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: Being a donor for Fecal Microbiota Transplants (FMT): Do good & earn easy money (up to 180k/y), published by Anton Rodenhauser on July 8, 2022 on LessWrong.

TLDR

Help to find super donors for Fecal Microbiota Transplants (FMT)! By donating stool for FMT, exceptionally healthy people can earn $180,000/yr with little effort. Simultaneously they can help chronically ill people improve their health with a novel treatment. Apply here.

More detailed summary

Fecal Microbiota Transplants (FMTs) is a procedure that transfers the stool of healthy people to the guts of sick people. The mechanism is to replace a dysbiotic gut microbiome with an eubiotic, disease-resistant gut microbiome.

So far, FMT has only been proven to be very effective in curing patients of C. difficile infections. However, since the gut microbiome impacts and regulates virtually every aspect of human health, function, and development, FMT is a promising treatment for a very wide range of chronic and acute health conditions[1][2].

FMTs can easily & safely be done at home, both for the donor and the recipient - no doctor needed.

A super-donor is a person who yields outstanding positive effects as a stool donor. Only exceptionally healthy & fit people are super-donors. They are very rare and a few people within the FMT movement have been searching for them for years. Most donors, even reasonable healthy ones, only yield modest benefits.

Human Microbes, an organization specialized in finding super-donors and connecting them with recipients, is paying super-donors $500 per stool. This adds up to $180,000/yr if donating a daily stool!

Purpose of this post

Spread the news that this opportunity for super donors exists: With little effort, they can help lots of ill people (“make them more effective”) while earning $500/stool. You should check whether you are a super donor. If you know someone who might be, please share this post with them.

Give a rough introduction to FMTs & Human Microbes, with the aim to start a rational discussion about FMT’s benefits. FMTs look very promising, but there is by no means a clear scientific consensus. I’d love for more EAs to look into this, to get more certainty about its efficacy!

If your health is suboptimal, maybe you could consider using FMTs. Not just for obviously gut health-related issues, but much more broadly than you might think. Potentially even e.g. unspecified subclinical low energy, low mood, or mental sluggishness.

At the end, this post provides you with a long list of useful further links on FMTs, the microbiome and its connection to a vast amount of (mental-) health conditions. This is to help you do your own research, both as a potential super donor or as someone who might benefit from FMTs.

Who might be a super donor?

Donors must be in exceptional physical and mental health. Ideally, top young athletes. Though there are non-athletes who qualify as well.

Ideally, donors would be under 30 years old. Donors can be under 18 provided they have signed consent from their parents. Many children need FMT and are ideally matched with young donors.

Donors should have minimal antibiotic use.

Donors should have a "Type 3 stool" as a Bristol Stool Type.

Check out the links at the end of this post for more information and sources.

As you can imagine, super donors are very rare and thus very high in demand. In fact, Human Microbes has screened over 25,000 donor applicants so far, including hundreds of college and professional athletes, and still hasn’t found that one “perfect” super donor that fully satisfies all their donor criteria! So please reach out to Human Microbes about potential super donor candidates!!

How much good can you do by finding a super donor?

Know an EA whose gut-borne suboptimal health is holding them back from doing the most good they can do? Find them a s...

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: Covid 7/7/22: Paxlovid at the Pharmacy, published by Zvi on July 7, 2022 on LessWrong.

Pharmacists are going to be allowed to prescribe Paxlovid, while doctors continue to be confused by the guideline of ‘you can give this life-saving medicine that cures disease to the people who are sick with that disease.’ The news on the vaccine update front is good. Cases and deaths went unexpectedly higher, which is slightly worrisome, but seems unlikely to indicate anything big.

On an unrelated note, I’ve also written a bit about a giving-people-money experiment in Chicago that did not go so well. With Covid quieting down, I’m likely going to be lowering the threshold for including non-Covid items in the weekly updates. The alternative was to have a growing backlog of things that I am unlikely to ever get to deal with.

Executive Summary

Get Paxlovid at your pharmacist, yay.

Vaccine update update is they are closer to updating.

Monkeypox continues to double every 10.5 days or so.

Let’s run the numbers.

The Numbers

Predictions

Prediction from last week (4th of July): 600k cases (-13%) and 1,800 deaths (-16%).

Results: 675k cases (-1%) and 2,272 deaths (+11%!).

Prediction for next week: 700k cases (+5%) and 2,250 deaths (-1%).

Cases and deaths both going unexpectedly higher implies there wasn’t much of a holiday effect. Deaths in particular have no reason to be higher even without a holiday effect, so that’s weird but I did adjust for Florida and the effect is still there in the South. My guess is some of this is random rather than permanent, but it’s hard to say, and I still have to presume that there was some amount of holiday effect to undo.

Deaths

Cases

Did you know this in many places represents medium or high levels of Covid?

All right then.

Paxlovid at the Pharmacist

You will be able to get Paxlovid prescribed to you directly by the pharmacist. Seems like an excellent development. I’m actually starting to notice a pattern of such pockets of sanity happening more often recently, after delaying long enough to let a lot of the value be destroyed, in quiet places where no one is looking. It’s as if they are worried someone will notice someone is optimizing for good outcomes in this hive of scum and villainy, and so they must be cautious.

You can also get Paxlovid at the Pop-Up here in NYC.

Meanwhile, doctors clamor for more clarity. Yes, we are allowed to prescribe life-saving medicine to (more or less) all our patients with Covid-19, but who should be taking Covid-19 sufficiently seriously that we should give them life-saving medicine for it? Should a healthy 50-year-old get life saving medicine, or are they at such little risk they shouldn’t bother, despite us locking them down for the better part of two years over this?

I mean, who needs a cure, anyway? So many confusing questions.

Vaccine Update Update

Matthew Yglesias reminds us that vaccines work better when you update them, and that we need to update our vaccines faster, and that other considerations don’t much matter.

Back in February of 2021, the FDA suggested that in the future, variant-optimized vaccine updates would be fast-tracked for approval.

They suggested at the time that companies “would need to submit new data that shows the modified vaccine produces a similar immune response and is safe, similar to the process for annual flu vaccines,” and this could be done “without the need for lengthy clinical trials.”

That would have been a good idea! But instead they seem to have opted for months of intensive clinical trials. The problem there, which should be obvious, is that while the clinical trials enhance our confidence that the new booster is in fact more effective against Omicron, they also radically decrease the real-world efficacy of the Omicron-optimized booster by ensuring that it gets into people’s arms far ...

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: When Giving People Money Doesn't Help, published by Zvi on July 8, 2022 on The Effective Altruism Forum.

(Cross-posted by request, always happy to do so if something is of sufficient interest.)

I’ve been thinking a bunch about this study (direct link) of CBT-style techniques used to try and get criminals onto a better life path in Liberia that showed what looks like spectacular results, curious if anyone wants to share thoughts. If it would replicate seems like it crosses the EA-worthiness threshold.

I’ve been thinking more about the topic of this post, which is the cash transfer study in Chicago. That worked out less well. It seems that small cash transfers to those in need here in America made them feel worse and maybe even actually makes them worse off somehow? (paper)

This prediction seems very reasonable, and I think I would have agreed with it if I’d thought about this in advance. My model would have said that obviously cash is good and cash is helpful, in the short term getting some of it will be helpful. In the long term I’d expect some positive impact from the wealth effect. And in some cases, that money will have a particular jam that it solves or opportunity it enables (e.g. fix your car, security deposit on a new apartment, maybe even give opportunity to start a small business). In those cases it would have a permanent impact, so I would predict some overall positive impact. But in my experience there are enough implicit taxes and poverty traps for such people, and such a cultural lack of the idea of hanging onto money, that it is tough to hang onto funds, so mostly they’d be back where they started.

It looks like within about three weeks the money was mostly gone for both groups. That is not what would happen for an economically rational agent, unless they had a lot of opportunities for very high return on capital or they were paying down debt.

Elizabeth had a similar reaction to mine.

There’s something very interesting in this graph. The three groups have completely converged by the end of the 180 day period, but the average bank balance is now considerably higher. They all started out with ~$200 and now they have ~$400. That feels like a meaningfully different average amount.

This happened during Covid, so I am guessing that this is due to the Covid checks or other similar factors. This certainly complicates the experiment.

Those are pretty different things to happen. If you spend the money on bills, that’s paying down debt. It makes sense that this should happen right away. Spending on food and drink in a way you wouldn’t have otherwise is true consumption, the opposite of that.

Note of course this is ‘what they intended to do’ with the money, not what actually happened. Gambling money tends to end up bigger than expected and all that.

I’d ask less ‘what do poor people spend money on?’ and more ‘what do poor people spend more money on when they get a windfall?’

This tells a very clear story. The control group is spending a lot on groceries, but the treatment groups are spending less. I don’t think they are splurging all that much here, more like they are spending money there anyway and money is fungible. This makes sense, this is smart.

Where they are spending it is on bills, utilities (more bills), and especially on housing. Also small on general debt, education and clothes.

Notice that non-durable household goods spending is stronger in the control group, and durable household goods spending is stronger in the treatment group.

Note that it’s not easy to spend only $500 or even $2000 on housing, in a way that you weren’t already spending it, other than paying debt. So my guess is the bulk of this is ‘was behind on rent or mortgage’ and now they aren’t, but the size of this shift says that perhaps some of them did use this to get a new place.

So I int...

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: Wild animal welfare in the far future, published by saulius on July 8, 2022 on The Effective Altruism Forum.

Opinions expressed here are solely my own and do not express the views or opinions of my employer.

Summary

In this article, I analyze which far-future wild animal welfare (WAW) scenarios seem most important from a utilitarian perspective, and what we can do about them. While I don’t think that WAW is among the most important longtermist considerations, perhaps a few small projects in this area are worthwhile.

Scenarios where humans have the biggest impact on WAW seem to be about spreading wildlife:

Humans could transform hundreds of millions of planets to be similar to Earth (including wildlife) so that humans could live on them. E.g., if an artificial general intelligence (AGI) is aligned, and humans in charge are only interested in space colonization with biological humans and animals.

Humans could create artificial settlements in space that contain some wildlife.

Humans could spread wildlife to other planets with life without colonizing them, perhaps because they value life in itself.

Advanced civilizations or AIs may create simulations of wildlife with sentient digital animals.

Some physicists theorize that it may be possible to create universes in a laboratory.

I don’t think that these scenarios are very likely, but they are still important because some of them could result in quadrillions of animals suffering on millions of planets for billions of years. Compared to the scenarios above, what happens with WAW on Earth in the far future seems relatively less important, as it only impacts one planet. Hence, perhaps WAW advocates should think more about how their actions affect these scenarios where wildlife is spread beyond Earth.

To reduce the probability of humans spreading of wildlife in a way that causes a lot of suffering, we could:

Directly argue about caring about WAW if humans ever spread wildlife beyond Earth

Lobby to expand the application of an existing international law that tries to protect other planets from being contaminated with Earth life by spacecrafts to planets outside of our solar system.

Continue building EA and WAW communities to ensure that there will be people in the future who care about WAW.

Spread the general concern for WAW (e.g., through WAW documentaries, outreach to academia).

However, I don’t think that the actions listed above should be a longtermist priority, partly because I think that digital minds are much more important. Digital minds can be much more efficient, thrive in environments where biological beings can’t, use many more resources, etc. Hence, I believe that they dominate in terms of importance in the far future. I think that spreading biological life is among the most important far-future considerations only if one believes with a very high credence that digital minds can’t or won’t be sentient. But EA is probably now big enough for it to be worthwhile for someone to do work on issues that might not be top priority.

In the last section of this article, I list open questions that could inform us on how to address far-future WAW concerns and could be researched with surveys.

At a few points in the article, I assume that spreading wildlife without addressing WAW issues might be undesirable. I explain why in the Why I’m against spreading wildlife section.

Everything in this article is very speculative. Many of the ideas I discuss have been written about before (particularly by Brian Tomasik), but I believe that I analyze some of them a bit deeper. Also, this is my first time looking into longtermist topics, so please point out in the comments if some of my thoughts or assumptions are naive.

Far-future WAW scenarios

Below, I analyze how WAW and WAW advocacy is relevant in various scenarios of the future.

Animals 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: Being a donor for Fecal Microbiota Transplants (FMT): Do good & earn easy money (up to 180k/y), published by Anton Rodenhauser on July 6, 2022 on The Effective Altruism Forum.

TLDR

Help to find super donors for Fecal Microbiota Transplants (FMT)! By donating stool for FMT, exceptionally healthy people can earn $180,000/yr with little effort. Simultaneously they can help chronically ill people improve their health with a novel treatment. Apply here.

More detailed summary

Fecal Microbiota Transplants (FMTs) is a procedure that transfers the stool of healthy people to the guts of sick people. The mechanism is to replace a dysbiotic gut microbiome with an eubiotic, disease-resistant gut microbiome.

So far, FMT has only been proven to be very effective in curing patients of C. difficile infections. However, since the gut microbiome impacts and regulates virtually every aspect of human health, function, and development, FMT is a promising treatment for a very wide range of chronic and acute health conditions.

FMTs can easily & safely be done at home, both for the donor and the recipient - no doctor needed.

A super-donor is a person who yields outstanding positive effects as a stool donor. Only exceptionally healthy & fit people are super-donors. They are very rare and a few people within the FMT movement have been searching for them for years. Most donors, even reasonable healthy ones, only yield modest benefits.

Human Microbes, an organization specialized in finding super-donors and connecting them with recipients, is paying super-donors $500 per stool. This adds up to $180,000/yr if donating a daily stool!

Purpose of this post

Spread the news that this opportunity for super donors exists: With little effort, they can help lots of ill people (“make them more effective”) while earning $500/stool. You should check whether you are a super donor. If you know someone who might be, please share this post with them.

Give a rough introduction to FMTs & Human Microbes, with the aim to start a rational discussion about FMT’s benefits. FMTs look very promising, but there is by no means a clear scientific consensus. I’d love for more EAs to look into this, to get more certainty about its efficacy!

If your health is suboptimal, maybe you could consider using FMTs. Not just for obviously gut health-related issues, but much more broadly than you might think. Potentially even e.g. unspecified subclinical low energy, low mood, or mental sluggishness.

At the end, this post provides you with a long list of useful further links on FMTs, the microbiome and its connection to a vast amount of (mental-) health conditions. This is to help you do your own research, both as a potential super donor or as someone who might benefit from FMTs.

Who might be a super donor?

Donors must be in exceptional physical and mental health. Ideally, top young athletes. Though there are non-athletes who qualify as well.

Ideally, donors would be under 30 years old. Donors can be under 18 provided they have signed consent from their parents. Many children need FMT and are ideally matched with young donors.

Donors should have minimal antibiotic use.

Donors should have a specific Bristol Stool Type.

Check out the links at the end of this post for more information and sources.

As you can imagine, super donors are very rare and thus very high in demand. In fact, Human Microbes has screened over 25,000 donor applicants so far, including hundreds of college and professional athletes, and still hasn’t found that one “perfect” super donor that fully satisfies all their donor criteria! So please reach out to Human Microbes about potential super donor candidates!!

How much good can you do by finding a super donor?

Know an EA whose gut-borne suboptimal health is holding them back from doing the most good they can do? Find them a...

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: Self investment I think community builders should do, published by Rob Gledhill on July 7, 2022 on The Effective Altruism Forum.

I frequently hear that community builders are dedicating so much time to their work, that they are not sufficiently prioritizing their self investment, or their own wellbeing. To try to establish some common knowledge, I want to share some things that I think

community builders should be regularly doing. I’ve tried to give a rough breakdown of the share of time dedicated to these - although I suspect there are many situations where these rough time allocations do not hold.

Self investment I think community builders should do

Rest (~10% of time)

Holiday - Community Building Grants are made in the expectation that people work ~normal working hours for their location. This includes taking holiday, without needing to “pay back” the hours. For people on other types of grants, I also think that the level of holiday that optimizes productivity for most people is >0

Example: EACH (EA Switzerland) make it explicit in their work contracts that they must take a certain amount of time off

Self development (~10% of time, more for those new to effective altruism)

Cause prioritization - If community builders are to be persuasive when they make the case for effective altruism, I think it is important that they have thought through their own opinions on cause prioritization. I would not be persuaded that I should invest a lot of time in learning and making detailed prioritizations if the person encouraging me had not done so themselves. Additionally, I think it’s motivating to feel you are working on something that is truly aligned with your values, and if you aren’t taking time to revisit the implications of your values, I don’t think this can be true.

Example: The coaching pilot for city and national groups led to multiple group leaders committing to re-evaluating their cause prioritization

Learning - In the recent post Bad Omens in Community Building the author explains “...in the first session, foundational arguments were laid out, and he was encouraged to offer criticism. So he did. According to him, the organisers were grateful for the criticism, but didn’t really give him any satisfying replies.” This seems bad, and I think it is worthwhile investing in learning to reduce incidents like this.

Example: In April and May of this year I read the texts for EA Cambridge’s AGI SF

Improving working practices (~5%)

Predictions - As CEA’s CBG manager, I get to read the strategies and plans of many different groups. It is exceedingly rare that these strategies or plans make explicit predictions about the impact that they will have. Planning is hard, and by not making predictions, community builders are depriving themselves of a major source of information about their planning.

Example: EA Netherlands make predictions about tactical decisions that they are uncertain on, to try and improve over time

Working on the collaboration or ways of working - For community builders that work in teams, I think it is important to take time to review how the collaboration is going. What do you enjoy about working together? What parts of the collaboration are frustrating? Are there things you can change to collaborate better in the future?

Examples: Organisers sharing their self perception of: Strengths, weaknesses, how they like to receive feedback, personal development objectives (and how the rest of the team can support this); Habit tracking; Identifying and tracking growth areas

Taking the above into account, this then implies 75% of your time should be on object level community building.

What I suspect is blocking self-investment

Start-up-ness: Community builders are working in small organisations without fully defined processes, which can lead to a sense of "as soon as we...

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: Babysitting as Parenting Trial?, published by jefftk on July 7, 2022 on LessWrong.

Julia recently posted an

interview with Kat Woods on how Kat has thought through decisions around having kids. In her 20s she started strongly wanting to:

I didn't want kids when I was younger, and then when I was about 20 I was in the park, minding my own business and a little toddler toddled up to me and he gave me this bent daisy. And my brain exploded! I was like, "Oh my God, I need babies now."

I just became absolutely obsessed. EA was my number one priority, and the next one was kids. I actually had a blog about parenting. I was reading all these books about it, reading studies and figuring out exactly how to deal with pregnancy and everything. I'd sit out outside of playgrounds and do my homework next to the playground so I could watch the kids play, just be around them.

Then she spent some time trying to understand her motivations better, and decided to try babysitting as a way to get practical experience with more aspects of taking care of kids:

Oh man, I get to spend time with the kids and develop this relationship with them. But also I get paid for it. Amazing!

I babysat three separate sets of kids, and they were all perfectly ordinary kids. They were not kids with behavior problems or anything like that. But that was enough for me to realize: oh, I do not want to be a parent. I do not want children at all.

Her advice to others based on this experience is to try babysitting before having kids:

I think the lesson from this that I took away is that before you have kids, it's really good to get some babysitting experience. It's a really good way to cheaply test being as close to being a parent as you can be without actually being a parent, that's pretty available to most people. ... I just want people to be informed, and do some quick, easy tests before they commit their lives to new beings.

In many ways this makes a lot of sense: the decision to have kids is a serious one with lifelong effects, and you want to make it with lots of information about yourself and how it's likely to go. Your responsibilities in the moment as a babysitter pretty similar to as a parent: taking care of the kid, and dealing with whatever comes up. And yet, I like parenting a lot and didn't enjoy babysitting. If I had used this method to decide whether to become a parent, I think I would have decided against it, and that would've been the wrong choice for me. I'm not completely sure, because, not liking babysitting, I didn't do very much of it, but I wanted to think through a bit where this difference is coming from.

Imagine you're considering whether to start a band. You decide to try it out by taking some gigs subbing in with other groups. You were initially excited about the idea, but when you're actually playing with them it's not much fun: the group is not gelling, you wouldn't have picked these songs, and this isn't really your kind of music anyway.

Parenting is made up of many small interactions, and when things are going well many of those interactions are individually fun or rewarding, but it's also a long-term project. You're helping this person who starts off completely incompetent and utterly dependent grow into the kind of person they want to become. When I'm interacting with someone else's child, as a friend, sitter, or even relative, I don't have that level of involvement.

In my experience (n=2, replication forthcoming) if you're thoughtful about those individual interactions, paying attention to what's working and what isn't, being predictable and consistent, and listening to your kids, you have fewer of the kinds of disappointing interactions Kat talks about in the interview. Which isn't to say that I don't ever have rough betimes with the kids or realize partway through an explanatio...

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: Aversion Factoring, published by CFAR!Duncan on July 7, 2022 on LessWrong.

Epistemic status: Anecdotally strong

The aversion factoring technique is not derived from any particular body of psych research, though it draws lightly on trigger-affect patterns and exposure therapy and takes advantage of the framework of reductionism. It was primarily developed for personal use by Andrew Critch, and then later shared and refined through iteration.

It’s quite common for people to talk about their aversions:

“Oh, I just hate going to the gym.”

“I never talk on the phone if I can help it.”

“They require a six-month commitment, and I really didn’t want to be locked into anything for that long.”

“I don’t know what it is. Something about his. . . attitude, I guess? The way he talks to people? Look, I’d just really rather you not invite him.”

Aversions lead to avoidance—they’re any sort of mental mechanism that causes us to be less likely to engage in a particular activity, or to do so only with pain, displeasure, or discomfort. Aversions can be conscious or unconscious, reasoned or felt, verbal or visceral, and they can range anywhere from a slight tinge of antipathy to outright phobias.

The purpose of the Aversion Factoring technique is to give you the tools you’ll need to identify and overcome aversions. Of course, not every aversion should be overcome—it would probably be counterproductive to lose your aversion to being hit by cars, for example—but there are many activities we might engage in and enjoy if we could just get past their one sticking point, be it self-consciousness or tedium or the aftertaste of mushrooms.

Reductionism: the LEGO principle

In the Goal Factoring section of this sequence, we attempted to draw out the positive elements of various plans, and weave them together into a new strategy that had fewer drawbacks (or none). In Aversion Factoring, we’ll take a slightly different approach, holding the larger context constant while we address specific aspects of it one at a time.

The key insight is that no activity simply is aversive, at its core. There’s no fundamental quantity called “running” that “just sucks,” for instance—running is a complex system of experiences, and summing or averaging across all of them causes us to lose valuable detail. Treating each experience separately, we may find that we’re more or less okay with all of them, and that the ones that are the most negative can be addressed individually.

Wearing running shoes and athletic clothes (verdict: seems fine)

Feet slapping against the pavement (verdict: kind of unpleasant?)

Legs and arms pumping, muscles burning (verdict: no problem)

Shins and knees aching (verdict: VERY BAD)

Rapid heart rate and breathing (verdict: not my favorite part of exercising, but it’s also not really a big deal)

Being outside on sidewalks in the heat (verdict: problematic)

Sweating until my shirt sticks to me (verdict: who cares?)

Wind in my face and hair (verdict: great!)

Feeling fast and light (verdict: AWESOME)

Other people looking at me and judging me (verdict: definitely bad)

This is the "factoring" part of the process—an application of the LEGO principle, which Goal Factoring and Aversion Factoring have in common. In the example above, what was previously just a vague sense that running is bad has been clarified into a much more specific set of issues, starting with aching shins and knees and including self-consciousness and a little bit of a negative reaction to the environment. Before, the obvious intervention was “don’t run;” now, this person has the necessary information to find a specific running strategy that works (if they want to).

Case study: the tenacious tree-climber

A few years ago, CFAR instructor Andrew Critch was looking to add more physical activity into his daily routine, and realized that it 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: Legal Impact for Chickens files its first lawsuit! Costco shareholder derivative case, published by alene on July 7, 2022 on The Effective Altruism Forum.

Dear Wonderful EAs,

Thank you so much to everyone in the EA community who has contributed to the launch of Legal Impact for Chickens (LIC).

I wanted to let you know that LIC has filed our first lawsuit!

LIC is representing two Costco shareholders. The shareholders are suing Costco executives for causing the company to violate Nebraska and Iowa laws against livestock neglect.

Our complaint follows a 2021 undercover investigation revealing that chickens used by Costco are bred to grow so fast that many of them can’t stand under their own weight. The complaint alleges that many of these young, disabled birds are unable to reach food or water, and that the company fails to provide them with individualized medical care. The complaint accuses Costco’s executives of violating their basic fiduciary duty to act lawfully.

Costco is responsible for the lives and deaths of approximately 100 million chickens per year.

Our lawsuit has already gotten coverage in The Washington Post, People, Yahoo News, Fox Business, CNN Business, The New York Post, Meatingplace, Watt Ag Net, and more.

Thank you again, so much, to everyone who helped make this possible!

Sincerely,

Alene and Legal Impact for Chickens

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: Safety considerations for online generative modeling, published by Sam Marks on July 7, 2022 on The AI Alignment Forum.

Summary: the online decision transformer is a recent approach to creating agents in which a decision transformer is pre-trained offline (as usual) before producing its own trajectories which are fed back into the model in an online finetuning phase. I argue that agents made with generative modeling have safety advantages – but capabilities disadvantages – over agents made with other RL approaches, and agents made with online generative modeling (like the online decision transformer) may maintain these safety advantages while being closer to parity in capabilities. I propose experiments to test all this. (There is also an appendix discussing the connections between some of these ideas and KL-regularized RL.)

I’ll start with some motivation and then introduce the decision transformer and the online decision transformer. If you’re already familiar with the decision transformer, you can probably skip to “Online generative modeling.” If you’re already familiar with the online decision transformer, you can skip to “Some remarks.”

Motivation

Suppose we want to get an AI system to produce a picture of a cat.

A naive approach is to take a trained image classifier M and optimize an image to maximally activate M's 'cat' classification. If you do this, you won't get anything that looks like a cat. Instead, you'll get some deep-dream-esque conglomeration of whiskers, cat ears, and fur which M happens to strongly classify as 'cat.' One way of describing what went wrong here is that the image of the cat you get is very off-distribution. You would rather get a normal-looking picture of a cat, i.e. one which is on-distribution for M's training data.

This can be taken as an analogy for alignment: if you have an agent optimize the world for a given utility function U, what you get won't look anything like a normal world, but rather a very weird-looking world which happens to maximize U. We would prefer to get a normal-looking world which also scores high according to U.

We've recently seen lots of reminders that modern ML does have ways to produce normal-looking pictures of cats. A key ingredient in all of these approaches is generative modeling. Recall that a generative model is something which is fed in data and learns to produce new data which "looks similar" to its training data; in other words, a generative model tries to produce new samples from its training distribution. Think GPT, which tries to produce text which looks like the text it's been trained on.

Generative models can also be conditioned on information which, roughly speaking, tells the generative model which part of its training distribution to sample from (e.g. conditioning GPT on a prompt tells it to sample from the part of its training distribution consisting of texts which start with that prompt; a caption given to DALL-E tells it to sample from the part of its training distribution consisting of images which would be given that caption). So to get a normal-looking picture of a cat, train a generative model on lots of images and captions, and then ask the generative model to sample an image from its training distribution, conditional on that image being one which would be captioned ‘cat.’

If "produce a normal-looking picture of a cat" is an analogy for alignment and generative models solve the “produce a normal-looking picture of a cat” problem, then what does an agent built via generative modeling look like?

Making agents via generative modeling

It looks like a decision transformer.

Recall that decision transformers work as follows. Suppose we want to train an agent to play the Atari game Breakout. We start by encoding the game states, rewards, and actions as tokens. We then treat playing Breako...