Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: Grokking “Semi-informative priors over AI timelines”, published by anson.ho on June 12, 2022 on The Effective Altruism Forum.

Notes:

I give visual explanations for Tom Davidson’s report, Semi-informative priors over AI timelines, and summarise the key assumptions and intuitions

The diagrams can be found here – you can click on the boxes to get linked to the part of the report that you’re interested in

Thanks to the Epoch team for feedback and support! Thanks especially to Jaime Sevilla and Tom Davidson for providing detailed feedback.

Executive Summary

The framework in Semi-informative priors over AI timelines assumes a model of AGI development which consists of a sequence of Bernoulli trials, i.e. it treats each calendar year as a “trial” at building AGI with constant probability p of succeeding.

However, we don’t know what this value of p is, so we use a generalisation of Laplace’s rule of succession to estimate P(AGI next year | no AGI yet). This is done by specifying a first-trial probability, the probability of successfully building AGI in the first year of AI research, together with the number of virtual successes, which tells us how quickly we should update our estimate for P(AGI next year | no AGI yet) based on evidence. The framework leans very heavily on the first-trial probability, which is determined using a subjective selection of reference classes (more here).

How much evidence we get depends on the number of trials that we see, which depends on the regime start-time – you can think of this as the time before which failure to develop AGI doesn’t tell us anything useful about the probability of success in later trials. For instance, we might think that 1956 (the year of the Dartmouth Conference) was the first year where people seriously started trying to build AGI, so the absence of AGI before 1956 isn’t very informative. If we think of each trial as a calendar year, then there have been 2021-1956 = 65 trials since the regime start-time, and we still haven’t developed AGI, so that’s 65 failed trials which we use to update P(AGI next year | no AGI yet), where “next year” now corresponds to 2022 rather than 1957.

But why should a trial correspond to a calendar year? The answer is that it doesn’t have to! In total, Davidson considers three candidate trial definitions:

Calendar-year trials: 1 trial = 1 calendar year

Compute trials: 1 trial = a 1% increase in the largest amount of compute used to develop an AI system to date

Researcher-year trials: 1 trial = a 1% increase in the total researcher-years so far

If we extend this reasoning, then we can predict the probability that AGI is built X years into the future. Davidson does this to predict P(AGI by 2036 | no AGI yet) as follows:

The idea is that this framework only incorporates a small amount of information based on observational evidence, giving “semi-informative priors” over AI timelines. This framework is shown in more detail below:

Since Davidson uses three different trial definitions, we actually get three of these diagrams!

All in all, Davidson uses this to get a central estimate of P(AGI by 2036 | no AGI yet)=8%, with the following cumulative probability function:

Motivation

One way of forecasting AI Timelines is to consider the inner workings of AI, guess what kinds of developments are the most important, and then generate a probability distribution over when Artificial General Intelligence (AGI) will be developed. This is the approach taken by Ajeya Cotra in Forecasting TAI with biological anchors, a really detailed draft report that draws analogy to the human brain to forecast when Transformative AI (TAI) will first be developed.

Tom Davidson’s report, Semi-informative priors over AI timelines, is also a detailed report forecasting AI timelines, but it takes a different approach to Cotra’s repor...

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: Training Trace Priors, published by Adam Jermyn on June 13, 2022 on The AI Alignment Forum.

I’m worried about scenarios involving deceptive models. We’ve failed at inner alignment so the model has goals that are not aligned with ours. It can somehow detect when it’s in training, and during training it pretends to share our goals. During deployment, surprise! The model paperclips the universe.

In this story deception is all about the model having hidden behaviors that never get triggered during training. There are many inputs where the model does what we want. These correspond to inputs where it ‘plays along’. Then on some rare input or an input outside of the training distribution it defects, triggering some catastrophic behavior that corresponds to maximizing an ulterior motive.

This suggests that we should be wary of models that have logic/subcircuits/states that are never used during training, and we can penalize models for having such hidden behaviors.

In this post I'm focused on examples of training trace priors, which are priors that make use of execution traces from the training distribution. I think these provide at least a few bits of optimization away from deception.

These are very much working notes, and feedback is very welcome.

One-Gate Trace Prior: Entropy

Consider a boolean circuit. The output of each gate is 0/1. One way to implement the entropy prior is to record traces of these outputs during training, and penalize the model for having strong bias between 0/1:

p∝exp(−∑i∑j∈[0,1]pi(j)lnpi(j))

Here pi is the observed fraction of the time gate i produces output j on the training distribution. We call this the one-gate trace prior because it relies on probabilities defined over single gates. This prior favors models where each gate has even odds of producing 0/1.

The intuition for why this might work is that deceptive models need some machinery specifying when and how they defect, and this machinery is never active during training. This is at the core of what makes them bad! If the defecting machinery activated during training we’d notice because the model would stop optimizing the outer goal, so we’d assign it a high loss and move on to better models.

Another framing is: It is inherently suspicious if a model has machinery that is rarely activated. That machinery wasn’t vetted during training, so we have no reason to trust it. If the model has a hidden love of paperclips, this is the natural place to hide it.

As a technical detail, in order to do useful computation the model may need access to some constant input wires, and these should be excluded from the prior.

Normalization

The one-gate trace prior isn’t normalized, and can’t be normalized without an additional prior over e.g. complexity. So in reasoning about it we need to remember that there will always be another piece to the prior setting the distribution over circuits. I’m agnostic to the details of that prior, and view the one-gate trace prior as a regularizer we apply to other (normalized) priors to make them safer.

Objection: Model Performance

This prior could cause a performance hit. The intuition is that sometimes a model will need to respond to unusual events even in the training distribution, so forcing the distribution over gate outputs to be as uniform as possible could make it hard to act in environments containing rare events.

In particular, this prior disfavors modularity. Suppose the model is meant to classify images. It might be natural have separate sub-circuits for recognizing cats and dogs. If cats are much more likely than dogs during training, our prior will disfavor models that separate that machinery out, because the ‘dog’ sub-circuit will rarely be active and the ‘cat’ sub-circuit will often be active.

If the pressure to perform well during training is strong enough this c...

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: Megaprojects for animals, published by James Ozden on June 13, 2022 on The Effective Altruism Forum.

Summary

Megaprojects ($100M+) for nonhuman animal welfare seem unlikely unless major new funding appears.

We still think it is worth laying out some ideas of comparatively smaller $10M megaprojects so that there are some shovel-ready projects if large funding arrives, and to perhaps entice funders to enter the space in a big way with a list of promising ideas.

This post lists some animal welfare projects that further research might reveal would cost-effectively absorb $10M+/year.

Where are the megaprojects for animals?

There has been a lot of buzz lately about megaprojects (projects that could deploy $100 million or more in funding per year). However, among the long lists of ideas presented so far (here, here, here, here, here, here) there are few ideas to reduce or end the suffering of animals. This invites the question: do we have similarly scalable and promising opportunities to cost-effectively help animals given additional resources?

Despite corporate campaigns being very cost-effective (Šimčikas 2019) and receiving a large share of EA animal funding (Kato 2022, Ozden 2021), it’s not obvious that this will continue indefinitely. How much can we scale up corporate campaigns and still achieve a sufficiently large total impact to offset any decline in cost-effectiveness? What are some potential large and effective projects if our current projects become fully funded or less cost-effective? What options do we have for other species and welfare asks? Can they scale?

Current spending by 2,000 organisations that include fighting factory farming as part of their work is just over $200 million per year (FAF 2021, 2020), and even less, about $40 million per year, is spent on “effective animal advocacy” (Ozden 2021 and Forristal 2020). Most (85%) of these organisations in the animal advocacy space operate on budgets less than $1M, and the largest operate on less than $20M (FAF 2020). So we’re very short of the level of spending of $100M+ megaprojects. However, it’s expected that as the EA movement grows, the resources available to EA-aligned animal advocacy will too (especially from increased funding in this space from Open Philanthropy and possibly FTX Community). Still, it’s unclear if/when this would reach the levels of funding available for human-focused longtermist projects. Given this practical limit, $10M might be a more reasonable operationalization of “mega” in this space.

Regardless, we still think it is worth laying out some ideas of megaprojects for animals so that there are some shovel-ready projects if funding scales up, and to perhaps entice funders to enter the space in a big way with promising ideas. Additionally, this list is a call to action to generate back-of-the-envelope calculations or detailed research into some of these ideas, to more rigorously identify highly scalable and cost-effective ways to help animals.

This list contains ideas from a range of reviews, contributors, and experts (though not all may endorse the broader idea of megaprojects or all ideas in this post). We haven’t yet taken the time to assess whether each of them could absorb $10M and at what level of cost-effectiveness. Partly this is because it’s unclear what the minimum bar should be. We are unsure which interventions can reach the cost-effectiveness of cage-free campaigns, especially when spending $10M per year (61-120 years of chicken life per average dollar spent/ ~7 animals spared a year of complete suffering). We perhaps should be willing to lower the cost-effectiveness bar because it’s plausible corporate campaigns are just a very rare good opportunity and it’s unfair to use it as a baseline (imagine if Givewell only granted to opportunities that were at least 5x AMF inst...

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: Nick Bostom - Sommar i P1 Radio Show, published by juliakarbing on June 12, 2022 on The Effective Altruism Forum.

Nick Bostrom participated in a Swedish radio show (Summer in P1) a few years ago, in which successful/famous people have one hour to talk about their lives, and choose the music to play. His talk is absolutely wonderful, and so I’ve translated it into English (imperfectly, but as well as I could..), and we’ve put it together into a video with the original audio with English subtitles :)

Would recommend watching the video with a few friends - it’s a fun experience to share :))

Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: Why all the fuss about recursive self-improvement?, published by So8res on June 12, 2022 on LessWrong.

This article was outlined by Nate Soares, inflated by Rob Bensinger, and then edited by Nate. Content warning: the tone of this post feels defensive to me. I don't generally enjoy writing in "defensive" mode, but I've had this argument thrice recently in surprising places, and so it seemed worth writing my thoughts up anyway.

In last year’s Ngo/Yudkowsky conversation, one of Richard’s big criticisms of Eliezer was, roughly, ‘Why the heck have you spent so much time focusing on recursive self-improvement? Is that not indicative of poor reasoning about AGI?’

I’ve heard similar criticisms of MIRI and FHI’s past focus on orthogonality and instrumental convergence: these notions seem obvious, so either MIRI and FHI must be totally confused about what the big central debates in AI alignment are, or they must have some very weird set of beliefs on which these notions are somehow super-relevant.

This seems to be a pretty common criticism of past-MIRI (and, similarly, of past-FHI); in the past month or so, I’ve heard it two other times while talking to other OpenAI and Open Phil people.

This argument looks misguided to me, and I hypothesize that a bunch of the misguidedness is coming from a simple failure to understand the relevant history.

I joined this field in 2013-2014, which is far from "early", but is early enough that I can attest that recursive self-improvement, orthogonality, etc. were geared towards a different argumentative environment, one dominated by claims like "AGI is impossible", "AGI won't be able to exceed humans by much", and "AGI will naturally be good".

A possible response: “Okay, but ‘sufficiently smart AGI will recursively self-improve’ and ‘AI isn’t automatically nice’ are still obvious. You should have just ignored the people who couldn’t immediately see this, and focused on the arguments that would be relevant to hypothetical savvy people in the future, once the latter joined in the discussion.”

I have some sympathy for this argument. Some considerations weighing against, though, are:

I think it makes more sense to filter on argument validity, rather than “obviousness”. What’s obvious varies a lot from individual to individual. If just about everyone talking about AGI is saying “obviously false” things (as was indeed the case in 2010), then it makes sense to at least try publicly writing up the obvious counter-arguments.

This seems to assume that the old arguments (e.g., in Superintelligence) didn’t work. In contrast, I think it’s quite plausible that “everyone with a drop of sense in them agrees with those arguments today” is true in large part because these propositions were explicitly laid out and argued for in the past. The claims we take as background now are the claims that were fought for by the old guard.

I think this argument overstates how many people in ML today grok the “obvious” points. E.g., based on a recent DeepMind Podcast episode, these sound like likely points of disagreement with David Silver.

But even if you think this was a strategic error, I still think it’s important to recognize that MIRI and FHI were arguing correctly against the mistaken views of the time, rather than arguing poorly against future views.

Recursive self-improvement

Why did past-MIRI talk so much about recursive self-improvement? Was it because Eliezer was super confident that humanity was going to get to AGI via the route of a seed AI that understands its own source code?

I doubt it. My read is that Eliezer did have "seed AI" as a top guess, back before the deep learning revolution. But I don't think that's the main source of all the discussion of recursive self-improvement in the period around 2008.

Rather, my read of the history is that MIRI was operatin...

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: You Don't Need To Justify Everything, published by ThomasWoodside on June 12, 2022 on The Effective Altruism Forum.

Thanks to Miranda Zhang, Teddy Tawil, and Matt Burtell for feedback on this post.

EA shares a key similarity with large family gatherings: you constantly feel the need to justify your life choices. You might have a really good justification, but you might also feel like you need to make something up.

I think it’s good for people to give explanations of their choices. After all, EA is about making the choices which will maximize your impact, and many of these choices are career-related. Your career is a major part of your life, and if you say you’re choosing a career path on the basis of EA principles, people in EA will sometimes ask you to give an explanation for why you're choosing that particular path. This isn't usually because they are trying to expose you, but because they genuinely want to hear your perspective and help you have the most impact that you can.

In fact, some people fill out applications just to speak with someone who can help you make life choices that are more justifiable (this is called 80,000 Hours advising).

A justification is supposed to indicate how you’ve thought about what you’re doing and why you’re doing it instead of any other options you had. In EA, this usually means why you think it’s likely to have the highest impact. Thinking carefully about your life choices is one way to be able to produce a legible justification, and thinking carefully is certainly an extremely important aspect of having impact. One way to practice thinking is to practice explaining.

But I’ve noticed that sometimes, people in EA try to justify seemingly everything.

Why are you ordering food instead of cooking? Because it saves me time so I can spend more time working. Why do you spend so much time gardening? Because it keeps me sane so I can do more one-on-ones with new group members. Why are you going to that party tonight? Because it will make me happier, which will help me produce more long-run undiscounted utility.

I think it can be very worthwhile to think about the major parts of your life on a semi-regular basis. Are they valuable to you? Do they help you achieve your goals (Julia Wise provides an excellent justification for having more than one)? Are they worth changing? If you think hard about decisions as large as your career, you will do much more good in expectation. The unexamined life is not worth many impact points.

However, even if you are an ardent utilitarian, it doesn’t make sense to make every single decision on the basis of a calculated utility for that decision. Many people have written great things about this, so I will not try to justify it fully here. But the basic gist is that act utilitarianism can be self-defeating: it can be time-consuming and emotionally difficult to decide every decision on the basis of the greater good, and that’s not even considering the risk of the unilateralist’s curse. Because of that, being a strict act utilitarian can reduce the utility of your acts.

But possible self-defeating dynamics aren’t the only issue. Another is that pressure to justify everything can cause people to come up with justifications after the fact. What do you do if you only heard about EA six months ago, and already made some decisions beforehand? What do you do if you just didn’t think very hard about working at a random tech company rather than something “more EA”? What if you just went to the party because you wanted to?

Justifying, in these cases, is also a way to get practice... in motivated reasoning. Why did you go to two parties last weekend? Maybe you just need two weekly parties to be happy enough to work. Why were you spending so much time trying to get an A in differential equations that you forgot to apply to t...

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: ALTER Israel - Mid-year 2022 Update, published by Davidmanheim on June 12, 2022 on The Effective Altruism Forum.

I’m excited to announce that the organization I started founding last year, the Association for Long Term Existence and Resilience, has been launched and funded, and we’re going to be working on a number of projects to build up an academic and policy focus in Israel on preventing catastrophic and existential risks and improving the trajectory of humanity for the long-term future. Given that, I wanted to give a public update about what has been happening, and invite anyone in Israel who we're not already in touch with to contact us.

Hiring

We recently hired an operations manager, Naham Shapiro, who will be doing much of the day-to-day work of running things.

We have made an offer to a researcher who will be doing mathematics research related to Vanessa Kosoy’s AI-safety research agenda, which was accepted. We are finalizing the details, but they will likely start in September or October.

Activities

We have incorporated officially as a non-profit in Israel, (Amutah,) and are slowly working our way through the byzantine mess of regulations and requirements for being able to actually function.

We are planning a conference for mid-October jointly with the Center for Humanities and AI, a new joint center between University of Haifa and the Technion, to introduce and promote AI safety research to Israeli academics and graduate students. Stuart Russell will be a keynote speaker, and we are working with a number of other AI-safety researchers internationally on the lineup. The goal is to encourage Israeli researchers' involvement in AI safety.

As a follow-up to the conference, we are hoping to offer some combination of programming, grants, and other support to encourage people in Israel to work on AI research, and connect them with the international community.

We have a website, which was designed in large part by Ben Schifman (Thanks!) and translated by Sariel Rabinowitz, with additional assistance and corrections from Joshua Fox and Vanessa Kosoy.

Funding

The organization’s initial startup funding came from the EA infrastructure fund, and David Manheim, the director, received a grant from the Long Term Future Fund to cover his salary and further startup costs of the organization for the initial 6 months, which has been used for the organization.

We have received a grant from FTXFF to fund most of our operations, regrants, policy work, and conferences over the next 18 months, as well as a separate recommended regrant from FTXFF to hire a researcher, mentioned above, to work on Vanessa Kosoy’s research agenda.

We have recently also tentatively received a substantial grant from the Survival and Flourishing Fund. We are currently planning to use this both to plan additional future conferences, and as an operating reserve to allow us to expand our work if and when we find particularly promising opportunities. We tentatively expect that this funding will be sufficient through 2024, and potentially longer, depending on whether we fund additional AI safety research.

Office

We are actively planning on splitting EA Israel’s office space, both for myself on some days, for our new operations manager, and for other future staff. Naham is working from there, and I have begun to meet at the current office and coworking space once a week, and will be working out an arrangement for sharing their future office space.

We're excited to see how all of this evolves, and again, feel free to contact us with questions, or feel free to comment below.

Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: How To: A Workshop (or anything), published by Duncan Sabien on June 12, 2022 on LessWrong.

Author's note: while the whole piece is a 55min read, there's a ~7min standalone essay at the top, and then a break and a shift into a different kind of content.

Back in February, I provided ops support and light-touch facilitation for a handful of mathematicians on a research retreat. Just prior to that, I helped MC a section of a two-week program for a batch of promising young EAs trying to refine and improve their long-term plans. Just after, I was the sole instructor for an abridged CFAR-esque applied rationality workshop for nine Circlers in southern California.

Two weeks ago, I solo led a non-abridged, 4.5-day workshop for a little over twenty participants, mostly EAs in their early twenties (thanks to Akash Wasil for laying the groundwork to make it happen).

By all accounts, my impact on each of these four very different events was strongly net positive. And as I reflected on each of them, it became clear to me that there was actually a fairly small number of things-going-on-in-my-brain that led to the good things that happened.

(Approximately just one, actually.)

I'd like to make that mental motion available to others, as I think it's useful for a wide range of people and pursuits, including ones that don't look like retreats or workshops at all.

So. How To: A Workshop (or anything, really).

While working as the curriculum director and workshop lead at CFAR circa 2017, I was attempting to hand over the reins to a young up-and-comer who asked me to write down what I was thinking about and focusing on in my role as MC. This was the result:

...this is a lot. My colleague was surprised, and a bit daunted.

And I do, in fact, believe that every item on this board is "pretty important," and they were all things that I was actively tracking or doing during approximately every workshop. I think that a workshop where any one of these things gets dropped is "noticeably" worse, i.e. staff or participants would be able to say "oh, yeah, that thing could have been better," even if the overall experience remains solid.

Pet theory, though:

The greatest value that a teacher or mentor can provide is usually in giving you discriminating principles that allow you to confidently ignore everything except the current most important thing to focus on.

(This claim by itself might be fully 1/3 of the value of this post, in expectation? I use something like this belief to select from among potential teachers and mentors and to better extract the wisdom I want from the teachers and mentors I have available.)

When you are first starting out in a new domain, there's often simply TOO MUCH INFORMATION coming in, and you often have no conceptual framework for digesting it, no buckets for the information to sort itself into, and no knowledge of how to effectively prioritize. You're trying to digest a dozen different things at once, which means you tend to fail at all of them for a while.

A good teacher can help by saying "forget everything except X for the next hour." And then, after an hour is up, "okay, now don't lose track of X, but start gradually diverting some of your spare attention to Y, as well."

And indeed, there is one single mental motion underlying all of the above; a single place where my attention stays during a workshop which, given sufficient context, generates all of the various items on the whiteboard. It generated all of them, in the moment, as I mentally traced my way through a Typical Workshop—I wasn't remembering what I usually did, or running through a mental checklist; I was running a function that outputs different answers at every step of the process of running a workshop.

If you, starting out in your Endeavors, can only juggle a single ball, it should be this one:

Do you kno...

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: Poorly-Aimed Death Rays, published by Thane Ruthenis on June 11, 2022 on LessWrong.

Alternate framing: Optimality is the tiger, and agents are its teeth.

Tonally relevant: Godzilla Strategies.

It's a problem when people think that a superintelligent AI will be just a volitionless tool that will do as told. But it's also a problem when people focus overly much on the story of "agency". When they imagine that all of the problems come from the AI "wanting" things, "thinking" things, and consequentializing all over the place about it. If only we could make it more of a volitionless tool! Then all of our problems would be solved. Because the problem is the AI using its power in clever ways with the deliberate intent to hurt us, right?

This, I feel, fails entirely to appreciate the sheer power of optimization, and how even the slightest failure to aim it properly, the slightest leakage of its energy in the wrong direction, for the briefest of moments, will be sufficient to wash us all away.

The problem isn't making a superintelligent system that wouldn't positively want to kill us. Accidentally killing us all is a natural property of superintelligence. The problem is making an AI that will deliberately spend a lot of effort on ensuring it's not killing us.

I find planet-destroying Death Rays to be a good analogy. Think the Death Star. Think

Imagine that you're an engineer employed by an... eccentric fellow. The guy has a volcano lair, weird aesthetic tastes, and a tendency to put words like "world" and "domination" one after another. You know the type.

One of his latest schemes is to blow up Jupiter. To that end, he'd had excavated a giant cavern underneath his volcano lair, dug a long cylindrical tunnel from that cavern to the surface, and ordered your team to build a beam weapon in that cavern and shoot it through the tunnel at Jupiter.

You're getting paid literal tons of money, so you don't complain (except about the payment logistics). You have a pretty good idea of how to do that project, too. There are these weird crystal things your team found lying around. If you poke one in a particular way, it releases a narrow energy beam which blows up anything it touches. The power of the beam scales superexponentially with the strength of the poke; you're pretty sure shooting one with a rifle will do the Jupiter-vanishing trick.

There's just one problem: aim. You can never quite predict which part of the crystal will emit the beam. It depends on where you poke it, but also on how hard you poke, with seemingly random results. And your employer is insistent that the Death Ray be fired from the cavern through the tunnel, not from space where it's less likely to hit important things, or something practical like that.

If you say that can't be done, your employer will just replace you with someone less... pessimistic.

So, here's your problem. How do you build a machine that uses one or more of these crystals in such a way that they fire a Death Ray through the tunnel at Jupiter, without hitting Earth and killing everyone?

You experiment with the crystals at non-Earth-destroying settings, trying to figure out how the beam is directed. You make a fair amount of progress! You're able to predict the beam's direction at the next power setting with 97% confidence!

When you fire it with Jupiter-destroying power, that slight margin of error causes the beam to be slightly misdirected. It grazes the tunnel, exploding Earth and killing everyone.

You fire the Death Ray at a lower, non-Earth-destroying setting that you know how to aim.

It hits Jupiter but fails to destroy it. Your employer is disappointed, and tells you to try again.

You line the cavern's walls and the tunnel with really good protective shielding.

The Death Ray grazes the tunnel, blows past the shielding, and kills everyone.

Yo...

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: New cause area: bivalve aquaculture, published by Brian Lui on June 12, 2022 on The Effective Altruism Forum.

Summary

Bivalve aquaculture improves animal welfare as a substitute for meat, fish, and other marine food.

Possibly improves economic welfare in developing countries.

Can be financially risky.

There is a huge opportunity to scale.

Improves water quality and reduces climate change because they are a carbon sink.

Second-order effects are likely to be beneficial.

Bivalve aquaculture means farming oysters, mussels, scallops, and other edible molluscs.

Bivalve aquaculture has multiple moderate benefits as summarized above. This means that it tends to fall between the cracks when EAs evaluate its effectiveness, because we often focus on maximizing a particular goal. Indeed, searching for “aquaculture” (not even bivalve aquaculture) returned very few “hits” on this forum.

This is an initial look into this cause area; approximately 10 hours was spent mainly on looking at sources.

Note: I may submit this to the Cause Exploration Prizes contest.

Animal welfare

Bivalves taste good (should be self-evident), and are healthy: “oysters, mussels, scallops, and clams are good for you. They’re loaded with protein, healthy fats, and minerals like iron and manganese.”

Eating bivalves causes less suffering than an equivalent amount of chickens, pigs, cows, and most other animals. Depending on what it substitutes for, it would also reduce crop farming and associated rodent/insect deaths, which are more sentient than bivalves.

Non-EAs are receptive to a proposal to substitute bivalves for other meat. They are not receptive to proposals to go vegetarian/vegan. Bivalves are also healthier than plant-based meat. Therefore, bivalves are the most effective way to reduce overall animal suffering.

Developing countries

Note: Epistemic confidence is low, and requires more research.

But it’s quite likely that bivalve aquaculture can have an enduring positive impact on developing countries. This report claims widespread benefits in Vietnam, for example: “There exists an opportunity to rapidly advance and sophisticate oyster aquaculture in Vietnam by exploring the full economic potential that has environmental, social, and sustainability benefits.”

Even if distribution isn’t developed enough for export, bivalve aquaculture can still provide highly nutritious food locally, as protein is generally rarer.

Financially risky

Note: Epistemic confidence is low, because these are comments on a preliminary request for feedback - more research needed.

Bivalve aquaculture has tail risk, e.g.:

“There are a ton of tail risks that can go wrong and wipe you out. Big one is disease, and the bigger the scale the harder that problem is. A lot of unknown unknowns.”

“Intensive aquaculture is tricky, diseases and anoxic conditions can wipe out entire crops.”

Counterintuitively, this is great for its attractiveness as a funding cause. It means that risk capital is valuable, and also means that the optimal risk-neutral level of bivalve aquaculture should be higher than it is now. It also means that we can be creative in how we approach spending money. For example, paying for insurance policies might be more impactful than investing directly.

Lots of room to scale

“Across the world there is an estimated 1.5 million sq km (579,000 sq miles) of coastline suitable for growing bivalve shellfish. According to Willer, developing just 1% of this could produce enough bivalves to fulfil the protein requirements of more than one billion people.”

It’s possible that a lot of this coastline is located in areas with bad governance. And not all of these places would be accessible or economically viable. But this is a tractable problem (scaling/new technologies) and shows that room to scale is unlikely to be the bottleneck.

Note: mo...