track flu rates. But during a pandemic, won't people's pattern of temperature-taking completely change? I've taken my temperature way more in the last two weeks than I would normally (which is close to zero) and I bet the same is true for many other people. So the base rates will have changed a lot.

Comment by erickball on Loss of sense of smell as marker of COVID-19 infection · 2020-03-22T18:58:30.423Z · score: 1 (1 votes) · LW · GW

Does it say what the threshold we would use to determine if someone has anosmia?

Comment by erickball on When to Donate Masks? · 2020-03-22T18:56:07.315Z · score: 3 (2 votes) · LW · GW

Thank you, this exactly answers a question I had been wondering about. I have only about a dozen N95 masks and would like to donate them, but since my area is not hard-hit yet I'm imagining that they would each be used once and thrown out. I will wait a few weeks as you suggest.

Comment by erickball on March Coronavirus Open Thread · 2020-03-22T05:40:01.020Z · score: 2 (2 votes) · LW · GW

Hmm, maybe you're right. The South Korean distribution of cases by age here suggests that it's actually most common by far among people in their twenties, and the larger number of confirmed cases among older people is a statistical artifact resulting from test criteria. The data do look a bit suspicious though.

Comment by erickball on March Coronavirus Open Thread · 2020-03-21T16:20:03.457Z · score: 1 (1 votes) · LW · GW

You found an age distribution for the infected population on the Diamond Princess, but you're using it as if it's the age distribution for everyone on the ship. Older people are more likely to get infected, so the infected population in the US will lean older as well--closer to the distribution on the ship. To do a good age adjustment we need to know the ages of the people on the ship who were not infected.

Comment by erickball on Preprint says R0=~5 (!) / infection fatality ratio=~0.1%. Thoughts? · 2020-03-20T14:03:34.924Z · score: 1 (3 votes) · LW · GW

The ~1% infection fatality rate on the Diamond Princess (where everyone was tested) is pretty solid evidence against this.

Comment by erickball on How useful are masks during an epidemic? · 2020-03-17T20:36:40.959Z · score: 4 (3 votes) · LW · GW

You can see the fit-testing on youtube. It's pretty involved and not something you can do at home--they put a hood over your head and spray in a bunch of super-sweet aerosol, and if you can taste it through the mask then the fit isn't good enough.

On the other hand, they do that once a year, just to make sure that model is capable of fitting your face shape. The masks are supposed to be one-size-fits most. On a daily basis, the fit procedure is to put it on and blow out and see if you can feel any air escaping around your nose. This seems very achievable and likely to give a decent fit for most people.

Comment by erickball on When to Reverse Quarantine and Other COVID-19 Considerations · 2020-03-17T20:25:42.608Z · score: 2 (2 votes) · LW · GW

From what I can google, typical dosage of quinine for malaria would be at least 500mg every 8 hours. If you drink 6 liters of tonic water every 8 hours you'll have more to worry about than coronavirus. What I don't understand is why they haven't started treating Covid with chloroquine yet--it's cheap and plentiful and has minimal side effects. Are desperate doctors in overwhelmed ICUs really holding off just because it would be off-label?

Comment by erickball on More Dakka for Coronavirus: We need immediate human trials of many vaccine-candidates and simultaneous manufacturing of all of them · 2020-03-15T00:24:12.596Z · score: 1 (1 votes) · LW · GW

Yes, I'm referring to this line:

" · Group 2: standard care + VCO (45 mL, approx. 3 three tablespoons, daily or higher,) "

Does this just mean you feed people 3 tbsp/day of regular coconut oil and they think it will have a positive effect on outcomes for people infected with nCov-2019?

Comment by erickball on What is a School? · 2020-03-15T00:21:42.076Z · score: 4 (3 votes) · LW · GW

I suspect if online schooling continues for more than a couple of months there will be a lot of complaints (likely justified, in my estimation) that it is not an adequate substitute for in-person education. In particular, I think maintaining student motivation would be an issue--more in some age groups than others, but even among college students I believe this is an existing problem with online courses.

Comment by erickball on More Dakka for Coronavirus: We need immediate human trials of many vaccine-candidates and simultaneous manufacturing of all of them · 2020-03-14T23:56:32.897Z · score: 1 (1 votes) · LW · GW

Is the coconut oil thing for real? Are they suggesting a clinical trial where people just eat 3 tbsp/day of coconut oil and it's expected to have a significant antiviral effect? Or is it something more complicated?

Comment by erickball on March Coronavirus Open Thread · 2020-03-14T22:00:19.021Z · score: 3 (2 votes) · LW · GW

Hopefully someone more knowledgeable can weigh in, but in the meantime I want to sound a note of caution on this: my experience with pulse oximeters is that by the time it drops noticeably below normal, you are already quite sick and walking a mile may not be practical. The point of the pulse ox is more to be able to quantify your symptoms so that you can have a specific threshold for when you should seek emergency treatment. I don't believe it's a leading indicator and so it may not be helpful for this situation.

Comment by erickball on March Coronavirus Open Thread · 2020-03-14T21:53:36.448Z · score: 3 (2 votes) · LW · GW

Please do not get on a plane if you have symptoms. If you are in a high risk category and/or very scared, you could go to Singapore preemptively, since they've already shown they can control their outbreak.

Comment by erickball on March Coronavirus Open Thread · 2020-03-13T13:31:34.695Z · score: 8 (3 votes) · LW · GW

Follow-up: Trevor Bedford has also debunked the claim in this twitter thread, saying that by the time Wuhan had 100,000 infections there were 1000 severe cases and 300 deaths. For Ohio to be in that state now the disease would have had to be spreading there since about mid-December.

Comment by erickball on March Coronavirus Open Thread · 2020-03-12T23:42:04.248Z · score: 5 (4 votes) · LW · GW

Ohio health official estimates 100,000 people in state have coronavirus:

https://thehill.com/policy/healthcare/487329-ohio-health-official-estimates-100000-people-in-state-have-coronavirus

This sounds crazy, but I don't understand the methodology so I'm not sure... Do people think it's plausible?

Comment by erickball on What I'm doing to fight Coronavirus · 2020-03-09T22:00:16.627Z · score: 2 (2 votes) · LW · GW

Cool idea! Do you need one for each hand to make it effective?

Comment by erickball on March Coronavirus Open Thread · 2020-03-09T21:21:59.290Z · score: 3 (2 votes) · LW · GW

Unfortunately it doesn't let you modify the assumptions about disease severity or number of undetected cases. It assumes that the majority of cases have been undetected (which seems questionable) and that 4.31% of cases are severe (which seems low even if the majority are undetected). It gives a case fatality rate of 0.97%, which doesn't seem to depend on any of the other parameters.

In their baseline scenario (for a small Swiss city with good infection control) 0.26% of the population dies.

With no infection control this goes up to 0.76% of the population dying, with no change in the CFR.

If you also increase the length of a hospital stay from 10 days to 20 days, the total number of deaths actually decreases slightly because the spread is slower. So while the graph is a nice way to see how long hospitals will be overwhelmed in different scenarios, it doesn't show you anything useful about how this affects outcomes. I would love to be able to add in some parameters for fatality rate for severe and critical cases with/without a hospital bed.

Comment by erickball on Seeing the Smoke · 2020-02-29T17:32:48.191Z · score: 4 (4 votes) · LW · GW

I believe this is in reference to the changes in European economies during/after the Black Death, where typical incomes rose significantly. However, my understanding is that this happened mainly because most capital was in the form of land, which of course was unaffected by the plague. Thus, the ratio of capital to labor went up and the value of labor increased. In the modern world where agriculture is a relatively small part of the economy, it's not at all clear that a reduction in population density would cause individual incomes to increase.

Comment by erickball on Operationalizing Newcomb's Problem · 2020-02-21T13:30:29.474Z · score: 2 (2 votes) · LW · GW
I'll be honest, this sentence confuses me. I don't know what to make of it.

Maybe I was mixing two different ideas together here.

One is about dualism, the assumption that the mind can be treated as a magic box that takes in only sensory inputs and outputs only motor signals to the muscles. The way we normally think about decision making is that our thoughts affect our subsequent decisions, which affect our subsequent actions, and those actions affect what happens to us. Causal decision theory is appropriate for situations that follow this pattern (causal, because the decision causes the action which causes the result). All you have to worry about when you make decisions is what effect your actions will have. But imagine that someone could see inside your brain and decipher what you're thinking about. Now even before you've decided what to do, your thoughts have affected the person reading your mind. They can react to what you were thinking in a way that affects you, even without you taking any action. The magic box assumption is broken because your mind has let something leak out besides muscle signals. Now, when you make a decision about how to act, you have to take into account not only how those actions will affect the world, but also how the thought process behind them will affect the mind-reader who is observing them. (In the OP, the polygraph plays the role of mind reader; in Newcomb's problem, the perfect predictor does.)

The other idea, which is related, is that your thought process may "affect" the world non-causally and even backwards in time. I use the scare quotes because of course if it's not causal it's not really affecting things--it's really just a correlation. But there are hypotheticals where it could seem a lot like a causal effect because the correlation with your thought process is perfect or near-perfect. The twin prisoner's dilemma is a good example of this. It relies on knowing that there's a perfect copy of yourself out in the world. Since it's a perfect copy (and the setup of the problem is symmetric; you and the copy both encounter identical situations), you know that the copy will decide whatever you decide. This is true even if the copy makes its decision before you make yours. If you decide to cooperate, then you will find that it already cooperated. Likewise in Newcomb's problem: time doesn't have to be a puddle in order for one-boxing to make sense. You cannot cause Omega to predict that you will one-box, because it already happened; but if you decide to one-box, then you always were the kind of person who would decide to one-box in this situation--effectively Omega had a near-perfect copy of you that it could observe when it made the prediction, even if the copy was just in its head, and just like in the twin prisoner's dilemma, that copy would have decided whatever you end up deciding. By choosing the decision process used by you and all copies of you and perfect predictions of you, you constrain the past decisions of those copies, which may in turn causally affect what situations you encounter.

Comment by erickball on How to Identify an Immoral Maze · 2020-01-17T03:04:23.297Z · score: 1 (1 votes) · LW · GW

One factor is that the military has a pretty consistent policy of moving officers around to different postings every few years. You never work with the same people very long, except maybe at the very top. This might help enable some of the outrunning-your-mistakes phenomenon mentioned above, but it also probably means you can't develop the kind of interpersonal politics you might see in a big corporation.

Comment by erickball on human psycholinguists: a critical appraisal · 2020-01-05T16:29:37.444Z · score: 2 (2 votes) · LW · GW

Wait, why is it impossible for a fully-connected network trained by backpropagation to generalize across unseen input and output nodes? Is this supposed to be obvious?

Comment by erickball on human psycholinguists: a critical appraisal · 2020-01-05T16:28:32.782Z · score: 7 (4 votes) · LW · GW
If compositionality is necessary, then this sort of “deep learning” implements compositionality, even if this fact is not superficially obvious from its structure. 

But compositionality mostly isn't necessary for the kind of writing GPT-2 does. Try getting it to tell you how many legs a dog has. One? Three? Five? It doesn't know, because people rarely write things like "a dog has four legs" in its input data. Here's GPT-2:

A dog has the same number of legs as  a man, but has fewer legs than a gorilla.  It has a lot of brains, but they are divided equally between the two front legs. 

It's very good at coming up with sentences that are grammatically and stylistically correct, but it has no concept of whether they're true. Now, maybe that's just a result of interacting exclusively with language--people generally learn how many legs a dog has by looking at one, not by hearing about them. But even when it comes to purely linguistic issues, it basically doesn't make true statements. This is typified by its habit of contradicting itself (or repeating itself) within the same sentence:

A convincing argument requires certain objects to exist. Otherwise it's not really science. For instance, according to Stromberg, all the atoms in a dog's body exist, and the image of the dog, even though it does not exist, exists (p. 120).
Comment by erickball on Operationalizing Newcomb's Problem · 2019-12-16T15:38:34.097Z · score: 3 (2 votes) · LW · GW

I actually think if you're new to the idea of logical decision theory, Newcomb's problem is not the best place to start. The Twin Prisoner's Dilemma is much more intuitive and makes the same basic point: that there are situations where your choice of thought process can help to determine the world you find yourself in--that making decisions in a dualist framework (one that assumes your thoughts affect the world only through your actions) can sometimes be leaving out important information.

Comment by erickball on Operationalizing Newcomb's Problem · 2019-12-13T16:37:46.838Z · score: 2 (2 votes) · LW · GW

I guess if you ran this experiment for real, any answer along the lines of "I don't know whether I'll stay" would have to result in getting $10.

Comment by erickball on Operationalizing Newcomb's Problem · 2019-12-13T16:33:52.320Z · score: 1 (1 votes) · LW · GW

Sorry if that was unclear--I meant that the lie detector has 10% false positives, so those people were telling the truth when they said they would stay, but got the $10 anyway because the lie detector thought they were lying.

Comment by erickball on Epistemic Spot Check: Fatigue and the Central Governor Module · 2019-12-04T22:32:03.819Z · score: 3 (2 votes) · LW · GW

Levine's comments seem to be a bit of a strawman of Noakes' position... though, based on your second quote, he may also be a strawman of himself.

A subject can elect to simply stop exercising on the treadmill while walking slowly because they don’t want to continue; no mystical ‘central governor’ is required to hypothesize or predict a VO2 below maximal achievable oxygen transport in this case.

Whether sub-maximal oxygen transport ever occurs is entirely beside the point; the implication of the central governor model would be that maximal oxygen transport (or maximal muscle contraction, or whatever) occurs only under extreme circumstances, because the brain subconsciously detects potential for damage to the body and reduces exercise effort. A VO2 max is generally determined by running some exercise test "until exhaustion", with no mechanism in place for knowing how hard the athlete is trying. Presumably the only motivation is "this doctor just told me to keep going until I'm tired" or maybe "I want a good test score for bragging rights". It's definitely not "baby trapped under the car". In this situation the evidence for Noakes' hypothesis would be something like, measure VO2 max again while being chased by a bear (or during some kind of competition, if the IRB doesn't like the bear thing) and it should be higher the second time. If you can improve performance through increased motivation, up to the highest levels of motivation you're willing/able to measure, then common sense would say you still probably haven't reached the actual "maximum". Anecdotally, in at least one instance of extreme motivation I've been able to cycle so hard that in a period of about 30 seconds I gave myself a nasty cough lasting several days; I am unable to repeat this under normal circumstances.

Whether the performance limitation occurs primarily through pain or whether there's also some kind of hard limit is unclear. Limitation of exercise performance by fatigue is familiar to anyone who's tried running a mile in gym class. On the other hand, I've read that major league pitchers are often limited in pitching speed by the stress on their elbow ligaments, rather than by the ability of the muscles to contract harder. If true that means the brain is detecting potentially injurious muscle contractions and putting a damper on them, but without necessarily causing pain.

Comment by erickball on (Reinventing wheels) Maybe our world has become more people-shaped. · 2019-12-04T18:50:16.669Z · score: 1 (1 votes) · LW · GW

But we have no problem observing causality in nature as well as in man-made environments. It seems like human culture has not so much made the world friendly to human concepts of causality, rather it has built up a standard set of human-friendly abstractions that are selected for their ability to fit causal models onto a complex world. There are lots of parts of the world where causality exists but is not observable through abstractions (e.g. butterfly effects). We generally ignore these.

Comment by erickball on Can you eliminate memetic scarcity, instead of fighting? · 2019-11-26T18:09:57.000Z · score: 1 (1 votes) · LW · GW

Interesting. I like the idea here of having reusable patterns of problem-solving across diverse engineered physical systems. It reminds of the idea of software patterns. I'm actually kind of disappointed that this was never taught in any of my engineering classes, especially given Wikipedia's big list of organizations that use it (including Samsung, GE, Boeing, NASA, HP, Intel, and a bunch of others). Now I'm excited to read about some case studies!

If you have any examples of how you've used it, I'd love to hear about those too.

Comment by erickball on Matthew Walker's "Why We Sleep" Is Riddled with Scientific and Factual Errors · 2019-11-17T16:52:09.565Z · score: 3 (2 votes) · LW · GW

Yes, this just-so-story is suspicious, especially because it's not "just so"--if there were such strong selection for variability, wouldn't you expect full coverage of the night? Some people could go to bed at four and others could wake up at three. As far as I know this does not generally happen (in the absence of electric lighting) and hence the dangers of having everybody asleep at once must be manageable.

Comment by erickball on Arguing about housing · 2019-11-15T22:42:33.129Z · score: 11 (3 votes) · LW · GW

Your solution is... a bunk bed with cabinets built in?

Squeezing everyone into college-dorm-style housing would certainly reduce living costs, but people who want that can already do it. Most don't.

Comment by erickball on Hard to find factors messing up experiments: Examples? · 2019-11-15T21:23:07.430Z · score: 15 (9 votes) · LW · GW

Not an experiment but similar situation: I talked with a scientist at NIST who told me that their nuclear reactor was shut down once for maintenance, and while the core was out they had the cleaning staff come in and scrub the aluminum panels that lined the walls, since they hadn't been cleaned in years. When the maintenance was done, the reactor wouldn't start back up. Eventually they figured out that the cleaning spray used on the walls contained borax. Boron is a potent neutron absorber and the trace amounts left behind were keeping the reactor from going critical. They had to toss out all the panels and replace them.

Comment by erickball on Operationalizing Newcomb's Problem · 2019-11-13T17:15:09.270Z · score: 1 (1 votes) · LW · GW

Yes, that is a disadvantage to this formulation... as with real-world analogues of the Prisoner's Dilemma, personal ethical principles tend to creep in and muddy the purely game-theoretic calculations. The key question, though, is not how well you can lie--it's whether, once you've decided to be honest either due to ethics or because of the lie detector, you can still say you'll stay and precommit to not changing your mind after the test is over.

As for why you should care, the truth is that for most situations where causal decision theory gives us a harmful answer, most people already tend not to use causal decision theory. Instead we use a set of heuristics built up over time and experience--things like altruism or desire for revenge. As long as the decisions you face more or less match the environment in which these heuristics were developed, they work pretty well, or at least better than CDT. For example, in the ultimatum game, the responses of the general population are pretty close to the recommendations of UDT, while economists do worse (sorry, can't find the link right now).

Really understanding decision theory, to the extent that we can understand it, is useful when either the heuristics fail (hyperbolic discounting, maybe? plus more exotic hypotheticals) or when you need to set up formal decision-making rules for a system. Imagine a company, for instance, that has it written irrevocably into the charter that it will never settle a lawsuit. Lawyer costs per lawsuit go up, but the number of payouts goes down as people have less incentive to sue. Generalizing this kind of precommitment would be even more useful.

UDT might also allow cooperation between people who understand it, in situations where there are normally large costs associated with lack of trust. Insurance, for instance, or collective bargaining (or price-fixing: not all applications are necessarily good).

Comment by erickball on Operationalizing Newcomb's Problem · 2019-11-12T19:38:21.827Z · score: 2 (2 votes) · LW · GW

I don't see how two-boxing is a Nash equilibrium. Are you saying you should two-box in a transparent Newcomb's problem if Omega has predicted you will two-box? Isn't this pretty much analogous to counterfactual mugging, where UDT says we should one-box?

Comment by erickball on Why so much variance in human intelligence? · 2019-11-08T19:41:37.743Z · score: 12 (3 votes) · LW · GW

Summary of the 10 findings in the linked paper (Plomin et al 2016):

They show that all of these have large effect sizes and are well replicated, except where noted below. I notice that the authors cite themselves a lot as support for many of these claims. I am not an expert in any of this, so if they're trying to push controversial ideas as widely accepted, I wouldn't be able to see through it.

1. Significant genetic influence is ubiquitous in cognitive and psychological traits. Intelligence has about 50% heritability. Twin studies show intelligence correlation about 0.85 in identical twins vs 0.6 in fraternal twins.

2. Although basically all psychological traits have some heritability (typically 30-50%) none of them have close to 100% heritability. Contrast this with physical traits like height, which has about 90% heritability.

3. Heritability of complex traits is caused by many genes of small effect that add up. Example: tendency for open-field activity in mice shows a linear response to selection pressure over 30 generations, rather than a clear separation that would occur if it were controlled by just a few genes. "Genome-wide association" studies hundreds of thousands or millions of nucleotides covering most of the genome to detect population associations between a single-nucleotide polymorphism and a trait. It generally finds that even the most significant genetic changes by themselves have tiny effects (far less than 1% of variation).

4. Correlations between traits are usually caused largely by genetics. For example, the strong correlation between types of intelligence (R=0.76 between reading and math) is due more to genetics than environment (the reading/math correlation is about 64% genetic). Anxiety and depression are correlated entirely for genetic reasons (they are affected by all the same genes). The schizophrenia/bipolar connection is largely genetic too, as is neuroticism/depression. Another finding (not yet replicated) is that the correlation of 0.3 between exercise behavior and attitudes toward exercise is 70% genetic; I interpret this to mean that most of the genetic influence on exercise behavior is caused by the influence of those same genes on attitudes toward exercise.

5. Counterintuitively, heritability of intelligence increases linearly throughout development (from 41% at age 9 to 66% at age 17 in one twin study, and maybe as high as 80% in adulthood).

6. Stability of traits from age to age is largely due to genetics; changes that occur with age are largely environmental. So then how does the heritability of intelligence increase over time? The authors suggest "genetic amplification": genetic nudges early in development get magnified as time goes by, perhaps due to genotype-environment correlation (kids choose or create environments that match their propensities). Some evidence supports this idea, but it may vary depending on the culture. The authors do not seem to consider genetic amplification one of the "replicated findings" noted in the title.

7. Most measures of the 'environment' show significant genetic influence. This is a generalization of the genotype-environment correlation noted above for intelligence. Parenting, social support, and some life events seem to be causally affected by a child's genetics (not just correlated); this can be shown in twin studies. Same goes for school and work environments. Heritability averages 0.27. This again varies with culture; parenting is more affected by the child's genetics in Japan than in Sweden. A child's genetics have even been shown to have some effect on the family's socioeconomic status.

8. Most associations between environmental measures and psychological traits are significantly mediated by genetics. Since genetic factors affect environmental measures as well as behavioral measures, we should not assume that correlations between parenting and children’s behavior are caused entirely by the environmental effect of parenting on children’s behavior. For instance, correlation between a child's developmental index and measures of their home environment is stronger for genetically related families (0.44) than adopted families (0.29). So, much of what appears to be the effect of parenting on behavior is actually effect of the parents' and child's shared genetics on both the behavior and the environment. Disentangling genetic and environmental influences is important because it allows us to tailor interventions more effectively.

9. Most environmental effects are not shared by children growing up in the same family: salient experiences are specific to each child. Similarity among siblings is mainly due to shared genetics. Non-shared environment has a bigger effect on phenotypic variance than shared environment does. Shared environment between siblings (including going to the same schools) accounts for 10-15% of variance in academic achievement. Shared environment's effect on intelligence decreases after adolescence. Specific non-shared environmental effects are hard to identify, and are likely due to additive effects of many seemingly inconsequential experiences.

10. Abnormal is normal: quantitative genetic methods suggest that common psychological disorders are the extremes of the same genetic factors responsible for heritability throughout the distribution. Reading disabilities, for instance, have been shown to have strong "group heritability," indicating a genetic link between the disorder and normal variation in quantitative measures of reading ability. This is supported by the finding that many genes of small effect determine heritability of traits (finding 3); polygenic scores that sum these effects are normally distributed. An interesting exception involves severe intellectual disability (IQ < 70), which this type of analysis suggests is etiologically distinct from the normal distribution of intelligence (no significant group heritability).

The authors suggest that the above findings have replicated because: the controversy of the nature/nurture debate has motivated bigger and better studies; behavioral genetics has historically used better statistical methods than much of psychology, partly because studies often have to be observational rather than experimental; focusing on the net effects of genetics and environment is more reliable than studying specific genes (polygenic scores work better); there are better incentives and opportunities (data) for replications; and because genetic effect sizes are larger than other factors studied in psychology (e.g. sex differences generally account for less than 1% of variance on psychological traits). Many of these advantages cannot easily transfer to other fields.

Comment by erickball on Rohin Shah on reasons for AI optimism · 2019-11-01T18:36:21.443Z · score: 4 (2 votes) · LW · GW
Suppose [...] you’ve got this AI system with this really, really good intelligence, which maybe we’ll call it a world model or just general intelligence. And this intelligence can take in any utility function, and optimize it, and you plug in the incorrect utility function, and catastrophe happens.

I've seen various people make the argument that this is not how AI works and it's not how AGI will work--it's basically the old "tool AI" vs "agent AI" debate. But I think the only reason current AI doesn't do this is because we can't make it do this yet: the default customer requirement for a general intelligence is that it should be able to do whatever task the user asks it to do.

So far the ability of AI to understand a request is very limited (poor natural language skills). But once you have an agent that can understand what you're asking, of course you would design it to optimize new objectives on request, bounded of course by some built-in rules about not committing crimes or manipulating people or seizing control of the world (easy, I assume). Otherwise, you'd need to build a new system for every type of goal, and that's basically just narrow AI.

If our superintelligent AI is just a bunch of well developed heuristics, it is unlikely that those heuristics will be generatively strategic enough to engage in super-long-term planning

If the heuristics are optimized for "be able to satisfy requests from humans" and those requests sometimes require long-term planning, then the skill will develop. If it's only good at satisfying simple requests that don't require planning, in what sense is it superintelligent?