[Question] Is running Folding@home /​ Rosetta@home beneficial?

[orenmn]

Two weeks ago, I found out about Folding@home, and have been running it since then.

Today I finally got around to googling it a bit (exam period is over), and here is what I found:

• A 2011 post by gwern that explains why it might be harmful, and points out that Rosetta@home seems like a better option. (There is also some discussion about gwern’s post in LessWrong).

• A 2012 reddit page with a top-voted answer that says that Folding@home unequivocally helped developing drugs to treat Alzheimer’s disease.

• A 2015 quora page in which two Stanford people highly praise Folding@home.

• A 2016 quora page in which it is pointed out that Folding@home influenced the development of Markov State Models (MSMs).

• A 2017 quora page with an offensive criticism of gwern’s post, whose main point (IIUC) is that gwern underestimated the potential of understanding protein folding.

• According to Wikipedia’s Folding@home page and the official website of Rosetta@home, the projects run 98.7 petaFLOPS and 270 teraFLOPS respectively, so Folding@home is (roughly speaking) 365 times faster. (FLOPS stands for ‘floating point operations per second’, i.e., how fast you can do arithmetic of real numbers.)

So what do you think? Is running Folding@home beneficial? Is running Rosetta@home more beneficial? Or maybe running either is harmful?

[BrownHairedEevee]

I just installed Folding@home on my laptop (partially inspired by this post), and for me, the cost so far has been close to zero. FAH took me about 10 minutes to install, and my time isn’t very valuable right now (spring break). It used up to 75% of my CPU capacity and I didn’t notice a drop in my laptop’s performance, and I don’t pay for electricity at my dorm. As for the external cost of running it, I don’t know what percentage of my school’s electricity comes from fossil fuels, so it’s hard for me to estimate my FAH instance’s carbon footprint.

However, I’m worried that FAH will cause my laptop’s fan to wear out more quickly because the laptop is not designed to crunch numbers ²⁴⁄₇. I think it’s best if I have a plan for maintaining whatever hardware I run FAH on, so I’m going to stop running it for now.

[Edit: I’m more concerned about the risk of damaging my laptop now.]

[Aaron Gertler]

Rather than get into the details, I’ll make the meta-level point that the impact of your action here is likely to be very small in one direction or another.

At best, you are one more computer in a network of millions*; at worst, you’ve added a tiny amount of pollution to the air, which might take a few minutes to an hour off of humanity’s collective lifespan, if we stick to Gwern’s reasoning—you might waste more human life in the course of spending time to install the software than you would actually running the program.

Meanwhile, the “indirect costs” are based mostly on money you could otherwise donate to charity, a consideration which could come up every time you spend money on anything (and which is generally better to ignore unless you’re making a big spending decision; I wouldn’t worry about $10/​year).

Given the complexity of the issue (e.g. trying to calculate your computer’s extra electricity usage, evaluating the expected value of papers produced through FAH), I would recommend against trying to make a serious calculation of your impact. As with many questions people ask in EA spaces, “don’t worry about it” is a reasonable answer.

----

*There are only about 100,000 machines in the FAH network right now, but many of those were designed specifically for high-performance computation; I’d be unsurprised if an average home machine contributed one-millionth or less of the project’s processing power.

[anonymous_ea]

I agree that the impact of this decision is likely to be very small, but trying to analyze a complicated phenomenon can be personally beneficial for improving your skills at analyzing the impact of other phenomenon. In general, it seems good for EAs to practice analyzing the impact of various interventions, as long as they keep in mind that the impact of the intervention and the direct value of the analysis might be small.

[Aaron Gertler]

This might be the case, though if someone has the time to analyze a complicated phenomenon and wants to get practice, I think they should take a bit more time to choose a phenomenon to start with, so that they can get one with other useful characteristics. For example, they might try to find something with a larger expected magnitude of impact, positive or negative, or to choose a question that is of direct relevance to the EA community (e.g. something which is an active topic of debate, or involves some very common thing many people in EA do).

Along those lines, I like Gwern’s study of melatonin, which involves a bit of self-experimentation but also expected-value calculations. Various other productivity tools/​strategies could also be solid candidates.

[Milan_Griffes]

cf. Gwern’s study of catnip.

Also Luke’s post on Scaruffi:

Sometimes I do blatantly useless things so I can flaunt my rejection of the often unhealthy “always optimize” pressures within the effective altruism community. So today, I’m going to write about rock music criticism.

[Aaron Gertler]

I certainly don’t endorse “always optimize”! I spend far too much time reading manga and trying to win Magic: the Gathering tournaments for that. I fully endorse analyzing things that are interesting/​entertaining. But it seems bad to get stuck with something that is both low-expected-impact and low-interest. Someone who really likes Folding@Home should totally give the analysis a go; someone who doesn’t care and just wants evaluation practice has many other options.

[anonymous_ea]

Since this post has gotten very little traction, I wanted to let you (orenmn) know that at least I found it valuable and interesting!

[orenmn]

Thank you :)

[Max Ghenis]

One element of this I’d like to understand better is the trade-off between donating your own compute (i.e. donating money you’re spending on electricity) compared to donating money directly to a protein folding effort that utilizes cloud computing. In general, how do the hardware savings of using an existing device compare to the greater efficiency of cloud computing? Is the key innovation of Folding@Home that they’re able to obtain in-kind electricity donations from people who wouldn’t otherwise donate money to fund the effort?