Director of Research at PAISRI
The people I know of who are best at mentorship are quite busy. As far as I can tell, they are already putting effort into mentoring and managing people. Mentorship and management also both directly trade off against other high value work they could be doing.There are people with more free time, but those people are also less obviously qualified to mentor people. You can (and probably should) have people across the EA landscape mentoring each other. But, you need to be realistic about how valuable this is, and how much it enables EA to scale.
The people I know of who are best at mentorship are quite busy. As far as I can tell, they are already putting effort into mentoring and managing people. Mentorship and management also both directly trade off against other high value work they could be doing.
There are people with more free time, but those people are also less obviously qualified to mentor people. You can (and probably should) have people across the EA landscape mentoring each other. But, you need to be realistic about how valuable this is, and how much it enables EA to scale.
Slight push back here in that I’ve seen plenty of folks who make good mentors but who wouldn’t be doing a lot of mentoring if not for systems in place to make that happen (because they stop doing it once they aren’t within whatever system was supporting their mentoring), which makes me think there’s a large supply of good mentors who just aren’t connected in ways that help them match with people to mentor.
This suggests a lot of the difficulty with having enough mentorship is that the best mentors need to not only be good at mentoring but also be good at starting the mentorship relationship. Plenty of people, it seems though, can be good mentors if someone does the matching part for them and creates the context between them and the mentees.
On a related but different note, I wish there was a way to combine conversations on cross-posts between EA Forum and LW. I really like the way AI Alignment Forum works with LW and wish EA Forum worked the same way.
I often make an adjacent point to folks, which is something like:
EA is not all one thing, just like the economy is not all one thing. Just as civilization as we know it doesn’t work unless we have people willing to do different things for different reasons, EA depends on different folks doing different things for different reasons to give us a rounded out basket of altruistic “goods”.
Like, if everyone thought saltine crackers were the best food and everyone competed to make the best saltines, we’d ultimately all be pretty disappointed that we had a mountain of amazing saltine crackers and literally nothing else, and so it makes sense even in the world where saltines really are the best food that generate the most benefit by their production that we instrumentally produce other things so we can enjoy our saltines in full.
I think the same is true of EA. I care a lot about AI x-risk and it’s what I focus on, but that doesn’t mean I think everyone should do the same. In fact, if they did, I’m not sure it would be so good, because then maybe we stop paying attention to other causes that, if we don’t address them, end up making trying to address AI risks moot. I’m always very glad to see folks working on things, even things I don’t personally think are worthwhile, both because of uncertainty about what is best and because there’s multiple dimensions along which it seems we can optimize (and would be happy if we did).
I think it’s worth saying that the context of “maximize paperclips” is not one where the person literally says the words “maximize paperclips” or something similar; this is instead an intuitive stand-in for building an AI capable of superhuman levels of optimization, such that if you set it the task, say via specifying a reward function, of creating an unbounded number of paperclips you’ll get it doing things you wouldn’t as a human do to maximize paperclips because humans have competing concerns and will stop when, say, they’d have to kill themselves or their loved ones to make more paperclips.
The objection seems predicated on interpretation of human language, which is aside the primary point. That is, you could address all the human language interpretation issues and we’d still have an alignment problem, it just might not look literally like building a paperclip maximizer if someone asks the AI to make a lot of paperclips.
I wrote about something similar about a year ago: https://forum.effectivealtruism.org/posts/Z94vr6ighvDBXmrRC/illegible-impact-is-still-impact
There’s a lot to unpack in that tweet. I think something is going on like:
fighting about who is really the most virtuous
being upset people aren’t more focused on the things you think are important
being upset that people claim status by doing things you can’t or won’t do
being jealous people are doing good doing things you aren’t/can’t/won’t do
spillover of culture war stuff going on in SF
None of it looks like a real criticism of EA, but rather of lots of other things EA just happens to be adjacent to.
Doesn’t mean it doesn’t have to be addressed or isn’t an issue, but I think also worth keeping these kinds of criticisms in context.
I find others answers about what the actual low resolution version of EA they see in the wild fascinating.
I go with the classic and if people ask I give them a three word answer: “doing good better”.
If they ask for more, it’s something like: “People want to do good in the world, and some good doing efforts produce better outcomes than others. EA is about figuring out how to get the best outcomes (or the largest positive impact) for time/money/effort relative to what a person thinks is important.”
I realize this is a total tangent to the point of your post, but I feel you’re giving short-shrift here to continental philosophy.
If it were only about writing style I’d say fair: continental philosophy has chosen a style of writing that resembles that used in other traditions to try to avoid over-simplifying and not compressing understanding down into just a few words that are easily misunderstood. Whereas you see unclear writing, I see a desperate attempt to say anything detailed about reality without accidentally pointing in the wrong direction.
This is not to say that there aren’t bad continental philosophers who hide behind this method to say nothing, but I think it’s unfair to complain about it just because it’s hard to understand and takes a lot of effort to suss out what is being said.
As to the central confusion you bring up, the unfortunate thing is that the worst argument in the world is technically correct, we can’t know things as they are in themselves, only as we perceive them to be, i.e. there is no view from nowhere. Where it’s wrong is thinking that just because we always know the world from some vantage point that trying to understanding anything is pointless and any belief is equally useful. It is can both be true that there is no objective way that things are and that some ways of trying to understand reality do better at helping us predict reality than others.
I think the confusion that the worst argument in the world immediately implies we can’t know anything useful comes from only seeing that the map is not itself the territory but not also seeing that the map is embedded in the territory (no Cartesian dualism).
I think this is often non-explicit in most discussions of morality/ethics/what-people-should-do. It seems common for people to conflate “actions that are bad because it ruins ability to coordinate” and “actions that are bad because empathy and/or principles tell me they are.”
I think it’s worth challenging the idea that this conflation is actually an issue with ethics.
Although it’s true that things like coordination mechanisms and compassion are not literally the same thing and can have expressions that try to isolate themselves from each other (cf. market economies and prayer) and so things that are bad because they break coordination mechanisms or because they don’t express compassion are not bad for exactly the same reasons, this need not mean there is not something deeper going on that ties them together.
I think this is why there tends to be a focus on meta-ethics among philosophers of ethics rather than directly trying to figure out what people should do, even when setting meta-ethical uncertainty aside. There’s some notion of badness or undesireableness (and conversely goodness or desirableness) that powers both of these, and so they are both different expressions of this same underlying phenomenon. So we can reasonably ties these two approaches together by looking at this question of what makes something seem good or bad to us, and simply consider these different domains over which we consider how we make good or bad things happen.
As to what good and bad mean, well, that’s a larger discussion. My best theory is that in humans it’s rooted in prediction error plus some evolved affinities, but this is an ongoing place where folks are trying to figure out what good and bad mean beyond our intuitive sense that something is good or bad.
Weird, that sounds strange to me because I don’t really regret things since I couldn’t have done anything better than what I did under the circumstances or else I would have done that, so the idea of regret awakening compassion feels very alien. Guilt seems more clear cut to me, because I can do my best but my best may not be good enough and I may be culpable for the suffering of others as a result, perhaps through insufficient compassion.
These cases seem not at all analogous to me because of the differing amount of uncertainty in each.
In the case of the drowning child, you presumably have high certainty that the child is going to die. The case is clear cut in that way.
In the case of the distant commotion on an autumn walk, it’s just that, a distant commotion. As the walker, you have no knowledge about what it is and whether or not you could do anything. That you later learn you could have done something might lead you to experience regret, but in the moment you lacked information to make it clear you should have investigated. I think this entirely accounts for the difference in feeling about the two cases, and eliminates the power of the second case.
In the second case, any imposition on the walker to do anything hinges on their knowledge of what the result of the commotion will be. Given the uncertainty, you might reasonably conclude in the moment that it is better to avoid the commotion, maybe because you might do more harm than good by investigating.
Further, this isn’t a case of negligence, where you failing to respond to the commotion makes you complicit in the harm, because you seem to have no responsibility to the machinery or the conditions by which the man came to be pinned under it. Instead it seems to be a case where you are morally neutral throughout because of your lack of knowledge, and your lack of active effort to avoid gaining knowledge that would otherwise make you complicit by trying to avoid becoming morally culpable. That is not the case here and so your example seems to lack the necessary conditions to make the point.
Could the seeming contradiction be resolved by greater specificity of statements?
For example, rather than abandoning “Everyone should sell everything that begins with a ‘C’, but nothing that begins with an ‘A’.” as a norm, we might realize we underspecified it to begin with and really meant “Everyone should sell everything that is called by a word in English that begins with a ‘C’, but nothing that begins with an ‘A’ in English.”. We could get even more specific if objections remained until we were not at risk of under specifying what we mean and suffering from relativity.
In the same vein, maybe the contradiction of the through experiment could be resolved by being more specific and including more context about the world. For example, cf. this attempt at thinking about preferences as conditioned on the entire state of the world. Maybe the same sort of technique could be applied here.
Where do you work, and what do you do?
I’m a software engineer at Plaid working on the Infrastructure team. My main project is leading our internal observability efforts.
What are some things you’ve worked on that you consider impactful?
In terms of EA impact at my current job, not much. I view this as an earning to give situation where I’m taking my expertise as a software engineer and turning it into donations. I think there’s some argument that Plaid has positive impact on the world by enabling lots of new financial applications built on our APIs, thereby increasing access to financial resources for those who historically had the least access to them. But I don’t work directly on that stuff, instead working on the things that enables the org to carry out its mission.
I will say I considered some other jobs, say working at Facebook or continuing to work on ads as I had been doing, and although the mission was not the primary reason I chose Plaid it is nice that I don’t worry I might work on something that harms the world.
What are a few ways in which you bring EA ideas/mindsets to your current job?
I often use the TIN framework informally in work and elsewhere in life. It’s sort of baked into my soul to think about tractability, impact, and neglectedness when thinking about what to do. Plaid has a big internal focus on the idea of impact, including having a positive impact on the world, and of course as an engineer there’s plenty of focus on doing things that are tractable (possible). Neglectedness considerations mostly show up in what I personally choose to work on: I look for things where I can have impact, that are tractable, and that are being neglected by others such that I can make things better in ways that are currently not being pursued. In a growing organization this is easy, because there’s often a lot of stuff we’d do if someone had more time to do it, so then it largely becomes a question of prioritizing between different neglected issues.
I think this holds true in more traditionally “quantitative” fields, too, because often things can be useful or not depending on how they are framed such that without the proper framing good numbers don’t matter because they are measuring the right thing.
This seems to suggest that a lot of what makes quantitative research successful also makes qualitative research successful, and so we should expect any extent to which expertise matters in quantitative fields to matter in qualitative fields (although I think this mostly points at the quant/qual distinction being a very fuzzy one that is only relevant along certain dimensions).
Jonas also mentioned to me that EA Funds is considering offering Donor-Advised Funds that could grant to individuals as long as there’s a clear charitable benefit. If implemented, this would also allow donors to provide tax-deductible support to individuals.
This is pretty exciting to me. Without going into too much detail, I expect to have a large amount of money to donate in the near future, and LTF is basically the best option I know of (in terms of giving based on what I most want to give to) for the bulk of that money short of having the ability to do exactly this. I’d still want LTF as a fall back for funds I couldn’t figure out how to better allocate myself, but the need for tax deductibility limits my options today (though, yes, there are donor lotteries).
LTF covers a lot of ground. How do you prioritize between different cause areas within the general theme of bettering the long term future?
How much room for additional funding does LTF have? Do you have an estimate of how much money you could take on and still achieve your same ROI on the marginal dollar donated?
Do you have any plans to become more risk tolerant?
Without getting too much into details, I disagree with some things you’ve chosen not to fund, and as an outsider view it as being too unwilling to take risks on projects, especially projects where you don’t know the requesters well, and truly pursue a hits-based model. I really like some of the big bets you’ve taken in the past on, for example, funding people doing independent research who then produce what I consider useful or interesting results, but I’m somewhat hesitant around donating to LTF because I’m not sure it takes enough risks that it represents a clearly better choice for someone like me whose fairly risk tolerant with their donations than donating to other established projects or just donating directly (but this has the disadvantage of making it hard for me to give something like seed funding and still get tax advantages).
I’m being strategic in 2020 and shifting much of my giving for it into 2021 because I expect a windfall, but here’s where I chose to give this year:
AI Safety Support
I think the work Linda (and now JJ) are doing is great and is woefully underfunded. I would give them more sooner but I have to shift that into 2021. They’ve had some trouble getting funding from more established sources for reasons I don’t endorse but don’t want to go into here, and I think giving to them now is especially high leverage to help AISS bootstrap.
I’ll be giving $5k soon and plan to donate more once the funds to do so are unlocked.
Read Linda’s post about AISS for more details.
MIRI keeps doing great work on AI safety, and I’ve been especially impressed with Scott and Abram in the last couple years. I’ve cut back on some of my funding to MIRI because I view them as less neglected now relative to other things I could fund, but I continue to support them via Amazon Smile.
This feels a little bit like paying for utilities I use, but I get a lot of value out of Wikipedia and think everyone who can should donate $5 or $10 to them. It also seems generally useful for maintaining and improving a source of facts in a world that increasingly uncertain about what facts even are.
I have a cryonics contract with Alcor, and I pay annual dues to them. Most of this is counted as charitable giving.
Bay Zen Center
This isn’t really EA giving, but it is charitable giving to a religious organization (full disclosure, I’m on the board of the Center). They get about 2% of my income. Listed for completeness.
Long Term Future Fund
LTF is generally aligned with my giving priorities and will get my marginal additional funding I don’t have a better idea about how to allocate.
Long term my objective is to donate 30-50% of my income (limited by tax incentives and marginal value of money until I resolve some large outstanding expenses), but today it’s closer to 5%.
I’ve had RSI in the past, but not from typing, but instead from repetitive motions loading paper into a machine for scanning. I didn’t need to see a doctor about it, and addressing it was ultimately pretty straight forward and I was able to keep doing the job that caused it while I recovered. Things I did:
wore a stabilizing wrist brace to alleviate the strain on my wrist that was causing pain, even when I was not engaged in an activity that would necessarily cause pain
payed attention to and changed my motions to reduce wrist strain
rearranged my work so I had more breaks and less long periods of continually performing the motion (I had other job responsibilities so it was easy to interleave breaks from one thing with work on another)
It’s now more than 10 years since I developed RSI, and maybe 4 years since I have needed the wrist brace (my need for it rapidly decreased once I left the job). I think never needing it correlated with increased strength, specifically from indoor rock climbing and related conditioning.