Case study: Folding@home
Latest in an irregular series, some of whose previous entries were Edge.org and the Girl Scouts...
I examine the Folding@home distributed computing project with reference to the costs (electricity resulting in air pollution causing deaths) and benefits (some papers): http://www.gwern.net/Charity is not about helping. Additional data on either side of the cost-benefit is welcome.
(I also recently split out my essay describing things I have changed my mind on.)
I have questioned before the wisdom of spending money to “save lives” in countries that do not have a functioning economy. If I save someone’s life, I’d prefer to think that their life will not need to be saved by my charity dollars again next year, and the year after that. It is not at all clear to me that GiveWell or other charities presents data that accounts for this. I’d be interested if you have done any analysis on the topic.
Instead of looking at a rough figure of the cost to “save a life” for an unspecified period of time, I’d prefer to see alternatives considered on the basis of the net present value of the external charity funds that will likely be required to continue saving an average’s person’s life for the next 70 years or so (without considering the related issue of cost-burden for geriatric medicine).
For example if we know a particular region has cycles of drought, war, famine and starvation every fifteen years or so, then instead of considering only the cost X—where X is the funds required to see a person through the present crisis—we should consider the cost of saving their life to be 70/15X. This method would favor spending in regions that are less likely to be in the same circumstances again and again.
I haven’t, because it seems obviously false to me. Africa’s population has been growing exponentially over the 20-21st centuries, aid to Africa has not, as far as I know. If conditions in Africa were so bad that aid contributed only a fixed number of life-days—if it only saved people to die another day, then I would not expect to see continued exponential growth but rather, re-hitting the Malthusian ceiling and quickly.
(One could try to explain this by saying the life expectancies and public health changes are thanks to non-aid Western inputs like regular commercial goods, but this strikes as an ad hoc explanation to try to deny the value of aid.)
In your cost analysis, you’re comparing how much electricity is used by folding@home to the number zero. But your thesis is that participants in folding-at-home are doing so to pursue status. Isn’t it important to consider how much electricity is used by alternative forms of status-seeking? This is hard to quantify, but the counterfactual world where there is no status-seeking, or even where there is less status-seeking, is sort of hard to imagine.
By the way is this number available in QALYs?
I see it more as a fake-altruism-signaling game—eliminate the harmful signal (Folding@home), and the activities aren’t 100% displaced into other activities, more or less harmful (I don’t think the demand is inelastic); to put it another way, https://www.xkcd.com/871/
I didn’t see QALYs mentioned. My general understanding is that air pollution disproportionately hurts the old, so the QALY loss is not so bad as it could be, but then, the comparison is to interventions in Africa focused on kids, who have many more potential QALYs than old Westerners… so I suspect on net it strengthens the case against Folding@home.
I think there there are many forgone QALY’s for the young due to air pollution that are just too hard to measure! My dream girl lives in Bangkok and doesn’t want to leave it...but I don’t want to live there because of the pollution and culture of scamming foreigners!
I found the part about the American Revolutionary War in “My Mistakes” very interesting—it was a perspective I hadn’t considered before.
Perhaps you should try and compare costs and benefits more directly?
In particular: how much research can get done by other means for $12 million a year? By paying researchers and technicians, maintaining lab space, etc. For example, MIT spends on the order of $1.2 billion a year on research. How does the contribution of folding@home compare to 1% of the research done at MIT each year? Of course, we’d be buying marginal research, not average research: we can get a good lower bound on the marginal benefit of more money by looking at the value of more equipment or somewhat skilled labor for labs (money that hires additional researchers may just be poaching from other efforts, so it’s a little harder to get conservative estimates).
This analysis may be slightly less straightforward, but it gives a much stronger argument against spending power on folding@home. Regardless of how the individual feels about the tradeoff between fundamental progress and lives in the developing world today, they would probably be better served by giving to a charity that does fundamental research.
There is also potentially a large benefit from understanding how efforts like folding@home could be better run. The most conservative estimate for its value would be the marginal benefit of the most effective similar distributed computing project (if the main bottleneck in improving effectiveness was experience with similar distributed computing projects, which may be unlikely).
I was going to say, I can see how I might estimate the value of 1% of MIT’s scientific output, but I don’t see how I would do this. Where do you get the value of more equipment or more skilled labor? Is there a data source or Fermi calculation I am overlooking?
I endorse the theory of the My Mistakes audit, but find your practice is unusual. One of my most valuable documents is a text file in which I have a few items—title of the file is “things I used to believe were true which I have since found out to be false”. One time I was at a seminar and told the seminar leader about my file and he admonished me:
“Don’t ever do that!”
I changed the subject but I gathered afterward his problem was that such a file is harmful to our precarious sense of self-esteem.
Anyway, the thing which you have done which I almost never do is share this information. To me it is a pretty private thing (they are largely private things).
Perhaps your “mistakes” file is only a partial listing?
Do you think I am making a mistake in publishing it?
It is, actually. I keep the personal mistakes in a separate file. (I don’t have any issues with revealing intellectual mistakes, mistakes of thought and belief—I have managed to keep my identity small but I do have a problem revealing actions.)
I thought I explicitly mentioned this, but checking, I see I only have
where I thought I had a parenthetical comment ‘(which go into a separate private file)’. Guess I should fix that...
The Revolutionary War angle is interesting, but I think it’s hard to figure out the implications of that huge a counterfactual.
Would the Civil War have been prevented because slavery would have been peacefully abolished? Or would there have been a war against that abolition? Perhaps not extremely likely—maybe the slaveholders would have been less willing to revolt against a bigger government, but it’s hard to be sure. Could a larger block of slaveholders have done more to prevent British abolition?
I’ve heard that the British were more careful with their other colonies because they didn’t want more of them breaking away.
WWI becomes really hard to predict. Would Germany have been more careful if Britain had been that much larger?
I admit I’m not sure if I’m arguing honestly on this. I may be assuming too much of a pull towards history as it happened.
That is true; but if one really is agnostic, doesn’t that argue against the Revolution, and actually strongly against it? You shouldn’t kill and wound 100,000 people (Wikipedia’s military count alone) and cause the self-exile of >62,000 refugees for something you don’t know to even be a net benefit!
Does my computer really use less energy when it’s running System Idle Process instead of Folding@home?
Yes. System Idle Process does not execute instructions, not even NOPs. It is instead in one of the power-saving idle states, where much of the cpu is turned off until it receives a hardware interrupt.
Then I’m going to stop running Folding@home, then.
ETA: I’m still going to leave my computer on at night, though, because it often freezes shortly after booting, but leaving it on lets me avoid this problem.
Yes, but to be clear it is using more power then when it is off. There’s a decent argument that when one works out the marginal power use it makes sense to run Folding@home or some similar program.
Your inline divisions (?) show up like this in Chrome for me:
Yeah, that happens with Chrome.
Luddism might be justified; at the same time, though, there seem to be potentially transient facts (like low productivity in the developing world) which may explain the problem. I’d also like to lean on them as an example: their lot is the result of being outside the factory system, which takes non-factory humans and produces factory workers with high productivity (cf. Hanson on same). It seems plausible that programming, etc. has that same feature now, so it’s not as if we’ve never done this before. We’ve done it at least twice (agriculture, factories).
I take it that you partially changed “my mistakes” to include nicotine. I enjoyed your article on it—but how are you using it?
Are you rotating with other stimulants on a regular basis, using when you like, using to promote habit formation etc.
See http://www.gwern.net/Nootropics#nicotine
I haven’t been doing anything systematically with it because I’m still getting a handle on dosage and effects—it is, unfortunately, not as powerful as modafinil or Adderall, so it can be hard for me to get a lock on how long it lasts, how to administer, etc to the point where I could try rotation or double-blinds.
WARNING. Nicotine is not a drug you want to be fucking around with. Treat the stuff with the same respect that you would treat cocaine or heroin.
One thing you should be aware of, if you aren’t already, is that, in users that haven’t already developed tolerance, the effects of nicotine last a long time. Even though the nicotine itself leaves the body within twelve hours, it can take as long as a week for rebound/withdrawal symptoms (including cravings) to set in.
I did not run into this in my reading, no. What is your source for these claims?
http://www.scientificamerican.com/article.cfm?id=hooked-from-the-first-cigarette (link is paygated, I read it in the dead tree edition of the magazine)
http://www.jfponline.com/Pages.asp?AID=5711
I quit the nicotine for 6 or 7 days following your comment, and didn’t notice anything, so at least for me, I’m not concerned. (Obviously this lack of concern presumes that I really am/was taking nicotine; the effects still aren’t clear to me, and I’m pondering buying some nicotine pills for an independent comparison.)
As far as your second link goes, I didn’t find it too persuasive; it deals exclusively with tobacco, as far as I can tell, which I had already cited as a relevant difference as to why I expected pure nicotine to be less addictive, and the scariest figures seem to be based on lumping in any positive answer on pretty general surveys (‘Have you ever felt like you really needed a cigarette?’ Gosh, who hasn’t?)
Well, since you didn’t report the equivalent of a “First Inhalation Relaxation Experience”, I suspect that you’re not in the quarter of the population at highest risk for nicotine addiction.
On the other hand, what good are those 12,000 people (poor Africans to boot) to you or me? Would you even notice if they lived or not?
Like solipsism, ethical egoism is consistent, irrefutable, and utterly unconvincing.
What is that even supposed to mean? For one thing, I don’t think it’s necessarily irrefutable, but don’t most people (even most good people) act more or less like that anyway, regardless of whether they are “convinced” by it? Virtually no-one even tries to maximize the good they do if you measure “good” as “human QALYs saved anywhere on Earth at the present time with high certainty”. Even what the SIAI is doing seems far closer to Folding to me than it does to “giving money to starving Africans”.
It means what you said is true and irrelevant. I would not notice if they lived or not. That doesn’t matter to my ethics. What good are they to me? Probably nothing; my track record on investing in charitable donations is precisely −100% (none have ever even paid interest!). That doesn’t matter to my ethics.
If you are going to attack consequentialism or valuing people besides oneself, this is entirely the wrong place to do so, and makes about as much sense as discussing Nagarjuna’s arguments that nothing exists as a refutation of the assertion ‘Obama is a bad president’.
They think that is not true. I guess you have a different opinion on the danger or likelihood of AI, or their effect on either one. You should take that up with them.
So why do you say that the Folding money would better be spent on starving Africans then? Shouldn’t it be donated to the SIAI instead, if you believe it? If not, why not criticize them on the same basis? I am also not claiming that one shouldn’t value other people, just that you don’t have to weight all lives equally and shouldn’t expect others to. And I don’t really believe that anyone truly maximizes “lives of others”, or would want to if they knew what it meant.
Also, “Charity X doesn’t optimize under my personal ethics” is not the same as “Charity is not about helping”, not that I disagree that signaling is important.
Because I am not writing for the tiny cluster of fellow zealots who agree about the high EV of donating to SIAI. I am writing for intelligent people in general, and one of the standard practices of philosophy—and heck, writing in general—is to not make highly controversial claims you do not need to make. I do not have to prove SIAI is the highest EV charity in existence in an essay about Folding@home; I only need to compare to a better charity, to establish a lower bound on how much harm choosing Folding@home does.
Fine, don’t look at my personal ethics. If you asked a random Folding@homer, ‘would you be willing to participate in murdering a few people just to make yourself look better’, what do you think they would say?
They would say no, but of course everyone “murders” some fraction of a person every day that they don’t maximize their life-saving effectiveness. If they were playing video games instead to make themselves feel better, isn’t that even “worse”?
People do folding because it’s easy and the costs are hidden. If you could magic it out of existence, they wouldn’t suddenly start donating equivalent money to maximally-efficient charities.
Of course they do; this follows directly from consequentialism, and is, in fact, an argument that could be used about donating to any charity but the most effective charity. If this was pointed to most people, they wouldn’t care and would ignore it or, like the XKCD link, simply indulge themselves that much more.
That’s why Folding@home is so interesting as a charity, because while most charities are simply committing sins of omission, Folding@home is committing sins of comission. (You did see the section headers, right? They aren’t meaningless.)
Sure, if you assume (as you do) that Folding won’t save many lives in the long run, it looks like a bad use of resources if you’re purely concerned about charity. But that assumption could be applied to any research that doesn’t pay off immediately or with certainty. The LHC uses 180 MW (http://lhc-machine-outreach.web.cern.ch/lhc-machine-outreach/faq/lhc-energy-consumption.htm) for example, or 12x as much as much as your numbers for F@H, and is arguably even less practical, despite no doubt producing many more papers.
So, you’re resorting to the same teapot argument I specifically addressed in the essay. No wonder this conversation is so frustrating for me—it looks to me like you didn’t read it, or skimmed it at best.
How is this situation comparable to any “teapot argument”? There are clear and obvious ways that something like Folding could produce valuable results. It might not, but it wouldn’t be shocking if it did, and it doesn’t need either a high probability or an especially large magnitude of success to make up for a few lives a year in EV. Comparing it to space teapots is getting things wrong by many orders of magnitude.
Anything looks bad if you incorrectly assume that the probability of it working is astronomically small.
edit: And the reason this conversation has been a bit confusing is because you made two different points (that F@H is a net negative even ignoring opportunity costs, and that giving charity to anything other than the #1 Givewell charity or close equivalent is bad), and while I don’t agree with either of them, they’re unrelated enough that it’s tricky to argue against both at once.
Which it has not. 10 years it has run, and the paper count seems to be dropping.
What is the prior probability for Folding@home justifying either its sins of commission or sins of omission? Now, what’s the posterior probability, conditional on what we have (not) observed, of it now retroactively justifying all past expenses, and then its ongoing expenses?
It costs up to a billion dollars (http://en.wikipedia.org/wiki/Drug_development) and up to 14 years (http://www.addictiontreatmentmagazine.com/addiction-treatment/what-it-takes-to-bring-new-treatment-drugs-to-market/) to create a new drug once the basic idea is discovered, and seeing as the companies doing this stay in business, that level of investment must be (reasonably) worthwhile economically. Folding is a drop in the bucket compared to that and even if it never achieves anything serious and is eventually shut down, it seems like it was worth trying—and there’s still a chance that it could discover something important about proteins.
edit: Also at best you don’t seem to have any better justification for thinking that the probability of significant success is astronomically low than I do for thinking that it’s low but not astronomically so, so this aspect of the argument isn’t really going anywhere.
Folding has already incurred somewhere around $100m in total expenses. Do the Bayesian update on a 1/10th chance not happening… It’s not epsilon or zero, I’ll tell you that!
Let me ask you something. I know what evidence would convince me that Folding was a good idea: show me a drug based on Folding results or a therapy change or something like that. But is there any evidence that could convince you that Folding is not a good idea? Because everything you’ve said seems like it could apply to any project.