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: A bunch of videos for intuition building (2x speed, skip ones that bore you), published by the gears to ascension on March 12, 2023 on LessWrong.

did I mention 2x speed? hit play on many, pause on almost as many.

This is a big list of the youtube videos I find myself linking to people most often. These are all from the same playlist: - comment here if you'd like edit access, I give it out readily. I'd love to have people moving the most important and insightful videos towards the beginning. I'd also love to see people clone the playlist and just make their own version.

These vary wildly in topic and difficulty level. I generally do not try to avoid watching things above my level, I just use it as inspiration for how to fill in what I'm missing. If something sounds basic to you, it probably is.

Many of these videos are quite short, many are quite long.

1min: neuron clip

23min: neuroscience overview (bio neuron interpretation)

or 10min with 2x speed!

10min: overview of learning techniques (bit clickbaity title but I include because I like it plenty anyhow)

or 5min with 2x speed!

2min: visual intuition - details of how one particular chaotic fluid flow move

11min: research talk on what collective intelligence is. (see also many more cool talks from MITCBMM!)

or 5min with 2x speed!

2min: visualization of a volume of neurons in a (mouse?) amygdala

8min: cognitive biases in practice

33min: absolutely incredible visual intro to physics sims focusing towards fluid simulation

or 15min with 2x speed!

15min: cs101 "ok, but what does it mean to abstract over the matter of a computer"

or 7min with 2x speed!

1min: visualization of particle lenia

20min: overview of Michael Levin's research on the bioelectric communication of cells for morphogenesis and morphogenic editing without genetic change

or 10min with 2x speed

11min: cs101 how a neural network is actually just line segments (with relu, anyway)

12min: nice intro to what chaos theory is actually about

18min: overview of ways visual proofs can mislead

4min: overview of some important additional notes on how to learn efficiently. this playlist does not satisfy them all.

14min: Visual intro to why neural networks work. goes into detail about the geometric interpretation of neural networks.

15min: geometric interpretation of bayes' rule. Useful for intuition building even if you get the math.

See also chris olah's blog post on the same topic from a few years prior.

4min: visualization of atoms that better communicates what the probability fields are fields of.

6min: nice intro to what claim the manifold hypothesis of neural network effectiveness makes about the structure of natural data.

20min: a perspective on why anecdotes are important for natural communication (very rough summary: humans natively think in sequences of embodied events)

20min: intro to the clocks of the brain

43min: visualization of inventing math from only physical shapes

As a strict philosophical materialist, this is what made me start believing in math again ;)

20min on 2x speed!

7min: visualization of one rather narrow simulation of abstract market agents and the effect that interest-bearing loans have on a simulation

There are several more videos in will ruddick's playlists that go over the various configuration changes to this sim, and he also has a version you can try online

35min: more steps through even larger scale abstractions of fluid behavior for simulation

10min: intro to why you'd want to know category theory - all math is secretly category theory (but not an intro to the actual math in detail)

15min: overview of some results from evolutionary game theory

25min: overview of a very common abstract model of phase transitions

37min: rehash of the percolation video but with slightly less grokkable explanation, but then gets into connection to how this...

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 Support Someone Who is Struggling, published by David Zeller on March 11, 2023 on LessWrong.

[Crossposted from my blog]

There’s no shortage of pain, tragedy and loss in the world. And if you’re anything like me, you don’t always know how to be helpful when a loved one is going through the worst of it.

Over the past few years, I’ve been trying to get better at that.

I’ve read a couple dozen therapy textbooks, I’ve done four hundred or so hours of client-centered counselling, and I’ve been in a handful of other official and unofficial helping roles. By no means am I an expert, but I sure know more than I used to.

For my first blog post, I wanted to write something that past-me might have found helpful when he started stumbling through it all. In time, there’s so much more that I want to say on the art of supporting others. But for now...

Here are four fundamentals for helping someone who’s having a rough time:

1 - Simply listen. It helps far more than most of us expect.

When a catastrophe happens, it can change the whole landscape of one’s world. The tectonic plates shift, things break, and everything comes to look bewilderingly different to how it did before.

In the aftermath, we may have no good choice other than to stop, watch the buildings fall, and slowly map out this strange new world we’re in. Perhaps only then we can move forward.

Unfortunately, processing such big changes purely in one’s own head is. hard. Thoughts are ephemeral and it’s easy to think in circles, to get stuck, to have blind spots, to ruminate.

This is where listening comes in. A good listener can be of much help with that working through process. Patiently, the listener can keep track of where a conversation is getting stuck, gently bring up the things that are being avoided or missed, help bring attention towards what is most important, and bring a genuine sense of connection that makes all the bad stuff a little easier to bear.

As simple as it seems, having someone there to just listen may be exactly what the person in front of you needs.

2 - Rather than focusing on the bright side, sit with the other person’s real feelings.

This next point comes straight from Brené Brown. I’ve been shown the same video of her so many times in different training courses that I’m starting to get Stockholm syndrome. All the same, what it says is important.

Often when we’re trying to support another person, we try to get them to focus on the bright side. Standing separately from the other’s experience, we attempt to offer them silver linings.

“You may have failed this class. but at least your other grades are good.”

“Your partner left you. but at least you’re free to find someone who’ll treat you better.”

“You may have a disease with no cure. but at least there are lots of scientists working to find new treatments.”

People use these silver linings with the intention to help the other person view their situation in a more positive light. Unfortunately, in most cases, this does not end up bringing them any relief.

When you’re going through a tough time, talking to someone who only focuses on the nicer aspects of your bad situation most often just feels disorienting. This happens because, at some level, you’re being told that your problems are not as bad as you think they are. Instead of feeling reassured, you feel like your grip on reality is being questioned. The good intentions get lost in translation.

Luckily, there’s an alternative that really does let us bring some relief to others’ suffering: Empathy.

Rather than try to look on the bright side, it’s helpful to sit with the other person in their pain. To attempt to really understand, at an emotional level, the whole landscape of what they’re going through. When we manage to do this, it brings a genuine sense of connection, and a feeling that one doesn...

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 Power of Intelligence - The Animation, published by Writer on March 11, 2023 on The Effective Altruism Forum.

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: Understanding and controlling a maze-solving policy network, published by TurnTrout on March 11, 2023 on LessWrong.

TL;DR: We algebraically modified the net's runtime goals without finetuning. We also found (what we think is) a "motivational API" deep in the network. We used the API to retarget the agent.

Summary of a few of the most interesting results:

Langosco et al. trained a range of maze-solving nets. We decided to analyze one which we thought would be interesting. The network we chose has 3.5M parameters and 15 convolutional layers.

This network can be attracted to a target location nearby in the maze—all this by modifying a single activation, out of tens of thousands. This works reliably when the target location is in the upper-right, and not as reliably when the target is elsewhere.

Considering several channels halfway through the network, we hypothesized that their activations mainly depend on the location of the cheese.

We tested this by resampling these activations with those from another random maze (as in causal scrubbing). We found that as long as the second maze had its cheese located at the same coordinates, the network’s behavior was roughly unchanged. However, if the second maze had cheese at different coordinates, the agent's behavior was significantly affected.

This suggests that these channels are inputs to goal-oriented circuits, and these channels affect those circuits basically by passing messages about where the cheese is.

This network decides whether to acquire cheese not only as a function of path-distance to cheese, but—after controlling for path-distance—also as a function of Euclidean/"perceptual" distance between the mouse and the cheese, even though the agent sees the whole maze at once.

Another simple idea: We define a "cheese vector" as the difference in activations when the cheese is present in a maze, and when the cheese is not present in the same maze. For each maze, we generate a single cheese vector and subtract that vector from all forward passes in that maze. The agent now ignores cheese most of the time, instead heading towards the top-right region (the historical location of cheese).

We propose the algebraic value-editing conjecture (AVEC): It's possible to deeply modify a range of alignment-relevant model properties, without retraining the model, via techniques as simple as "run forward passes on prompts which e.g. prompt the model to offer nice- and not-nice completions, and then take a 'niceness vector' to be the diff between their activations, and then add the niceness vector to future forward passes."

Introducing the training process and visualizations

In this post, we'll mostly discuss what we found, not what our findings mean.

Let's run through some facts about Langosco et al.'s training process. Mazes had varying effective sizes, ranging from 3×3 to 25×25:

Each 64×64 RGB observation is processed by a deeply convolutional (15 conv layers!) network, without memory (i.e. no recurrent state):

Why does the agent go to the cheese sometimes, and the top-right corner other times?

It's not that the agent wasn't trained for long enough.

Sampling rollouts from the trained policy adds a lot of noise. It's also hard to remember what the agent did in what part of the maze. To better understand this mouse, we'll take a bird's-eye view.

A nicer way to view episodes is with a vector field view, which overlays a vector field representing the agent policy for a given maze.

We consider two kinds of vector fields:

While the net probability vector field leaves open two degrees of freedom per net probability vector, in practice it seems fine for eyeballing mouse behavior.

Behavioral analysis

When in doubt, get more data. When Alex (TurnTrout) was setting directions but didn't know what to do, he'd think "what data firehydrants can I crack...

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: Understanding and controlling a maze-solving policy network, published by Alex Turner on March 11, 2023 on The AI Alignment Forum.

TL;DR: We algebraically modified the net's runtime goals without finetuning. We also found (what we think is) a "motivational API" deep in the network. We used the API to retarget the agent.

Summary of a few of the most interesting results:

Langosco et al. trained a range of maze-solving nets. We decided to analyze one which we thought would be interesting. The network we chose has 3.5M parameters and 15 convolutional layers.

This network can be attracted to a target location nearby in the maze—all this by modifying a single activation, out of tens of thousands. This works reliably when the target location is in the upper-right, and not as reliably when the target is elsewhere.

Considering several channels halfway through the network, we hypothesized that their activations mainly depend on the location of the cheese.

We tested this by resampling these activations with those from another random maze (as in causal scrubbing). We found that as long as the second maze had its cheese located at the same coordinates, the network’s behavior was roughly unchanged. However, if the second maze had cheese at different coordinates, the agent's behavior was significantly affected.

This suggests that these channels are inputs to goal-oriented circuits, and these channels affect those circuits basically by passing messages about where the cheese is.

This network decides whether to acquire cheese not only as a function of path-distance to cheese, but—after controlling for path-distance—also as a function of Euclidean/"perceptual" distance between the mouse and the cheese, even though the agent sees the whole maze at once.

Another simple idea: We define a "cheese vector" as the difference in activations when the cheese is present in a maze, and when the cheese is not present in the same maze. For each maze, we generate a single cheese vector and subtract that vector from all forward passes in that maze. The agent now ignores cheese most of the time, instead heading towards the top-right region (the historical location of cheese). Furthermore, a given maze's cheese vector transfers across mazes to other mazes with cheese in the same location.

We propose the algebraic value-editing conjecture (AVEC): It's possible to deeply modify a range of alignment-relevant model properties, without retraining the model, via techniques as simple as "run forward passes on prompts which e.g. prompt the model to offer nice- and not-nice completions, and then take a 'niceness vector' to be the diff between their activations, and then add the niceness vector to future forward passes."

Introducing the training process and visualizations

In this post, we'll mostly discuss what we found, not what our findings mean.

Let's run through some facts about Langosco et al.'s training process. Mazes had varying effective sizes, ranging from 3×3 to 25×25:

Each 64×64 RGB observation is processed by a deeply convolutional (15 conv layers!) network, without memory (i.e. no recurrent state):

Why does the agent go to the cheese sometimes, and the top-right corner other times?

It's not that the agent wasn't trained for long enough.

Sampling rollouts from the trained policy adds a lot of noise. It's also hard to remember what the agent did in what part of the maze. To better understand this mouse, we'll take a bird's-eye view.

A nicer way to view episodes is with a vector field view, which overlays a vector field representing the agent policy for a given maze.

We consider two kinds of vector fields:

While the net probability vector field leaves open two degrees of freedom per net probability vector, in practice it seems fine for eyeballing mouse behavior.

Behavioral analysis

When in doubt, get m...

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 my community successfully reduced sexual misconduct, published by titotal on March 11, 2023 on The Effective Altruism Forum.

[Content warning: this post contains discussions of sexual misconduct, including assault]

In response to the recent articles about sexual misconduct in EA and Rationalism, a lot of discussion has ended up being about around whether the level of misconduct is “worse than average”. I think this is focusing on the wrong thing. EA is a movement that should be striving for excellence. Merely being “average” is not good enough. What matters most is whether EA is the best it could reasonably be, and if not, what changes can be made to fix that.

One thing that might help with this is a discussion of success stories. How have other communities and workplaces managed to “beat the average” on this issue? Or substantially improved from a bad place? For this reason I’m going to relay an anecdotal success story below. If you have your own or know of others, I highly encourage you to share it as well.

Many, many, years ago, I joined a society for a particular hobby (unrelated to EA), and was active in the society for many, many years. For the sake of anonymity, I’m going to pretend it was the “boardgame club”. It was a large club, with dozens of people showing up each week. The demographics were fairly similar to EA, with a lot of STEM people, a male majority (although it wasn’t that overwhelming), and an openness to unconventional lifestyles such as kink and polyamory.

Now, the activity in question wasn’t sexual in nature, but there were a lot of members who were meeting up at the activity meetups for casual and group sex. Over time, this meant that the society gained a reputation as “the club you go to if you want to get laid easily”. Most members, like me, were just there for the boardgames and the friends, but a reasonable amount of people came there for the sex.

As it turns out, along with the sex came an acute problem with sexual misconduct, ranging from pushing boundaries on newcomers, to harassment, to sexual assault. I was in the club for several years before I realised this, when one of my friends relayed to me that another one of my friends had sexually assaulted a different friend.

One lesson I took from this is that it’s very hard to know the level of sexual misconduct in a place if you aren’t a target. If I was asked to estimate the “base rate” of assault in my community before these revelations, I would have falsely thought it was low. These encounters can be traumatic to recount, and the victims can never be sure who to trust or what the consequences will be for speaking out. I’d like to think I was trustworthy, but how was the victim meant to know that?

Eventually enough reports came out that the club leaders were forced to respond. Several policies were implemented, both officially and unofficially.

Kick people out.

Nobody has a democratic right to be in boardgame club.

I think I once saw someone mention “beyond reasonable doubt” when it comes to misconduct allegations. That standard of evidence is extremely high because the accused will be thrown into jail and deprived of their rights. The punishment of “no longer being in boardgame club” does not warrant the same level of evidence. And the costs of keeping a missing stair around are very, very high.

Everyone that was accused of assault was banned from the club. Members that engaged in more minor offenses were warned, and kicked out if they didn’t change. To my knowledge, no innocent people were kicked out by mistake (false accusations are rare). I think this made the community a much more pleasant place.

2. Protect the newcomers

When you attend a society for the first time, you do not know what the community norms are. You don’t know if there are avenues to report misconduct. You don’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: Thoughts on the OpenAI alignment plan: will AI research assistants be net-positive for AI existential risk?, published by Jeffrey Ladish on March 10, 2023 on LessWrong.

Note: I really appreciate the work that the OpenAI alignment team put into their alignment plan writeup and related posts, especially Jan Leike, the leader of that team. I believe open discussions about alignment approaches make it more likely that the whole community will be able to find flaws in their own plans and unappreciated insights, resulting in better alignment plans over time.

Summary: OpenAI’s alignment plan acknowledges several key challenges of aligning powerful AGI systems, and proposes several good ideas. However, the plan fails to sufficiently address:

The dual-use nature of AI research assistants and the high risk that such assistants will improve capabilities more than alignment research in ways that net-increase AI existential risk.

The likely challenges involved in both generating and evaluating AI alignment research using AI research assistants. It seems plausible that generating key insights about the alignment problem will not be possible before the development of dangerously powerful AGI systems.

The nature and difficulty of the alignment problem. There are substantial reasons why AI systems that pass all tests in development may not stay safe once able to act in the world. There are substantial risks from goal misgeneralization, including deceptive misalignment, made worse by potential rapid increases in capabilities that are hard to predict. Any good alignment plan should address these problems, especially since many of them may not be visible until an AI system already has dangerous capabilities.

The dual-use nature of AI research assistants and whether these systems will differentially improve capabilities and net-increase existential risk

There has been disagreement in the past about whether “alignment” and “capabilities” research are a dichotomy. Jan Leike has claimed that they are not always dichotomous, and this is important because lots of capabilities insights will be useful for alignment, so the picture is not as worrisome as a dichotomous picture might make it seem.I agree with Jan that these alignment and capabilities research are not dichotomous, but in a way I think actually makes the problem worse, not better. Yes, it’s probable that some AI capabilities could help solve the alignment problem. However, the general problem is that unaligned AGI systems are far easier to build - they’re a far more natural thing to emerge from a powerful deep learning system than an aligned AGI system.

So even though there may be deep learning capabilities that can help solve the alignment problem, most of these capabilities are still easier applied to making any AGI system, most of which are likely to be unaligned even when we’re trying really hard.

Let’s look at AI research assistants in particular. I say “AI research assistant” rather than “alignment research assistant” because I expect that it's highly unlikely that we will find a way to build an assistant that is useful for alignment research but not useful for AI research in general. Let’s say OpenAI is able to train an AI research assistant that can help the alignment team tackle some difficult problems in interpretability. That’s great! However, a question is, can that model also help speed up AGI development at the rest of the company? If so, by how much? And will it be used to do so?

Given that building an aligned AGI is likely much harder than building an unaligned AGI system, it would be quite surprising if an AI research assistant was better at helping with AGI safety research differentially over AGI development research more broadly. Of course it’s possible that a research tool that sped up capabilities research more ...

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: Questions about Conjecure's CoEm proposal, published by Akash on March 9, 2023 on LessWrong.

Conjecture recently released an AI safety proposal. The three of us spent a few hours discussing the proposal and identifying questions that we have. (First, we each re-read the post and independently brainstormed a few questions we had. Then, we discussed the post, exchanged questions/uncertainties, and consolidated our lists).

Conjecture's post is concise, which means it leaves out many details. Many of our questions are requests for more details that would allow us (and others) to better understand the proposal and evaluate it more thoroughly.

Requesting examples and details

What are the building blocks that the CoEms approach will draw from? What are examples of past work that has shown us how to build powerful systems that are human-understandable?

What are examples of “knowledge of building systems that are broadly beneficial and safe while operating in the human capabilities regime?” (see Wei_Dai’s comment)

What’s an example of an experiment that would be considered part of the CoEm agenda? (see Garret Baker’s comment)

What kinds of approaches does Conjecture intend to use to extract alignment insights “purely from mining current level systems”? (Is this the same as interpretability research and digital neuroscience?)

The “minimize magic” section feels like that is where the juice is, but it’s not really explained much, which makes it difficult to evaluate. Can you offer more details about how you intend to minimize magic?

Conceptual questions

Assume you had a fully human-understandable system, and you could understand its current capabilities. How would you be able to forecast its future capabilities (e.g., if deployed or if given certain commands)?

If we solve human neuroscience such that we could understand the brain of a 2-year-old, we would be able to accurately assess the (current) capabilities of the 2-year-old. However, we would not necessarily be able to predict the (future) capabilities of this brain once it is 30 years old. Analogously, if we had a human-understandable AI (that may be superintelligent) through the CoEms agenda, would we only be able to understand its current capabilities, or would there be a reliable way to forecast its future capabilities?

Charlotte thinks that humans and advanced AIs are universal Turing machines, so predicting capabilities is not about whether a capability is present at all, but whether it is feasible in finite time with a low enough error rate. Predicting how such error rates decline with experience and learning seems roughly equally hard for human-understandable AIs and other AIs.

How easy is it to retarget humans?

When you refer to “retargetability”, we assume you refer to something like the following: “Currently the AI has goal X, you want to train it to have goal Y. If you do that, you truly change its goals to Y (rather than making it pretend to follow Y and then when you are not in control anymore, it switches back to X”.

We agree that in some sense, humans are retargetable. For example, if someone has very advanced persuasion tools or if the “persuader” is significantly stronger than the “persuadee” (e.g., a dictator persuading a citizen).

But even that is very hard, and often one just changes their incentives/strategy rather than their actual goals.

However, humans seem to be much less retargetable by other agents who are similarly powerful. For example, how would you retarget the goals of an (equally intelligent and equally powerful) neighbor?

Alternatively, you might refer to a much weaker version of “retargability”, e.g. very weak version of corrigible alignment. If this is what you mean, I am wondering why this is a particularly important property?

Other questions

Does Conjecture believe this approach is comp...

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: Share the burden, published by 2ndRichter on March 11, 2023 on The Effective Altruism Forum.

My argument: let’s distribute the burden of correcting and preventing sexual misconduct through collective effort, not letting the burdens and costs fall overwhelmingly on those who have experienced it.

CW: sexual assault/harassment

I work at CEA but write this in an individual capacity. Views expressed here are my own, not CEA's.

I encourage you to reach out to me individually on Twitter (@emmalrichter) if you want to discuss what I raise in this post. I’d love to engage with the variety of potential responses to what I’ve written and would love to know why you upvote or downvote it.

Intro and Context

Some of you already know that I’m a survivor. I was sexually assaulted, harassed, or abused in independent situations at the ages of 16, 17, 18, and 20. I am intentionally open and vocal about what I’ve gone through, including a PTSD diagnosis a few years ago.

Recent events in the EA community have reminded me that the mistreatment of people through sexual or romantic means occurs here (as it does everywhere). Last week at EAG, I received a Swapcard message that proposed a non-platonic interaction under the guise of professional interaction. I went to an afterparty where someone I had just met—literally introduced to me moments before—put their hand on the small of my back and grabbed and held onto my arm multiple times. These might seem like minor annoyances, but I have heard and experienced that these kinds of small moments happen often to women in EA. These kinds of experiences undermine my own feelings of comfort and value in the community.

This might be anecdata, as some people say, and I know obtaining robust data on these issues has its own challenges. Nonetheless, my experience and those of other women in EA indicate that there’s enough of a problem to consider doing more.

I’m writing this post for a few reasons:

I want to draw attention to the suffering of women here in the community.

I want to convey the costs placed on survivors seeking justice and trying to prevent further harm to others.

I want to share just how taxing it can be for survivors to work on these problems on their own, both due to the inherent pain of reliving experiences and the arduousness of most justice processes.

Above all, I want to make this request of our community: let’s distribute the burden of correcting and preventing sexual misconduct as fairly as we can, not letting the burdens and costs fall overwhelmingly on those who have experienced it. They have suffered so much already—they have suffered enough. I hope we can be as agentic and proactive in this domain as we strive to be in other areas of study and work.

Here are sub-arguments that I’ll explore below:

Before placing the burden of explanation on the survivor, we can employ other methods to learn about this constellation of social issues. We can listen to survivors more effectively and incorporate the feedback of those who want to share while also finding other resources to chart paths forward.

Good intentions can still lead to negative outcomes. This can apply to both bystanders who refrain from engaging with the subject out of the intention of not making things worse and might also apply to those who perpetrate harmful behaviors (as I discuss in my own experience further down).

Why write about the meta-level attitude and approach when I could have written something proposing object-level solutions?

Because how we approach finding object-level solutions will affect the quality of those solutions—particularly for those who are most affected by these problems. I don’t feel informed enough to propose institutional reforms or particular policies (though I intend to reflect on these questions and research them). I do feel informed enough 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: Tyler Johnston on helping farmed animals, consciousness, and being conventionally good, published by Amber Dawn on March 10, 2023 on The Effective Altruism Forum.

This post is part of a series of six interviews. As EAs, we want to use our careers or donations to do the most good - but it’s difficult to work out what exactly that looks like for us. I wanted to interview effective altruists working in different fields and on different causes and ask them how they chose their cause area, as well as how they relate to effective altruism and doing good more generally. During the Prague Fall Season residency, I interviewed six EAs in Prague about what they are doing and why they are doing it. I’m grateful to my interviewees for giving their time, and to the organisers of PFS for supporting my visit.

I’m currently working as a freelance writer and editor. If you’re interested in hiring me, book a short call or email me at [email protected]. More info here.

Tyler Johnston is an aspiring effective altruist currently based out of Tulsa, Oklahoma. Professionally, he works on corporate campaigns to improve the lives of farmed chickens, and is interested in cause prioritisation, interspecies comparisons, and the suffering of non-humans. He’s also a science-fiction fan and an amateur crossword puzzle constructor.

We talked about:

his work on The Humane League’s corporate animal welfare campaigns

how he became a vegan and animal advocate

whether animals are conscious

how being conventionally good is underrated

On his work at The Humane League

Amber: Tell me about what you’re doing.

Tyler: I work for The Humane League. We run public awareness campaigns to try to get companies to make commitments to improve the treatment of farmed animals in their supply chains. This strategy first gained traction in 2015, and was immediately really powerful. Since then, it has got a lot of interest from EA funders.

Amber: Did The Humane League always do that, or was it doing something else before 2015?

Tyler: It was a long journey; The Humane League’s original name was Hugs for Puppies

Amber: Aww, that’s very cute!

Tyler: Yeah, I feel like we’d be a more likeable organisation if we were still called that. They started doing demonstrations around issues like fur bans, and other animal welfare issues there was already a lot of energy around. They then switched to focussing on vegan advocacy, which involved things like leafleting, and sharing recipes and resources.

Amber: So the strategy at that time then was to encourage people to go vegan, which would lower demand for factory farming, which would mean there were fewer factory-farmed animals?

Tyler: That’s right. There was some early evidence that showed this was promising, and it also just made sense to them, since most vegans would attribute their own choice to be vegan to a time in the past when they heard and agreed with the arguments. So they thought, ‘why wouldn’t this export to other people?’

Amber: But you said the strategy is different now - it’s to lobby actual food producers to treat the animals that they’re farming better. Say more about that.

Tyler: That’s our dominant strategy now, yeah. It’s part of a broader shift in the [animal advocacy] movement toward institutional change rather than individual change. If for some given company, you either have to change the minds of, like, 10 million consumers, or a dozen executive stakeholders - the latter is just a lot more tractable. It started with running small campaigns to persuade companies to source cage-free eggs, and it turned out that this worked. Around 2015 there was a sharp turning point in the number of farmed birds that are cage-free - before 2015, the percentage was growing very slowly, from 3% to 5%, but between 2015 and today, the percentage went up from 5% to 36%. And people attr...