We discussed this, but it might make more sense to add/expand the section of the main GUV article: https://en.wikipedia.org/wiki/Ultraviolet_germicidal_irradiation Maybe opening a discussion in wikitalk about whether it merits its own page, too?
jvb
Karma: 230
Anecdotally, some of the worst conditions I’ve seen were on “pasture-raised” farms.
I’m very surprised by that anecdote actually! Would you mind elaborating? I was prepared for “better than other certifications, but still awful” but not “worse, actually”
I’m also curious for your personal take on “beef and dairy only” diets from a welfare impact perspective
I just wanted to note that I’m very glad people are still producing evocative, classic-EA-style pieces like this. Thank you!
Thank you for doing this. However, there are some inaccuracies and misconceptions in the UVC section. I won’t list them all here—chief among them is the equivocation between far-UVC and conventional UVC—but if I had comment access I could go through it.
I wrote a few of the reports you link to (though the executive summary is deprecated and does not represent our current thinking—this one does) so I will go ahead and answer the questions you asked in the doc to avoid unnecessary duplicated work:
What makes inexpensive, energy-efficient UVC light on the market ineffective?
Operational difficulty. Trained technicians are required to install upper-room systems safely, and such technicians are in short supply. Installing it in-duct has the same issues, with additional validation and transparency problem. Check out Jesse’s post on this subject.
Do they emit unspecific or varying wavelengths?
No, they emit nearly monochromatically at 254 nm.
What is the efficiency of different Upper-room UVC devices?
You’ve already read Zollner, so you know the base emitter efficiencies. Low-pressure mercury is around 40%, UVC LEDs aren’t really in use for upper-room GUV but they vary by wavelength and commercial stage and usually hover around 1-10%--though big caveat, efficiency is not the only important factor, and the issues UVC LEDs have with reliability are much more restrictive than the efficiency itself. Check out this 2021 DOE report for a less rosy outlook than Zollner’s—personally, I think the UVC LED industry is being far too optimistic in their predictions about UVC LEDs ever getting as efficient and cheap as white LEDs.
However, the louvers required to use cylindrical LP mercury lamps for upper-room GUV actually cause quite significant efficiency losses, down to 10%. Check out Kowalski (2009) Chapter 9--there hasn’t really been a ton of innovation since then.
How can we ensure that certain light wavelengths are harmless?
I won’t get into the specifics of what safety studies are considered the standard of evidence here. You do it the normal way—run the intervention (exposure to UV light), and check the endpoints of interest. Here is an extreme exposure study, here is a long-term exposure one in mice, there are many more—here’s a review, though it is from 2021. You would also need to think about environmental effects—for example, UV wavelengths below 240nm produce small (but increasing at lower wavelengths) amounts of ozone, which can then react with compounds in the air. UV can also degrade various materials, though shorter wavelengths don’t penetrate as deeply.
I do want to stress, however, that ‘harmless’ shouldn’t be the bar; nothing on this earth is ‘harmless’.
What is the current cost of UVC lamps, and how have prices trended in the past? What is the expected cost reduction?
I am going to assume you mean far-UVC lamps specifically. If you want to get one fast, $2500. If you want to get one that’s more reasonably priced, you can’t right now, but in the imaginable future you can get them from Acuity or Beacon. Good luck actually buying the Acuity one, though—I think they’re basically only B2B, and so is the source supplier Ushio who actually makes the lamp that all these fixture manufacturers are selling.
They cost only a couple hundred bucks or so to produce, so that’s probably what the cost floor is in the near to medium term future. Solid-state sources will have different scaling economics, but they don’t exist yet.
What light bulbs currently emit UVC light? How much energy do these lightbulbs require?
Basically just those Ushio bulbs. All the alternatives are worse or don’t exist yet. They consume 12 watts.
Could pathogens mutate and evolve under UVC, creating resistant bacteria?
Probably not, nobody in the UV world is really worried about this. GUV is not new. This is unlikely because the mechanism of inactivation is pretty fundamental to life (for both conventional GUV and far-UV), which is not how most antibiotics work. It is imaginable to me that constant exposure to far-UV specifically might promote biofilm-forming bacteria to form biofilms more, but I believe the Esvelt lab is currently looking at this.
What are the projected timelines for this technology? What is the projected timeline for setting up UVC lights in offices?
I don’t know what it means to have a projected timeline for a technology; that’s too general a question. If you want far-UV, you can contact FarUV Technologies. It will not be cheap but it will be faster than alternatives. If you want upper-room, there are plenty of suppliers you can ask. Here’s one I found on google. I know Kaleem Ahmid has done office installations in the past.
Could you elaborate on what you consider to be the specific marginal improvements of newer UV-C technologies?
We (the biosecurity team at Convergent Research) also have curated a number of specific, tarmac-ready far-UVC projects with teams and timelines attached, that range from a few 100k to a few million and address various bottlenecks in far-UVC adoption. We’ve got a whole catalogue of 1-pagers if any funders reading this are interested :P
I think I have a slightly different read on USGOV’s interest in far-UVC than Gavriel does. ~Every agency we’ve spoken to so far has gone “cool, but not our problem,” which I understand to mean that far-UVC is not being prioritized even within the already-deprioritized category of pandemic funding. Hell, even NIAID isn’t interested. This might change, of course...but no real idea of what might be a turning point. That being said, the CDC doesn’t seem actively unfriendly to the tech, so there is hope?
I think karma is awarded more by generality of subject matter than writing style per se? This post I spent a few hours on (including the hour I spent rereading the book) has x3 the karma of another post I put up around the same time that represents the outcome of about a year of part-time research.
And this is perfectly natural! Everyone on the EA forum has some reason to care about good writing, only some small subset of people on the EA forum have some reason to care about genetic engineering detection.
Very good to know! I’ve never heard of a US master’s program being paid.
I wonder if the interest in US-based PhDs has something to do with the larger US academic offerings—or maybe it’s just that unusually energetic people are both more likely to have early research experience and more likely to go to the US.
It’s actually quite remarkable—the way we teach writing to students is anti-useful. You could possibly do a worse job than we’re currently doing but I don’t immediately see how.
There’s something of a pepperoni airplane effect here. Everyone wants to think they’re Proust. I analogize it to the Picasso thing—you need to “learn the rules” before you can usefully take the training wheels off and start “breaking” them. Scare quotes intentional.
I disagree re: Murakami (haven’t read the others). I find him to be communicating extremely clearly. The actual book is full of specific examples of things that we think of as artful and indirect but that are actually bending the full force of themselves into conveying a very bright and specific concept.
I worked with Cass on the project mentioned and moving the needle on PPE seems much harder than I initially thought. The demand/scale thing is a real killer. There might be some solution here but it seems really muddled to me and I don’t think it’s throwing scrappy young engineers at it. Though of course the funding situation is different now too.
Good post—not least because I think this gives me insight into what the skeptics of EA-in-politics are thinking. I have a few responses:
It seems to me that two things are being conflated here: EAs individually running for office, and EA as a movement exercising political power. The latter, I agree, sounds like a terrible idea, for all the reasons you point out. But most of the arguments you bring up don’t apply to the former. My model of EA in electoral politics looks a lot more like individual EAs (who are themselves pre-selected for being unusually charismatic, well connected, or otherwise well-fit for politics) behaving as basically conventional politicians, being team players, but making EA issues their top priorities on the rare occasions when they come up. That’s perfectly consistent with pulling the rope sideways.
There’s a lot of middle ground between making EA as a movement a faction of the Democratic party and limiting EA involvement in politics to lobbying (though I think we should of course also be doing lobbying.)
As far as I know, advocates for EAs running for political office are much more excited for EAs to run as Republicans than as Democrats, for precisely the reasons you outline (and several others).
Finally, I want to comment that things like the Flynn campaign were always longshots, and that’s fine; this is a numbers game and is largely going to be paying off over decades, not years.
Completely agree, and I’ll second the impression that EA marketers primarily activate and recruit other EA marketers.
Another way to frame this dynamic -
If someone is making the claim to you that these problems are the overwhelmingly most important things to do with your life, then why aren’t they working on them? In a way it almost feels self-defeating. The people who best activated and motivated me to work directly on hard problems were people who themselves were working on them.
This is all anecdotal, of course, but I do think we need an epistemically healthier model of community building, for these and other reasons.
I think this is pretty much the case for many (especially non-STEM) fields in the US, too—my sense is that it’s a consequence of funding/competition.
I totally agree that they’re not useless—prestige/signalling in general is useful! And I think the median student is probably not going to be the kind of person who can fail out and still be wildly successful.
But, I think they are way overvalued. If the choice is between getting straight A’s and honor societies and awards, or getting B’s and also getting paid to do research, I think too many people choose the former over the latter.
Thanks for this wonderful comment! Let me try and address your questions:
I’m uncertain about the compensation/signaling/networking value for the research tech role. It’s not clear why it offers more returns than available to a few years in industry, even as a non-prestigious, entry level graduate.
I think actually a few years in industry is almost certainly better, though I think there’s a lot of overlap, and of course heavily depends on the field/industry. Major cruxes include I would say that if you have a substantial interest in later pursuing a PhD, that probably indicates being a research tech
The reason I recommend these roles is explicitly because they are easy to get. I remember how I felt nearing the end of my undergraduate physics degree. I had no idea how to even begin applying to industry jobs. It all seemed terribly scary and overwhelming; the returns on spending countless hours applying to jobs seemed low. If your counterfactual role to working as a research tech is going into industry, I would say you should probably go to industry.
But if your counterfactual sort of feels like it’s going to be just getting through classes, or going for unpaid internships, or sitting on the couch panicking about the future, then consider sending some emails to smart interesting people absolutely desperate for labor. Particularly, if you are contacting people at the university you currently attend, it’s pretty much part of the professor’s job to train you, even if they don’t really need labor.
If you’re still a student, an academic lab is also likely to be more flexible about letting you do interesting part time work. It’s all about accessibility.
In addition to the fact that many academic labs are exploitative (as the OP does touch on), I am concerned that even good and kind academic labs can give an off-color work experience/incentives/worldview, as I think they are not quite “real world environments”. I think a technician will get the worst of this while losing a lot of the positives?
“Technician” is sort of an imprecise title. I’ve designed and lead research projects as a “research assistant”/”research tech.” I find this is heavily dependent on the lab, and something that you should absolutely try to feel out during the interview. I’ll edit the post with some advice on this front.
The nice thing about roles like these is that they are relatively informal so that if it sucks it will not be that hard to just leave.
That said, while a lab is more like a “real world” environment than a class, this is a real weakness. Again, if you can easily get an industry job (or paid internship), that is probably a better choice, unless you are explicitly trying to boostrap yourself into a PhD without running the application gauntlet.
I don’t have a strong model of how this approach can lead to research that many in the EA community think is most valuable—”lead researchers with field leading potential”.
I’m not sure I entirely understand this point. Probably roles like this are not going to be in terribly directly impactful areas. I think the value of this approach is for bootstrapping yourself into an impactful role if you had a rough start—or more generally, doing better than just going to class and sitting around panicking about the future. I think this approach offers a good package deal for young EAs who don’t feel very effective or impressive and have absolutely no idea what to do next, or how.
This is great! You should consider doing a whole series of posts like this, especially focusing on common misconceptions—the time cost thing was something I fuzzily knew about, but hadn’t ever explicitly considered as the single major cost of a policy, such that money cost can be basically ignored.
I’m surprised that a policy only fails the cost-benefit test if the FTE cost five times more expensive than the benefit, and anything less ineffective than that is simply not a priority. What’s the reasoning behind that?