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: What's the Theory of Change/Theory of Victory for Farmed Animal Welfare?, published by Linch on December 1, 2021 on The Effective Altruism Forum.

Similar to Mark Xu’s concerns about Meta-EA, and Lizka’s on IIDM, I see lots of work in farmed animal welfare, with no clear explicated path/Theory of Victory.

Obviously this could just be due to my own ignorance, but often I find asking dumb questions to be helpful in both learning and education, as I’m sure others have been confused about the same thing.

So I’d be interested in seeing a sketch of what Theory of Victory people are either explicitly or implicitly drawing on.

“Victory” here can mean many things, but in the spirit of high ambition, I’m interested in operationalizations like a) the complete end of factory farming or b) enough success to a point where all or almost all of the current EA farmed animal welfare folks willingly choose to leave the field to work on other projects (e.g., biosecurity, broad longtermism, s-risks, wild animal welfare, etc).

Some toy examples I can imagine (Note these are very toy models, not to be taken too seriously):1. (slow/no Transformative AI (TAI) baseline, Western hegemony). Assuming cost-competitive cultured or plant-based meat is plausible. We work hard on speeding up meat alternative timelines, while simultaneously pushing for broad support for welfare and antipathy towards factory farming companies. If/once meat alternatives are cheap and good enough, this in conjunction with large-scale disapproval of factory farming is enough to swing more and more legislation in our favor, eventually culminating in more and more countries banning factory farming outright. We all eat alternative proteins, factory farming legally ends.

2. (medium-term TAI) Something directly relating to AI. Assume mediumish timelines and good AI safety progress. If there is enough elite support for animal welfare, especially for people in decisionmaking positions for TAI (e.g. major AI scientists, all the heads of labs, AI gov folks), that any aligned plan for the future would not involve large-scale animal suffering. Magic happens, animals don’t suffer.

3. (slow/no TAI, technological/economic cornucopia in other domains). TAI itself turned out to be very hard, but progress in ML and other scientific domains accelerate. Plant-based meat stalls out at a high price point, and cultured meat continues to look too expensive to be directly competitive. But we do a really strong job of moral circle expansion, such that once the entire world is rich enough, people voluntarily stop eating meat and eat just as tasty but much more expensive alternatives.

4. (slow/no TAI, business-as-usual technological growth, no strong priors on hegemony). Plant-based and cultured meat fails to pan out, but EAs systematically explore a wide range of alternative protein options, eventually coming across several replacements that systematically go down in price over decades. Concurrently, people work on corporate campaigns, welfare regulations, carbon taxes, and other measures that systematically increases the price of meat. Eventually, the two curves cross, and a combination of economic and political pressures slowly phase out factory farming.

Of course, these models are far from detailed enough. That’s why they aren’t real Theories of Victory! And in some ways they are too detailed (I don’t believe in any of them that much). But I offer this question here in the hopes that others have substantially more detailed theories, especially people working in leadership organizations in EAA, or researchers working on animal welfare prioritization questions.

Thanks to Lizka for first bringing this question to my attention, Saulius for Slack prompts in this general direction, and Neil_Dullaghan for verbally brainstorming this question with me.

N...

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: What's the Theory of Change/Theory of Victory for Farmed Animal Welfare?, published by Linch on December 1, 2021 on The Effective Altruism Forum.

Similar to Mark Xu’s concerns about Meta-EA, and Lizka’s on IIDM, I see lots of work in farmed animal welfare, with no clear explicated path/Theory of Victory.

Obviously this could just be due to my own ignorance, but often I find asking dumb questions to be helpful in both learning and education, as I’m sure others have been confused about the same thing.

So I’d be interested in seeing a sketch of what Theory of Victory people are either explicitly or implicitly drawing on.

“Victory” here can mean many things, but in the spirit of high ambition, I’m interested in operationalizations like a) the complete end of factory farming or b) enough success to a point where all or almost all of the current EA farmed animal welfare folks willingly choose to leave the field to work on other projects (e.g., biosecurity, broad longtermism, s-risks, wild animal welfare, etc).

Some toy examples I can imagine (Note these are very toy models, not to be taken too seriously):1. (slow/no Transformative AI (TAI) baseline, Western hegemony). Assuming cost-competitive cultured or plant-based meat is plausible. We work hard on speeding up meat alternative timelines, while simultaneously pushing for broad support for welfare and antipathy towards factory farming companies. If/once meat alternatives are cheap and good enough, this in conjunction with large-scale disapproval of factory farming is enough to swing more and more legislation in our favor, eventually culminating in more and more countries banning factory farming outright. We all eat alternative proteins, factory farming legally ends.

2. (medium-term TAI) Something directly relating to AI. Assume mediumish timelines and good AI safety progress. If there is enough elite support for animal welfare, especially for people in decisionmaking positions for TAI (e.g. major AI scientists, all the heads of labs, AI gov folks), that any aligned plan for the future would not involve large-scale animal suffering. Magic happens, animals don’t suffer.

3. (slow/no TAI, technological/economic cornucopia in other domains). TAI itself turned out to be very hard, but progress in ML and other scientific domains accelerate. Plant-based meat stalls out at a high price point, and cultured meat continues to look too expensive to be directly competitive. But we do a really strong job of moral circle expansion, such that once the entire world is rich enough, people voluntarily stop eating meat and eat just as tasty but much more expensive alternatives.

4. (slow/no TAI, business-as-usual technological growth, no strong priors on hegemony). Plant-based and cultured meat fails to pan out, but EAs systematically explore a wide range of alternative protein options, eventually coming across several replacements that systematically go down in price over decades. Concurrently, people work on corporate campaigns, welfare regulations, carbon taxes, and other measures that systematically increases the price of meat. Eventually, the two curves cross, and a combination of economic and political pressures slowly phase out factory farming.

Of course, these models are far from detailed enough. That’s why they aren’t real Theories of Victory! And in some ways they are too detailed (I don’t believe in any of them that much). But I offer this question here in the hopes that others have substantially more detailed theories, especially people working in leadership organizations in EAA, or researchers working on animal welfare prioritization questions.

Thanks to Lizka for first bringing this question to my attention, Saulius for Slack prompts in this general direction, and Neil_Dullaghan for verbally brainstorming this question with me.

N...

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: Hypotheses about Finding Knowledge and One-Shot Causal Entanglements, published by Jemist on December 1, 2021 on The AI Alignment Forum.

Epistemic status: my own thoughts I've thought up in my own time. They may be quite or very wrong! I am likely not the first person to come to these ideas. All of my main points here are just hypotheses which I've come to by the reasoning stated below. Most of it is informal mathematical arguments about likely phenomena and none is rigorous proof. I might investigate them if I had the time/money/programming skills. Lots of my hypotheses are really long and difficult-to-parse sentences.

What is knowledge?

I think this question is bad.

It's too great of a challenge. It asks us (implicitly) for a mathematically rigorous definition which fits all of our human feelings about a very loaded word. This is often a doomed endeavour from the start, as human intuitions don't neatly map onto logic. Also, humans might disagree on what things count as or do not count as knowledge. So let's attempt to right this wrong question:

Imagine a given system is described as "knowing" something. What is the process that leads to the accumulation of said knowledge likely to look like?

I think this is much better.

We limit ourselves to systems which can definitely be said to "know" something. This allows us to pick a starting point. This might be a human, GPT-3, or a neural network which can tell apart dogs and fish. In fact this will be my go-to answer for the future. We also don't need to perfectly specify the process which generates knowledge all at once, only comment on its likely properties.

Properties of "Learning"

Say we have a very general system, with parameters θ, with t representing time during learning. Let's say they're initialized as θ0 according to some random distribution. Now it interacts with the dataset which we will represent with X, taken from some distribution over possible datasets. The learning process will update θ0, so we can represent the parameters the parameters after some amount of time as θ(θ0; X; t). This reminds us that the set of parameters depends on three things: the initial parameters, the dataset, and the amount of training.

Consider θ(θ0; X; 0). This is trivially equal to θ0, and so it depends only on the choice of θ0. The dataset has had no chance to affect the parameters in any way.

So what about as t→∞? We would expect that θ∞(θ0; X)=θ(θ0; X; ∞) depends mostly on the choice of X and much less strongly on θ0. There will presumably be some dependency on initial conditions, especially for very complex models like a big neural network with many local minima. But mostly it's ω which influences θ.

So far this is just writing out basic sequences stuff. To make a map of the city you have to look at it, and to learn your model has to causally entangle itself with the dataset. But let's think about what happens when ω is slightly different.

Changes in the world

So far we've represented the whole dataset with a single letter X, as if it were just a number or something. But in reality it will have many, many independent parts. Most datasets which are used as inputs to learning processes are also highly structured.

Consider the dog-fish discriminator, trained on the dataset Xdog/fish. The system θ∞(θ0; Xdog/fish) could be said to have "knowledge" that "dogs have two eyes". One thing this means if we instead fed it an X which was identical except every dog had three eyes (TED) then the final values of θ would be different. The same is true of facts like "fish have scales", "dogs have one tail". We could express this as follows:

θ∞(θ0; Xdog/fish+ΔXTED)

Where ΔXTED is the modification of "photoshopping the dogs to have three eyes". We now have:

θ∞(θ0; Xdog/fish+ΔXTED)=θ∞(θ0; Xdog/fish)+Δθ∞(θ0; Xdog/fish; ΔXTED)

Now let's consider how...

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: Hypotheses about Finding Knowledge and One-Shot Causal Entanglements, published by Jemist on December 1, 2021 on The AI Alignment Forum.

Epistemic status: my own thoughts I've thought up in my own time. They may be quite or very wrong! I am likely not the first person to come to these ideas. All of my main points here are just hypotheses which I've come to by the reasoning stated below. Most of it is informal mathematical arguments about likely phenomena and none is rigorous proof. I might investigate them if I had the time/money/programming skills. Lots of my hypotheses are really long and difficult-to-parse sentences.

What is knowledge?

I think this question is bad.

It's too great of a challenge. It asks us (implicitly) for a mathematically rigorous definition which fits all of our human feelings about a very loaded word. This is often a doomed endeavour from the start, as human intuitions don't neatly map onto logic. Also, humans might disagree on what things count as or do not count as knowledge. So let's attempt to right this wrong question:

Imagine a given system is described as "knowing" something. What is the process that leads to the accumulation of said knowledge likely to look like?

I think this is much better.

We limit ourselves to systems which can definitely be said to "know" something. This allows us to pick a starting point. This might be a human, GPT-3, or a neural network which can tell apart dogs and fish. In fact this will be my go-to answer for the future. We also don't need to perfectly specify the process which generates knowledge all at once, only comment on its likely properties.

Properties of "Learning"

Say we have a very general system, with parameters θ, with t representing time during learning. Let's say they're initialized as θ0 according to some random distribution. Now it interacts with the dataset which we will represent with X, taken from some distribution over possible datasets. The learning process will update θ0, so we can represent the parameters the parameters after some amount of time as θ(θ0; X; t). This reminds us that the set of parameters depends on three things: the initial parameters, the dataset, and the amount of training.

Consider θ(θ0; X; 0). This is trivially equal to θ0, and so it depends only on the choice of θ0. The dataset has had no chance to affect the parameters in any way.

So what about as t→∞? We would expect that θ∞(θ0; X)=θ(θ0; X; ∞) depends mostly on the choice of X and much less strongly on θ0. There will presumably be some dependency on initial conditions, especially for very complex models like a big neural network with many local minima. But mostly it's ω which influences θ.

So far this is just writing out basic sequences stuff. To make a map of the city you have to look at it, and to learn your model has to causally entangle itself with the dataset. But let's think about what happens when ω is slightly different.

Changes in the world

So far we've represented the whole dataset with a single letter X, as if it were just a number or something. But in reality it will have many, many independent parts. Most datasets which are used as inputs to learning processes are also highly structured.

Consider the dog-fish discriminator, trained on the dataset Xdog/fish. The system θ∞(θ0; Xdog/fish) could be said to have "knowledge" that "dogs have two eyes". One thing this means if we instead fed it an X which was identical except every dog had three eyes (TED) then the final values of θ would be different. The same is true of facts like "fish have scales", "dogs have one tail". We could express this as follows:

θ∞(θ0; Xdog/fish+ΔXTED)

Where ΔXTED is the modification of "photoshopping the dogs to have three eyes". We now have:

θ∞(θ0; Xdog/fish+ΔXTED)=θ∞(θ0; Xdog/fish)+Δθ∞(θ0; Xdog/fish; ΔXTED)

Now let's consider how...

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: Creating Individual Connections via the Forum, published by Ben West on November 30, 2021 on The Effective Altruism Forum.

Summary

I believe that creating personal connections is one of the most valuable things CEA can do.

There's some evidence that the Forum could be useful in building these connections.

In this document, I list possible use cases/types of connections people might want.

I would appreciate any feedback readers have on these use cases, including "I was too busy to read your document but think being connected to someone in XYZ way would be useful".

This document should be taken as "initial ideas Ben has". CEA is not committing to implement any of these ideas, and my colleagues at CEA have varying levels of agreement with these ideas. That being said, I would be surprised if CEA does not put meaningful effort towards building individual connections through our online properties over the next year in some manner.

I am grateful for conversations with many EA group organizers, the EA Hub team, my coworkers at CEA, and other community builders about these ideas. I generally am not the person who deserves the most credit for each of these ideas.

Motivation

Individual connections are a major reason why people get involved with EA, and why they stay involved with EA. For example, the most recent EA survey found that personal contacts were the most-cited reason why people got involved with EA, by a substantial margin.

Several platforms for building and maintaining individual connections between EAs already exist. The most recent EA survey listed the following online sources through which respondents made new interesting and valuable connections:

EA Facebook groups (8.8% of respondents)

International EA social events (8.7% of respondents)

EA Forum (5.9% of respondents)

LessWrong (5.3% of respondents)

EA Hub (1.9% of respondents)

I suspect that Twitter is responsible for a similar number of connections as the other ones listed, though it was not on the survey. There have also been smaller projects not listed on this survey, such as EA Pen Pals.

It's interesting to note that the EA Forum is responsible for a similar number of connections as Facebook, despite having very little functionality that is targeted towards instigating these connections.

Classic new product advice is that you should look for features your users like despite them being very poorly made, because if they're willing to use a poorly made version of your product, they will probably love a well-made version. It seems like we have initial evidence of people making connections using features not well-made to facilitate this (comments, private messages).

In summary: We have evidence that

Personal connections are extremely valuable, and

People are interested in forming these connections on the Forum

What types of connections are people looking for?

It's possible that we should just naïvely build out a list of EAs and let users filter it however they want. My guess, though, is that it's better to be targeted towards some specific use cases. Here are the most common use cases I've heard from others in the community, and my thoughts on how to provide solutions.

New person has new person questions

Most people have some questions when they first encounter EA. Some standard ones include:

Where should I donate?

My career so far has been in (whatever). How can I change that to be more impactful?

Basic questions about longtermism, AI safety, animal welfare, etc.

Currently, answering these questions is mostly done by group organizers, and my understanding from speaking with them is that these questions generally don't require a large amount of expertise to answer: it's helpful to simply point these newcomers to 80,000 hours for career advice, some key blog posts for questions about longtermism, etc.

A possible...

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: Creating Individual Connections via the Forum, published by Ben West on November 30, 2021 on The Effective Altruism Forum.

Summary

I believe that creating personal connections is one of the most valuable things CEA can do.

There's some evidence that the Forum could be useful in building these connections.

In this document, I list possible use cases/types of connections people might want.

I would appreciate any feedback readers have on these use cases, including "I was too busy to read your document but think being connected to someone in XYZ way would be useful".

This document should be taken as "initial ideas Ben has". CEA is not committing to implement any of these ideas, and my colleagues at CEA have varying levels of agreement with these ideas. That being said, I would be surprised if CEA does not put meaningful effort towards building individual connections through our online properties over the next year in some manner.

I am grateful for conversations with many EA group organizers, the EA Hub team, my coworkers at CEA, and other community builders about these ideas. I generally am not the person who deserves the most credit for each of these ideas.

Motivation

Individual connections are a major reason why people get involved with EA, and why they stay involved with EA. For example, the most recent EA survey found that personal contacts were the most-cited reason why people got involved with EA, by a substantial margin.

Several platforms for building and maintaining individual connections between EAs already exist. The most recent EA survey listed the following online sources through which respondents made new interesting and valuable connections:

EA Facebook groups (8.8% of respondents)

International EA social events (8.7% of respondents)

EA Forum (5.9% of respondents)

LessWrong (5.3% of respondents)

EA Hub (1.9% of respondents)

I suspect that Twitter is responsible for a similar number of connections as the other ones listed, though it was not on the survey. There have also been smaller projects not listed on this survey, such as EA Pen Pals.

It's interesting to note that the EA Forum is responsible for a similar number of connections as Facebook, despite having very little functionality that is targeted towards instigating these connections.

Classic new product advice is that you should look for features your users like despite them being very poorly made, because if they're willing to use a poorly made version of your product, they will probably love a well-made version. It seems like we have initial evidence of people making connections using features not well-made to facilitate this (comments, private messages).

In summary: We have evidence that

Personal connections are extremely valuable, and

People are interested in forming these connections on the Forum

What types of connections are people looking for?

It's possible that we should just naïvely build out a list of EAs and let users filter it however they want. My guess, though, is that it's better to be targeted towards some specific use cases. Here are the most common use cases I've heard from others in the community, and my thoughts on how to provide solutions.

New person has new person questions

Most people have some questions when they first encounter EA. Some standard ones include:

Where should I donate?

My career so far has been in (whatever). How can I change that to be more impactful?

Basic questions about longtermism, AI safety, animal welfare, etc.

Currently, answering these questions is mostly done by group organizers, and my understanding from speaking with them is that these questions generally don't require a large amount of expertise to answer: it's helpful to simply point these newcomers to 80,000 hours for career advice, some key blog posts for questions about longtermism, etc.

A possible...

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: Infra-Bayesian physicalism: a formal theory of naturalized induction, published by Vanessa Kosoy on November 30, 2021 on LessWrong.

This is joint work by Vanessa Kosoy and Alexander "Diffractor" Appel. For the proofs, see 1 and 2.

TLDR: We present a new formal decision theory that realizes

naturalized induction. Our agents reason in terms of infra-Bayesian hypotheses, the domain of which is the cartesian product of computations and physical states, where the ontology of "physical states" may vary from one hypothesis to another. The key mathematical building block is the "bridge transform", which, given such a hypothesis, extends its domain to "physically manifest facts about computations". Roughly speaking, the bridge transforms determines which computations are executed by the physical universe. In particular, this allows "locating the agent in the universe" by determining on which inputs its own source is executed.

0. Background

The "standard model" of ideal agency is Bayesian reinforcement learning, and more specifically, AIXI. We challenged this model before due to its problems with non-realizability, suggesting infra-Bayesianism as an alternative. Both formalisms assume the "cartesian cybernetic framework", in which (i) the universe is crisply divided into "agent" and "environment" and (ii) the two parts interact solely via the agent producing actions which influence the environment and the environment producing observations for the agent. This is already somewhat objectionable on the grounds that this division is not a clearly well-defined property of the physical universe. Moreover, once we examine the structure of the hypothesis such an agent is expected to learn (at least naively), we run into some concrete problems.

The modern understanding of the universe is that no observer plays a privileged role[1]. Therefore, the laws of physics are insufficient to provide a cartesian description of the universe, and must, to this end, be supplemented with "bridge rules" that specify the agent's location inside the universe. That is, these bridge rules need to translate the fundamental degrees of freedom of a physical theory (e.g. quantum wavefunction) to the agent's observations

(e.g. values of pixels on a camera), and translate the agent's actions

(e.g. signal to robot manipulators) in the other direction[2]. The cost of this is considerable growth in the description complexity of the hypothesis.

A possible retort is something like "how many bits can it really take to pick out the computer within the universe state and describe how to translate universe state to observations? Sure, it might be a considerable chunk of data. But, since Solomonoff induction only needs to make a kilobyte worth of predictive mistakes to learn to predict the inputs as well as any thousand-bit predictor, it should learn this sort of stuff pretty fast." This objection is addressed more later on in this post. A first counter-objection is that, for practical algorithms, this can be a bigger obstacle, especially when sample complexity scales superlinearly with description complexity.

For example, the Russo-Van Roy regret bound for Thompson sampling in multi-armed bandits has the time horizon necessary to get a particular regret bound scale as the square of entropy, and the computational complexity cost can also go way up as the description complexity rises, because you need to evaluate more hypotheses more times to test them. For reinforcement learning it's even worse, as in the case of a game where an agent must enter an n-digit password and if it gets it right, it gets reward, and if it gets it wrong, it does not get reward and can try again. The learning time for this game scales as Ω(2n), far superlinearly with description complexity.

Another assumption of AIXI is the simplicity prior, and ...

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: The EA Animal Welfare Fund Has Significant Room For More Funding, published by kierangreig on December 1, 2021 on The Effective Altruism Forum.

I have heard some people report the impression that the Animal Welfare Fund lacks room for more funding. I don’t think that impression is true, and that addressing it could be a pretty important point for EAs looking to make donations within the animal sector.

Briefly, here’s why I think the Animal Welfare Fund has some pretty significant room for more funding at this point:

Right now, there’s currently ~$600k in the Animal Welfare Fund (we just completed a round and are granting out another ~$1.1-$1.2M—note: still in the process of finalizing that total grant amount and the amount in fund metric doesn’t now include that total)

Our typical grant volume this year was ~$1.5M per round (ranging from ~$1M-$2M across three rounds). In total, we will grant out ~$4.5M (having raised ~$5M) for grants this year.

We have some plans for significant growth next year through some internal expansion plans in the works (e.g., possibly adding further fund managers and doing more active grantmaking).

Also, a lot of our grantees have grown, so they’ll have more room for funding. As a lot of the groups we give to are relatively small, they can grow at such a rate that they’d often be looking to absorb twice as much funding in the next year. If we zoom in on just say Crustacean Compassion, last year we granted $35k to them, and this year $92k, which comfortably corresponds to a greater than 100% growth in grant amount. Generally, it seems that if we have more money in the fund, it encourages some good organizations to request more funding for some quality projects.

Relatedly, some of the areas we grant in just tend to be pretty high growth and grow at comfortably >20% year on year. For instance, years ago there was basically very little that could be granted to Wild Animal Welfare, but this year we made several hundred thousand dollars in grants within that area. Similarly, previously there was not much that could be granted regarding invertebrates, but this year we granted several hundred thousand dollars in grants within that area.

So next year, I think that we could fairly comfortably and productively absorb and grant out in the realm of $6M-$9M (that’s a ~33%-100% increase on this year) without any significant decreases to the quality of our grants.

Note too, that in comparison to 2020 where we granted out $2.1M in total, this year we more than doubled that to $4.5M. I think we’re on track to handle that level of either absolute growth or percentage growth in grants for next year, which will put us in that $6M-$9M range.

So one way to look at this is that we now have ~$600k in the fund but next year we could do something like at least ~$6M in grants. So in that sense, we have several million dollars in room for more funding.

It could be worth thinking about how much we’ll likely raise for grants for next year too though. This year, we typically raised between $100k-$200k per month. Historically, we have seen about a ~4x-8x increase on that monthly total for the month of December and January (some end-of-year donations come in on the books in January).

Another way then to look at this then, is based on the current trends and growth in them year to year, we would now be looking at raising something like ~$5M (~$190k per month + ~4x that total for Dec and ~8x for Jan) between now and the end of next year (presuming trends hold about the same as other years, and I don't see too much of a reason to think they won't at this point). Again, as we have ~$600k now in the fund, that brings the total to ~$5.6M. But I think we could productively absorb at least ~$400k on top of that to grant out $6M, and again as much as ~$4.5M on top relatively comforta...

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: The EA Animal Welfare Fund Has Significant Room For More Funding, published by kierangreig on December 1, 2021 on The Effective Altruism Forum.

I have heard some people report the impression that the Animal Welfare Fund lacks room for more funding. I don’t think that impression is true, and that addressing it could be a pretty important point for EAs looking to make donations within the animal sector.

Briefly, here’s why I think the Animal Welfare Fund has some pretty significant room for more funding at this point:

Right now, there’s currently ~$600k in the Animal Welfare Fund (we just completed a round and are granting out another ~$1.1-$1.2M—note: still in the process of finalizing that total grant amount and the amount in fund metric doesn’t now include that total)

Our typical grant volume this year was ~$1.5M per round (ranging from ~$1M-$2M across three rounds). In total, we will grant out ~$4.5M (having raised ~$5M) for grants this year.

We have some plans for significant growth next year through some internal expansion plans in the works (e.g., possibly adding further fund managers and doing more active grantmaking).

Also, a lot of our grantees have grown, so they’ll have more room for funding. As a lot of the groups we give to are relatively small, they can grow at such a rate that they’d often be looking to absorb twice as much funding in the next year. If we zoom in on just say Crustacean Compassion, last year we granted $35k to them, and this year $92k, which comfortably corresponds to a greater than 100% growth in grant amount. Generally, it seems that if we have more money in the fund, it encourages some good organizations to request more funding for some quality projects.

Relatedly, some of the areas we grant in just tend to be pretty high growth and grow at comfortably >20% year on year. For instance, years ago there was basically very little that could be granted to Wild Animal Welfare, but this year we made several hundred thousand dollars in grants within that area. Similarly, previously there was not much that could be granted regarding invertebrates, but this year we granted several hundred thousand dollars in grants within that area.

So next year, I think that we could fairly comfortably and productively absorb and grant out in the realm of $6M-$9M (that’s a ~33%-100% increase on this year) without any significant decreases to the quality of our grants.

Note too, that in comparison to 2020 where we granted out $2.1M in total, this year we more than doubled that to $4.5M. I think we’re on track to handle that level of either absolute growth or percentage growth in grants for next year, which will put us in that $6M-$9M range.

So one way to look at this is that we now have ~$600k in the fund but next year we could do something like at least ~$6M in grants. So in that sense, we have several million dollars in room for more funding.

It could be worth thinking about how much we’ll likely raise for grants for next year too though. This year, we typically raised between $100k-$200k per month. Historically, we have seen about a ~4x-8x increase on that monthly total for the month of December and January (some end-of-year donations come in on the books in January).

Another way then to look at this then, is based on the current trends and growth in them year to year, we would now be looking at raising something like ~$5M (~$190k per month + ~4x that total for Dec and ~8x for Jan) between now and the end of next year (presuming trends hold about the same as other years, and I don't see too much of a reason to think they won't at this point). Again, as we have ~$600k now in the fund, that brings the total to ~$5.6M. But I think we could productively absorb at least ~$400k on top of that to grant out $6M, and again as much as ~$4.5M on top relatively comforta...

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: Rowing, Steering, Anchoring, Equity, Mutiny, published by Holden Karnofsky on November 30, 2021 on The Effective Altruism Forum. Audio version available at Cold Takes (or search Stitcher, Spotify, Google Podcasts, etc. for "Cold Takes Audio")

This post interrupts the Has Life Gotten Better? series to talk a bit about why it matters what long-run trends in quality of life look like.

I think different people have radically different pictures of what it means to "work toward a better world." I think this explains a number of the biggest chasms between people who think of themselves as well-meaning but don't see the other side that way, and I think different pictures of "where the world is heading by default" are key to the disagreements.

Imagine that the world is a ship. Here are five very different ways one might try to do one's part in "working toward a better life for the people on the ship."

Which of these is the "right" focus for improving the world? One of the things I like about the ship analogy is that it leaves the answer to this question totally unclear! The details of where the ship is currently trying to go, and why, and who's deciding that and what they're like, matter enormously. Depending on those details, any of the five could be by far the most important and meaningful way to make a positive difference.

If the ship is the world, then where are we "headed" by default (what happens if we have more total technology, wealth, and power over our environment)? Who has the power to change that, and how has it been going so far?

These are important questions with genuinely unclear answers. So people with different assumptions about these deep questions can get along very poorly with each other.

I think this sort of taxonomy provides a different angle on people’s differences from the usual discussions of pro/anti-government interventionism.

Next I will give some somewhat more detailed thoughts on the case for and against each of these pictures of "improving the world." This is not so much to educate the reader as to help them understand where I stand, and why I have some of the unusual views I do.

I talk a bit about the "track record" of each category. A lot of the point of this analogy is to highlight the importance of big-picture judgments about history.

Rowing

I use "rowing" to refer to the idea that we can make the world better by focusing on advancing science, technology, growth, etc. - all of which ultimately result in empowerment, helping people do whatever they want to do, more/faster. The idea is that, in some sense, we don't need a specific plan for improving lives: more capabilities, wealth, and empowerment ("moving forward") will naturally result in that.

Rowing is a contentious topic, and it’s contentious in a way that I think cuts across other widely-recognized ideological lines.

To some people, rowing seems like the single most promising way to make the world a better place. People and institutions who give off this vibe include:

Tech entrepreneurs and VCs such as Marc Andreessen (It's Time to Build) and Patrick Collison (progress studies).

Libertarian-ish academics such as Tyler Cowen and Alex Tabarrok (see their books for a sense of this).

The many institutions and funders dedicated to general speeding up of science, such as the Moore Foundation, Howard Hughes Medical Institute, and CZI Biohub.2

The global development world, e.g. nonprofits such as Center for Global Development and institutions such as the World Bank, which seeks to help "developing" countries "develop." This generally includes a large (though not exclusive) focus on economic growth.

I sometimes see rowing-oriented arguments tagged as “pro-market” or even “libertarian,” but I think that isn't a necessary connection. You could argue - and many do - that a lot of the biggest and mo...