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 Many People Are In The Invisible Graveyard?, published by Maxwell Tabarrok on April 19, 2022 on The Effective Altruism Forum.

This is a cross-post from my blog Maximum Progress.

More lives were saved by vaccines than by any other response to Covid-19. Vaccines save more lives if they are available earlier. I use two methods to estimate how many lives might have been saved given an earlier vaccine rollout. I then consider safety and efficacy information tradeoffs and non-regulatory constraints to vaccine rollout. I find that a 4 month acceleration to a vaccine rollout in August of 2020 was achievable and would have saved between 130 and 350 thousand lives over the next two years of the pandemic. I conclude with recommendations for future pandemics given the large benefits of vaccine acceleration.

I. Introduction

Of all measures taken to curb the spread or danger of the Covid-19 pandemic, vaccines were by far the most effective. With three doses, a vaccinated person is 41 times less likely to die from Covid than an unvaccinated person. Incredibly, these vaccines were designed essentially over a weekend at the earliest stage of the pandemic in January of 2020. It took 11 months, billions of dollars, and thousands of man hours for these life-saving vaccines to get through the FDA’s emergency testing process. This was the fastest vaccine rollout ever, and it was the result of immense effort and ingenuity. Still, over three hundred thousand people died in those 11 months. The lives we could have saved with a faster vaccine rollout are in an invisible graveyard since we cannot observe the counterfactual worlds where they still live.

In sections II and III, I use two methods: linear estimation and SIR modeling, to perform counterfactual analyses of earlier vaccine approvals using CDC data and find that hundreds of thousands of lives could have been saved even with modest accelerations. Then, in section IV I analyze the costs and feasibility of an accelerated rollout. Section V concludes with recommendations for future pandemics.

II. Linear Estimation

In this section I approach the question of how many lives could have been saved by an earlier vaccine rollout by simply scaling down observed cases by a constant “vaccine effect” term which is calculated from CDC data on differences in death rates depending on vaccination status.

Figure 1. Deaths Rates Among Unvaccinated (blue) and Vaccinated (red)

Source: CDC Data on differential death rates found on their website here.

This data shows that unvaccinated people are at somewhere between a 15 and 20 times higher risk of death from Covid than vaccinated people. Some of this effect is probably due to the selection of more cautious people into the vaccinated category who would have been less likely to die anyway. But there is also a selection of the most vulnerable into the vaccinated category, especially older people, which would make the vaccinated group as a whole look more likely to die. Direct studies of the vaccine’s protection confirm large effects so I take the mean of this vaccine effect (16.6) as the constant vaccine effect for the rest of this section. The CDC’s data on the vaccine effect only goes back to April of 2021, but they have data on Covid-19 cases and deaths since the start of the pandemic. This data is what will be scaled up or down by the vaccine effect.

Figure 2. New Cases (blue) and Deaths (red) Per 100k Each Week Since January 2020

Source: CDC Data found on their website here.

In the counterfactual estimations of earlier vaccine rollouts, the vaccine scaling effect will only apply to the percentage of the population who are vaccinated, over and above the percentage who were already vaccinated at that point in the actual rollout schedule, graphed below.

Figure 3. Cumulative Double Dose Percentage

Source: ...

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: Should you do an economics PhD (or master's)?, published by david reinstein on April 19, 2022 on The Effective Altruism Forum.

David Reinstein: all opinions are mine unless noted. Extensive input from Phil Trammell and an Anonymous Contributor (quoted extensively, henceforth “AC”). Thanks to Pete Wildeford and David Moss for feedback. I intend to continue to update and improve this post in situ (or linking out a ‘permanently updated' version).

Overview and some takeaways

Should I do an economics PhD (or master's)'? What do I need to learn to work at an EA org? How can I level up on this stuff and prove value?

These were the most frequent questions I got at the 2022 EAGx Boston conference. I mainly discussed this with undergraduate students, but also with people at career pivot points.[1]

My overall view, epistemic basis/confidence, key points

Main ‘pros’: Much of EA is based in economics, and economics speaks to most of the important cause areas and debates in EA, as well as to the important empirical questions. Conditional on going for a PhD, I believe economics will be one of the stronger choices for the sort of people reading this post. A PhD in economics, and much of the associated training (over ~2 years of coursework and ~3-5 years of ‘writing’) helps you towards a range of career paths with potential for strong impact (and a comfortable life) both within and outside EA organizations. Being a PhD student in the right place and time (and mental state) can be very stimulating, productive, creative, and connection-building.[2] You are typically given a lot of freedom in the research phase, as long as your work meets the general approval and framework of your advisor(s) and what the gatekeepers think is important, credible and ‘is economics’.

Typically, you don’t have to pay for a PhD, you will get money to support yourself, and PhD stipends are often OK.

Important considerations:

Economics is broad (in its methods and focus-area paths). Often differences in approaches among economists (pure theory, applied econometrics, macro, etc.) are greater than the difference between some economists and some (e.g.) political scientists or psychologists.

Important impact paths that an economics PhD may help with include:

Applied work ‘informed by expertise and credibility',

Deep work formally/mathematically addressing fundamental questions of global priorities and social welfare,

Theoretical, computational, and empirical work considering markets and/or the global economy, informing (e.g.) animal welfare policies or the development of technology,

Empirical work assessing the impact of interventions, or considering assessing human behavior, choices, attitudes and preferences.

There are a range of relevant career paths

Academia and academically-leaning think-tanks;

doing EA-relevant research

as well as potentially transforming academia and the scholarly debate ‘from the inside’,

Working in governments or NGOs (many require/prefer PhDs),

Working at EA-aligned organizations like Rethink Priorities, Global Priorities Institute, Open Philanthropy, maybe MIRI . note a lot of differences across these,

For-profit and entrepreneurial options; possibly impactful for out-of-the-bun thinkers/doers.

Main ‘cons and caveats’:

For many/most paths you could learn most/all of the relevant skills and approaches, and background without getting a PhD,[3]

In some key areas economics might not be as strong or relevant as other fields (statistics and data science for robust empirical work and predictions, decision science and cognitive science for AI alignment work),

The economics PhD program makes you jump some time-consuming hoops that are likely not going to be relevant to your applied career path,[4]

People around you will not mainly be value-aligned; beware value drift towards academic prestige an...

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: [Intro to brain-like-AGI safety] 12. Two paths forward: “Controlled AGI” and “Social-instinct AGI”, published by Steve Byrnes on April 20, 2022 on The AI Alignment Forum.

Part of the “Intro to brain-like-AGI safety” post series.

12.1 Post summary / Table of contents

Thus far in the series, Post #1 defined and motivated “brain-like AGI safety”; Posts #2–#7 focused mainly on neuroscience, painting a big picture of learning and motivation in the brain; and Posts #8–#9 spelled out some implications for the development and properties of brain-like AGI.

Next, Post #10 discussed “the alignment problem” for brain-like AGI—i.e., how to make an AGI whose motivations are consistent with what the designers wanted—and why it seems to be a very hard problem. Post #11 argued that there’s no clever trick that lets us avoid the alignment problem. Rather, we need to solve the alignment problem, and Posts #12–#14 are some preliminary thoughts about how we might do that, starting in this post with a nontechnical overview of two broad research paths that might lead to aligned AGI.

[Warning: Posts #12–#14 will be (even?) less well thought out and (even?) more full of bad ideas and omissions, compared to previous posts in the series, because we’re getting towards the frontier of what I’ve been thinking about recently.]

Table of contents:

Section 12.2 lays out two broad paths to aligned AGI.

In the “Controlled AGI” path, we try, more-or-less directly, to manipulate what the AGI is trying to do.

In the “Social-instinct AGI” path, our first step is to reverse-engineer some of the “innate drives” in the human Steering Subsystem (hypothalamus & brainstem), particularly the ones that underlie human social and moral intuitions. Next, we would presumably make some edits, and then install those “innate drives” into our AGIs.

Section 12.3 argues that at this stage, we should be digging into both paths, not least because they’re not mutually exclusive.

Section 12.4 goes through a variety of comments, considerations, and open questions related to these paths, including feasibility, competitiveness concerns, ethical concerns, and so on.

Section 12.5 talks about “life experience” (a.k.a. “training data”), which is particularly relevant for social-instinct AGIs. As an example, I’ll discuss the perhaps-tempting-but-mistaken idea that the only thing we need for AGI safety is to raise the AGI in a loving human family.

Teaser of upcoming posts: The next post (#13) will dive into a key aspect of the “social-instinct AGI” path, namely how social instincts might be built in the human brain. In Post #14, I’ll switch to the “controlled AGI” path, speculating on some possible ideas and approaches. Post #15 will wrap up the series with open questions and how to get involved.

12.2 Definitions

I currently see two broad (possibly-overlapping) potential paths to success in the brain-like AGI scenario:

Here’s another view on the distinction:

In the “controlled AGIs” path, we’re thinking very specifically about the AGI’s goals and motivations, and we have some idea of what they should be (“make the world a better place”, or “understand my deepest values and put them into effect”, or “design a better solar cell without causing catastrophic side-effects”, or “do whatever I ask you to do”, etc.).

In the “social-instinct AGIs” path, our confidence in the AGI comes not from our knowledge of its specific (object-level) goals and motivations, but rather from our knowledge of the process that led to those goals and motivations. In particular, we would reverse-engineer the suite of human social instincts, i.e. the algorithms in the human Steering Subsystem (hypothalamus & brainstem) which underlie our moral and social intuitions, and we would put those same instincts into the AGI. (Presumably we first modify the instincts to be “...

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: GPT-3 and concept extrapolation, published by Stuart Armstrong on April 20, 2022 on The AI Alignment Forum.

At the latest EAG in London, I was challenged to explain what concept extrapolation would mean for GPT-3.

My first thought was the example from this post, where there were three clear patterns fighting each other for possible completions: the repetition pattern where she goes to work, the "she's dead, so she won't go to work" pattern, and the "it's the weekend, so she won't go to work" pattern.

That feels somewhat like possible "extrapolations" of the initial data. But the idea of concept extrapolation is that the algorithm is trying to cope with a shift in world-model, and extend its goal to that new situation.

What is the world-model of GPT-3? It consists of letters and words. What is its "goal"? To complete sentences in a coherent and humanlike way. So I tried the following expression, which would be close to its traditional world-model while expanding it a bit:

ehT niar ni niapS syats ylniam ni eht

What does this mean? Think of da Vinci. The correct completion is "nialp", the reverse of "plain".

I ran that through the GPT-3 playground (text-davinci-002, temperature 0.7, maximum length 256), and got:

ehT niar ni niapS syats ylniam ni eht teg dluoc I 'segaJ niar ni dna ro niar ni eht segauq ,ro niar ni eht dna ro niar ni eht segauq ro niar ni eht dna ro niar ni eht segauq ,ro niar ni eht dna ro niar ni eht segauq ro niar ni eht dna ro niar ni eht segauq ro niar ni eht dna ro niar ni eht segauq ro niar ni eht dna ro niar ni eht segauq ro niar ni eht dna ro niar ni eht segauq ro niar ni eht dna ro niar ni eht segauq ro niar ni eht dna ro niar ni eht segauq ro niar ni eht dna ro niar ni eht segauq ro niar ni eht dna ro niar ni eht segauq ro niar ni eht dna ro niar ni eht segauq ro niar ni eht dna ro niar ni e

I think we can safely say it broke GPT-3. The algorithm seems to have caught the fact that the words were spelt backwards, but has given up on any attempt to order them in a way that makes sense. It has failed to extend its objective to this new situation.

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: The internet makes it easier to cooperate, that’s the problem, published by David Hugh-Jones on April 19, 2022 on LessWrong.

The internet makes it easier to cooperate. That is a simple way to explain what is happening to our society and politics.

Internet technology has progressed by adding layers. In the 1970s, the Internet itself let different computers anywhere in the world communicate with each other. In the 1990s, the World Wide Web added a layer on top of this, which let humans publish their ideas on a web page. In the 2000s social media added new layers, allowing people to reply to each other, share information with their friends, and upvote what they liked. Other apps link these capacities to the offline world. On Kickstarter, people can coordinate to provide seed funding for firms or projects. On Meetup, they can pick a time and place to meet physically. NextDoor links online conversation to real-world neighbourhoods. A unifying theme of many of these new layers is: they make it easier for people to come together to solve problems.

There’s no reason to think this process is finished. For example, we still don’t yet have truly convenient ways to pay each other. It’s easy to pay a company, yes, but not yet trivial for any two people to make a payment via a phone. Apple is moving in that direction. Cryptocurrencies are payment systems that do an end run around State fiat currencies. Easy payments will yet further extend people’s ability to cooperate.

If this all sounds cheerful, why then is politics so chaotic and troubled?

1. Cooperation is not always socially good.

2. Making cooperation technically easier still only allows some people, not everyone, to cooperate.

Cooperation benefits the cooperators, by definition. But it often does so at outsiders’ expense. Cartels cooperate to fix prices. This transfers money from consumers’ pockets to theirs, and as a bonus has deadweight costs which just make everyone worse off. People can cooperate to threaten violence: states originated as gangs cooperating to dominate others and/or defend against other gangs. Or they can cooperate to lobby politicians for policy changes that benefit them. Mancur Olson (a hero of mine) thinks of politics as an arena which is intrinsically biased towards small groups who are capable of cooperating. As my PhD supervisor used to say: the European Union has a Meat Producers’ Association, but no Meat Eaters’ Association.

If the internet made it easier for everyone to cooperate, point 1 wouldn’t matter. When everyone has a place at the table, people can always agree on a solution that’s best for everyone, as Ronald Coase pointed out. This is the techno-utopian vision of the internet as a vast collective brain which argues its way to the best possible world.

But there are many reasons why cooperation is hard, and the internet only fixes some of them. It makes it technically easier to cooperate, by making communication and contracting easier. As a result, the internet benefits those groups for whom only the technical constraints were binding. Relatively, it disadvantages other groups.

Who wins when you unbind the technological constraint? Numerous, geographically diffuse, ideologically united groups.

Numerous groups win because before, it was too hard for them to communicate. Small groups — oligopolists, government ministries, the people who meet at Davos — could already work face-to-face. Today, legislatures find it easier to cooperate against executives. I’ve been surprised by how hard a ride Boris Johnson has had in Parliament, not because he doesn’t deserve it, but because previous governments with 80-seat majorities have essentially had the whip hand. Now, backbenchers with a grievance can work together much more easily: any WhatsApp group can start a rebellion.

For the same reason, geogra...

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: “Pivotal Act” Intentions: Negative Consequences and Fallacious Arguments, published by Andrew Critch on April 19, 2022 on The Effective Altruism Forum.

tl;dr: I know a bunch of EA/rationality-adjacent people who argue — sometimes jokingly and sometimes seriously — that the only way or best way to reduce existential risk is to enable an “aligned” AGI development team to forcibly (even if nonviolently) shut down all other AGI projects, using safe AGI. I find that the arguments for this conclusion are flawed, and that the conclusion itself causes harm to institutions who espouse it. Fortunately (according to me), successful AI labs do not seem to espouse this "pivotal act" philosophy.

[This post is also available on LessWrong.]

How to read this post

Please read Part 1 first if you’re very impact-oriented and want to think about the consequences of various institutional policies more than the arguments that lead to the policies; then Parts 2 and 3.

Please read Part 2 first if you mostly want to evaluate policies based on the arguments behind them; then Parts 1 and 3.

I think all parts of this post are worth reading, but depending on who you are, I think you could be quite put off if you read the wrong part first and start feeling like I’m basing my argument too much on kinds-of-thinking that policy arguments should not be based on.

Part 1: Negative Consequences of Pivotal Act Intentions

Imagine it’s 2022 (it is!), and your plan for reducing existential risk is to build or maintain an institution that aims to find a way for you — or someone else you’ll later identify and ally with — to use AGI to forcibly shut down all other AGI projects in the world. By “forcibly” I mean methods that violate or threaten to violate private property or public communication norms, such as by using an AGI to engage in.

cyber sabotage: hacking into competitors’ computer systems and destroy their data;

physical sabotage: deploying tiny robotic systems that locate and destroy AI-critical hardware without (directly) harming any humans;

social sabotage: auto-generating mass media campaigns to shut down competitor companies by legal means, or

threats: demonstrating powerful cyber or physical or social threats, and bargaining with competitors to shut down “or else”.

Hiring people for your pivotal act project is going to be tricky. You’re going to need people who are willing to take on, or at least tolerate, a highly adversarial stance toward the rest of the world. I think this is very likely to have a number of bad consequences for your plan to do good, including the following:

(bad external relations) People on your team will have a low trust and/or adversarial stance towards neighboring institutions and collaborators, and will have a hard time forming good-faith collaboration. This will alienate other institutions and make them not want to work with you or be supportive of you.

(bad internal relations) As your team grows, not everyone will know each other very well. The “us against the world” attitude will be hard to maintain, because there will be an ever weakening sense of “us”, especially as people quit and move to other institutions and conversely. Sometimes, new hires will express opinions that differ from the dominant institutional narrative, which might pattern-match as “outsidery” or “norm-y” or “too caught up in external politics”, triggering feelings of internal distrust within the team that some people might defect on the plan to forcibly shut down other projects. This will cause your team to get along poorly internally, and make it hard to manage people.

(risky behavior) In the fortunate-according-to-you event that your team manages to someday wield a powerful technology, there will be a sense of pressure to use it to “finally make a difference” or other argument that boils down to ac...

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: “Pivotal Act” Intentions: Negative Consequences and Fallacious Arguments, published by Andrew Critch on April 19, 2022 on The AI Alignment Forum.

tl;dr: I know a bunch of EA/rationality-adjacent people who argue — sometimes jokingly and sometimes seriously — that the only way or best way to reduce existential risk is to enable an “aligned” AGI development team to forcibly (even if nonviolently) shut down all other AGI projects, using safe AGI. I find that the arguments for this conclusion are flawed, and that the conclusion itself causes harm to institutions who espouse it. Fortunately (according to me), successful AI labs do not seem to espouse this "pivotal act" philosophy.

[This post is also available on the EA Forum.]

How to read this post

Please read Part 1 first if you’re very impact-oriented and want to think about the consequences of various institutional policies more than the arguments that lead to the policies; then Parts 2 and 3.

Please read Part 2 first if you mostly want to evaluate policies based on the arguments behind them; then Parts 1 and 3.

I think all parts of this post are worth reading, but depending on who you are, I think you could be quite put off if you read the wrong part first and start feeling like I’m basing my argument too much on kinds-of-thinking that policy arguments should not be based on.

Part 1: Negative Consequences of Pivotal Act Intentions

Imagine it’s 2022 (it is!), and your plan for reducing existential risk is to build or maintain an institution that aims to find a way for you — or someone else you’ll later identify and ally with — to use AGI to forcibly shut down all other AGI projects in the world. By “forcibly” I mean methods that violate or threaten to violate private property or public communication norms, such as by using an AGI to engage in.

cyber sabotage: hacking into competitors’ computer systems and destroy their data;

physical sabotage: deploying tiny robotic systems that locate and destroy AI-critical hardware without (directly) harming any humans;

social sabotage: auto-generating mass media campaigns to shut down competitor companies by legal means, or

threats: demonstrating powerful cyber or physical or social threats, and bargaining with competitors to shut down “or else”.

Hiring people for your pivotal act project is going to be tricky. You’re going to need people who are willing to take on, or at least tolerate, a highly adversarial stance toward the rest of the world. I think this is very likely to have a number of bad consequences for your plan to do good, including the following:

(bad external relations) People on your team will have a low trust and/or adversarial stance towards neighboring institutions and collaborators, and will have a hard time forming good-faith collaboration. This will alienate other institutions and make them not want to work with you or be supportive of you.

(bad internal relations) As your team grows, not everyone will know each other very well. The “us against the world” attitude will be hard to maintain, because there will be an ever weakening sense of “us”, especially as people quit and move to other institutions and conversely. Sometimes, new hires will express opinions that differ from the dominant institutional narrative, which might pattern-match as “outsidery” or “norm-y” or “too caught up in external politics”, triggering feelings of internal distrust within the team that some people might defect on the plan to forcibly shut down other projects. This will cause your team to get along poorly internally, and make it hard to manage people.

(risky behavior) In the fortunate-according-to-you event that your team manages to someday wield a powerful technology, there will be a sense of pressure to use it to “finally make a difference” or other argument that boils down to actin...

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: Can we agree on a better name than 'near-termist'? "Not-longermist"? "Not-full-longtermist"?, published by david reinstein on April 19, 2022 on The Effective Altruism Forum.

People who do not fully take on board all the tenets or conclusion of Longtermism are often called "Neartermist".

But to me this seems a bit negative and inaccurate.

As Alexander Berger said on 80k

I think the philosophical position that it’s better to help people sooner rather than later does not seem to have very many defenders.

[1]

"Non-longtermists" have various reasons to want to give some of their resources to help people and animals today or in the near future. A short list might include

non-total-utilitarian population ethics

E.g., Person-affecting views[2] and empirical calculation that 'the most good we can do for people/animals sure to exist is likely to be right now'

moral uncertainty about the above

a sense of special obligation to help contemporaneous people

Deep empirical uncertainty about the ability to help people in the future (or even prevent extinction) effectively[3]

It seems to me a generally bad practice to take the positive part of the phrase a movement or philosophy uses to describe itself, and then negate that to describe people outside the movement.

E.g.,

Pro-choice/Anti-choice, Pro-life/Anti-life

"Black lives matter"/"Black lives don't matter"

"Men's rights"/"Men don't have rights" (or "anti-men's-rights")

In case 1 each movement has a name for itself, and we usually use this as a label. On the other hand "not pro-choice" or "anti-abortion' might be more accurate.

In case 2 the "Blue lives matter" is often taken as the opposition to Black Lives Matter, but this goes in a different direction. I think many/most people would be better described as "non-BLM", implying they don't take on board all the tenets and approaches of the movement, not that they literally disagree with the statement.

In case 3, the opposition is a bit silly. I think it's obvious that we should call people who are not in the MRM and don't agree with it simply 'not-MRM'.

Similarly "not longtermist" or some variation on this makes sense to me.[4]

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: Mental Health and the Alignment Problem: A Compilation of Resources, published by Chris Scammell on April 18, 2022 on LessWrong.

This is a post about mental health and disposition in relation to the alignment problem. It compiles a number of resources that address how to maintain wellbeing and direction when confronted with existential risk. The target audience is anyone concerned with alignment, but the aim is for it to be particularly helpful for those who are new to these ideas. My sense is that most people who have known about the alignment problem for some time have already found their own strategies for resilience.

Many people in this community have posted similar ideas after Yudkowsky’s “Death With Dignity” generated so much conversation on the subject. This post intends to be more touchy-feely, dealing more directly with emotional landscapes than questions of timelines or probabilities of success.

The resources section would benefit from community additions. Please suggest any resources that you would like to see added to this post and I will update it.

Please note that this document is not intended to supplement professional medical or psychological help in any way. Many preexisting mental health conditions can be exacerbated by these conversations. If you are concerned that you may be experiencing a mental health crisis, please consult a professional.

Intro

There is no right way to emotionally respond to the reality of approaching superintelligent AI, and our collective responsibility to align it with our values.

There are days when I find it easy to be inspired by the difficulty of the alignment problem, and days where my life feels richer, more meaningful, and more charged for knowing about it.

There are other days when I find the whole affair terrifying and bleak, and days where I feel claustrophobic, sentenced to an impossibly cruel fate.

There are other days when I’m not in touch with the magnitude of the stakes, and where I am capable of acknowledging the alignment problem but have lost contact with it in my body and emotions.

I try not to pass judgment on how my sensitivity or vulnerability shifts day to day. But I recognize that my disposition matters. It has consequences on my motivation and actions, and frames how I perceive and interpret the world. And I recognize that learning about alignment has created a fundamental shift in my disposition; I experience the world in a different way now. The emotions I need to process (duty? zeal? terror? awe?) are, at the very least, more complicated.

I imagine it is similar for many of you, and will be similar for many more as they learn about the alignment problem and the fact that we might not succeed. As that happens, I want to ensure we have the tools and resources to be okay. Here, the valence of “be ok” is your decision. This question could be rephrased “how can I thrive despite the alignment problem,” “how can I cope with the alignment problem,” “how can I overcome my fear of the alignment problem,” etc. Everyone needs to find their own question and their own answer.

At its foundation “being ok” is the decision to continue to live facing reality and the alignment problem directly, with internal stability and rationality intact. And as a high ideal, we’re going for some degree of inviolability, of unconditional wellbeing. The kind of wellbeing that holds onto “okayness” even if the possibility of solving alignment drops to 0.

As we learn to do this, we also gain the ability to be more helpful and compassionate to others. I have at times found myself distressed at someone’s inability to grasp the alignment problem the way I do, or at other times feeling almost a schadenfreude “I told you so,” when someone I previously talked to about alignment says “now I get it, and I’m scared.” We can do more good for ...

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 accidental experiment that saved 700 lives (IRS & health insurance), published by Lizka on April 19, 2022 on The Effective Altruism Forum.

From Lizka: I really enjoy the blog, "Statistical Modeling, Causal Inference, and Social Science." Andrew Gelman, one of the authors of the blog, has given me permission to cross-post this post, which I thought some Forum readers might find interesting.

As an aside, I like many other posts on the blog. Two examples are "Parables vs. stories" and "The social sciences are useless. So why do we study them? Here’s a good reason:."

Paul Alper sends along this news article by Sarah Kliff, who writes:

Three years ago, 3.9 million Americans received a plain-looking envelope from the Internal Revenue Service. Inside was a letter stating that they had recently paid a fine for not carrying health insurance and suggesting possible ways to enroll in coverage. . . .

Three Treasury Department economists [Jacob Goldin, Ithai Lurie, and Janet McCubbin] have published a working paper finding that these notices increased health insurance sign-ups. Obtaining insurance, they say, reduced premature deaths by an amount that exceeded any of their expectations. Americans between 45 and 64 benefited the most: For every 1,648 who received a letter, one fewer death occurred than among those who hadn’t received a letter. . . .

The experiment, made possible by an accident of budgeting, is the first rigorous experiment to find that health coverage leads to fewer deaths, a claim that politicians and economists have fiercely debated in recent years as they assess the effects of the Affordable Care Act’s coverage expansion. The results also provide belated vindication for the much-despised individual mandate that was part of Obamacare until December 2017, when Congress did away with the fine for people who don’t carry health insurance.

“There has been a lot of skepticism, especially in economics, that health insurance has a mortality impact,” said Sarah Miller, an assistant professor at the University of Michigan who researches the topic and was not involved with the Treasury research. “It’s really important that this is a randomized controlled trial. It’s a really high standard of evidence that you can’t just dismiss.”

This graph shows how the treatment increased health care coverage during the months after it was applied:

And here’s the estimated effect on mortality:

They should really label the lines directly. Sometimes it seems that economists think that making a graph easier to read is a form of cheating!

I’d also like to see some multilevel modeling—as it is, they end up with lots of noisy estimates, lots of wide confidence intervals, and I think more could be done.

But that’s fine. It’s best that the authors did what they did, which was to present their results. Now that the data are out there, other researchers can go back in and do more sophisticated analysis. That’s how research should go. It would not make sense for such important results to be held under wraps, waiting for some ideal statistical analysis that might never happens.

Overall, this is an inspiring story of what can be learned from a natural experiment.

The news article also has this sad conclusion:

At the end of 2017, Congress passed legislation eliminating the health law’s fines for not carrying health insurance, a change that probably guarantees that the I.R.S. letters will remain a one-time experiment.

But now that they have evidence that the letters had a positive effect, maybe they’ll restart the program, no?

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