The Nonlinear Library

The Nonlinear Library

By The Nonlinear FundEducation
Download on the App Store

All The Nonlinear Library episodes

  • EA - How many people die from the flu? (OWID) by Lizka
    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 die from the flu? (OWID), published by Lizka on October 24, 2022 on The Effective Altruism Forum.
    Saloni Dattani and Fiona Spooner at Our World in Data (OWID) have recently published a new article on the annual death toll of the flu — I thought the article was interesting and am link-posting it.
    The article also explains some issues with measuring this, how this has changed over time, factors that increase the chances of dying from the flu, and why some flu seasons are worse than others. The headline (emphasis mine):
    The risk of death from influenza has declined over time, but globally, hundreds of thousands of people still die from the disease each year.
    The annual mortality caused by seasonal influenza was estimated by the Global Pandemic Mortality Project II using data between 2002 and 2011. They estimated that, during this period, seasonal influenza caused between 294,000 and 518,000 deaths each year globally.
    For reference on what 300,000 to 500,000 looks like, here's a chart (from a different article) that shows deaths by cause:
    An excerpt and a couple of charts — or just see the article itself.
    Globally, hundreds of thousands of people die from seasonal influenza every year. During large flu pandemics, when influenza strains evolved substantially, the death toll was even higher.
    But the risk of dying from influenza has declined substantially over time from improvements in sanitation, healthcare, and vaccination.
    People born in 1940 had around a third of the risk of dying from influenza as those born in 1900 – even when they reached the same age. This decline continued, and those born in 1980 have a risk of half that of those born in 1940.
    Influenza still remains a large burden around the world, because of an aging population and a lack of access to healthcare and sanitation in many countries.
    In this article, we look into these developments in detail: how many people die from seasonal influenza and how this has changed over time.
    We will also look at which factors increase the risk of dying from the flu and understand why, in some years, influenza has led to large pandemics that caused millions of deaths. This knowledge can inform us about the risks of influenza in the future.
    Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.
    3 min
  • EA - Ways in which EA could fail by michel
    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: Ways in which EA could fail, published by michel on October 24, 2022 on The Effective Altruism Forum.
    This is the first post in a series of posts attempting to make EA strategy discussions more productive. I try to lay out a framework and shared vocabulary for navigating important risks and opportunities faced by EA decision-makers. The following posts will be linked here.
    Summary
    Following an influx of funding, media attention, and influence, the EA movement is speeding along an exciting, yet perilous, trajectory.
    A lot of the EA community’s future impact rests on this uncertain growth going well (and thereby avoiding movement collapse scenarios).
    Yet, discussions or critiques of EA’s trajectory rarely feel action-guiding. Even when critiques propose course corrections that are tempting to agree with (e.g., EA should be bigger!), proposed course corrections to make EA more like X often don’t rigorously engage with the downsides of being more like X, or the opportunity cost of not being like Y. Proposals to make EA more like X also often leave me with only a vague understanding of what X looks like and how we get from here to X.
    In hopes of making discussions of the EA community’s trajectory more productive (and to clarify my own thinking on the matter), I will lay out a series of posts that provide an overview of:
    (1) Important ways in which the EA movement could fail(2) “Domains” in which EA could make course corrections (e.g., more cause-area tailored outreach, new professional networks and events, etc.)(3) Key considerations that inform course corrections (i.e., places to disagree about course corrections)(4) Next steps to help guide the EA movement through exciting and perilous times
    This is the first post in this series: Ways in which EA could fail. Consider it an attempt at bounding the later discussions of strategy updates.
    Ways in which EA could fail
    The EA movement could collapse. Many movements before us have, and we’re not that special. But other movements, like abolitionism, have left lasting social change. In hopes of being the type of movement that does so well it doesn’t need to exist anymore, this section outlines many of the ways EA could fail.
    In this post, I’ll define “failure” as the EA/EA-adjacent ecosystem achieving substantially less impact than it could have along some other trajectory. Note that this is a pretty broad definition. Depending on your worldview and person-affecting ethical views, failure could look more like millions or billions of people alive in the near-future suffering in ways that could have been prevented – or failure could look more like an existential catastrophe that permanently stops sentient life from achieving its full potential.
    Implicit in this definition of failure is a statement about the influence EA already has, or could grow to have: I think our ideas and resources are powerful enough to seriously influence how many beings suffer today and how many beings live beautiful lives in the future, such that there’s a massive difference between the most good a flourishing EA ecosystem could achieve (i.e., upper bound) and the good, or possibly even harm, a collapsed EA ecosystem leaves us with (i.e., lower bound).
    In order to land closer to the heights of positive impact, let’s think concretely about the worst cases that we must collectively avoid. In what ways might the EA movement fail?
    I identify four clusters of failures:
    Reputation failures: failures that result in EA losing a substantial amount of possible impact because the EA community develops a negative reputation.
    Resource failures: failures that result in EA losing a substantial amount of possible impact because the community becomes preventably constrained by financial, human, or infrastructure resources.
    Rigor failures: failures that result in EA losing ...
    18 min
  • EA - UGAP Starter Program Retrospective by jessica mccurdy
    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: UGAP Starter Program Retrospective, published by jessica mccurdy on October 24, 2022 on The Effective Altruism Forum.
    TLDR
    The CEA uni groups team piloted running a starter program before the University Group Accelerator Program (UGAP) to:
    Provide resources to more groups.
    Gain more context on groups to see whether UGAP was right for them.
    Improve our mentor-group matching.
    We had 81 groups participate in the starter program from over 20 countries.
    We faced challenges due to rapid scaling, capacity constraints, and trying to accommodate different target audiences. We are pleased with the results and are glad we scaled quickly but have noted many improvements to be made going forward.
    We found that having the starter program separate from UGAP was confusing for organizers and that we would like both programs to have the same criteria for admissions. We also were able to better provide specific start-of-the-year resources to newer groups.
    Therefore, next semester we will integrate the starter program into UGAP as one single program. This program will likely be open only to new groups. We are looking into different programming for existing groups such as a continued mentorship network (which will be explained further in a future post)
    This post is part of our efforts to communicate more transparently what we are doing. This is the first post in a series of posts and will serve as mostly an update about the starter program specifically. Future posts will dive further into our strategy, Metrics & Evaluation (M&E), and our future plans.
    What was the UGAP Starter Program?
    We had a virtual two-week starter program that was open to a larger group of schools than UGAP itself. It provided groups with virtual training and resources to prepare for the semester. This was our first time running this program and 81 universities participated, of which 56 were accepted to UGAP.
    Groups were given a starter program guide that pointed to specific resources, attended a kickoff session focused on strategy and getting ready for the semester, and had a meeting with one of our starter program mentors where they discussed a plan for the coming semester. Participants who were preparing to facilitate fellowships were able to participate in facilitator training through EA Virtual Programs. We also offered two optional workshops, one on outreach and another on reasoning.
    During the program, individuals from groups were invited to apply for UGAP where they could receive continued mentorship and a stipend. Ultimately, 56 groups ended up having at least one organizer participate in UGAP (which is going on now).
    Why the starter program?
    Overall, the starter program itself aimed to provide the following:
    Help advise and steer group strategy for new groups
    We think that group strategy is one of the highest leverage ways to make a big difference in group outcomes. We go over the core of our current advice in this post.
    Increase motivation and excitement about running a group
    We focused on why we believe community building at universities can be so impactful, tried to create the sense of being part of a much larger international project, and expressed encouragement and gratitude to organizers.
    Make sure groups were ready for the start of the semester
    We think the start of the semester is one of, if not the most important time of the year for university groups. We wanted to make sure that groups were going into the semester with a plan and the resources they needed.
    Running the starter program separate from UGAP was an experiment to do the following:
    Provide resources to a larger number of groups, unconstrained by UGAP mentor capacity.
    Last summer we realized that there was a lot of interest in UGAP. While we only had ~30 groups participate in the Spring, our interest form had ~50 additional groups interes...
    15 min
  • LW - The optimal timing of spending on AGI safety work; why we should probably be spending more now by Tristan Cook
    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 optimal timing of spending on AGI safety work; why we should probably be spending more now, published by Tristan Cook on October 24, 2022 on LessWrong.
    Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.
    1 min
  • LW - AI researchers announce NeuroAI agenda by Cameron Berg
    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: AI researchers announce NeuroAI agenda, published by Cameron Berg on October 24, 2022 on LessWrong.
    Last week, 27 highly prominent AI researchers and neuroscientists released a preprint entitled Toward Next-Generation Artificial Intelligence: Catalyzing the NeuroAI Revolution. I think this report is definitely worth reading, especially for people interested in understanding and predicting the long-term trajectory of AI research.
    Below, I’ll briefly highlight four passages from the paper that seemed particularly relevant to me.
    Doubts about the 'prosaic' approach yielding AGI
    The authors write:
    The seeds of the current AI revolution were planted decades ago, largely by researchers attempting to understand how brains compute (McCulloch and Pitts 1943). Indeed, the earliest efforts to build an “artificial brain” led to the invention of the modern “von Neumann computer architecture,” for which John von Neumann explicitly drew upon the very limited knowledge of the brain available to him in the 1940s (Von Neumann 2012). The deep convolutional networks that catalyzed the recent revolution in modern AI are built upon artificial neural networks (ANNs) directly inspired by the Nobel-prize winning work of David Hubel and Torsten Wiesel on visual processing circuits in the cat (Hubel and Wiesel 1962; LeCun and Bengio 1995). Similarly, the development of reinforcement learning (RL) drew a direct line of inspiration from insights into animal behavior and neural activity during learning (Thorndike and Bruce 2017; Rescorla 1972; Schultz, Dayan, and Montague 1997).
    Now, decades later, applications of ANNs and RL are coming so quickly that many observers assume that the long-elusive goal of human-level intelligence—sometimes referred to as “artificial general intelligence”—is within our grasp. However, in contrast to the optimism of those outside the field, many front-line AI researchers believe that major new breakthroughs are needed before we can build artificial systems capable of doing all that a human, or even a much simpler animal like a mouse, can do [emphasis added].
    To the degree we take this final comment seriously—that many within the field think that major breakthroughs (plural) are needed before we get AGI—we should probably update in the direction of being relatively more skeptical of prosaic AGI safety. If it appears increasingly likely that AGI won't look like a clean extrapolation from current systems, it would therefore make increasingly less sense to bake the prosaic assumption into AGI safety research.
    Researchers eyeing brain-based approaches
    While many key AI advances, such as convolutional ANNs and RL were inspired by neuroscience, much of the current research in machine learning is following its own path by building on previously-developed approaches that were inspired by decades old findings in neuroscience, such as attention-based neural networks which were loosely inspired by attention mechanisms in the brain (Itti, Koch, and Niebur 1998; Larochelle and Hinton 2010; Xu et al. 2015). New influences from modern neuroscience exist, but they are spearheaded by a minority of researchers. This represents a missed opportunity. Over the last decades, through efforts such as the NIH BRAIN initiative and others, we have amassed an enormous amount of knowledge about the brain. This has allowed us to learn a great deal about the anatomical and functional structures that underpin natural intelligence.
    The emerging field of NeuroAI, at the intersection of neuroscience and AI, is based on the premise that a better understanding of neural computation will reveal basic ingredients of intelligence and catalyze the next revolution in AI, eventually leading to artificial agents with capabilities that match and perhaps even surpass those of humans [emphasis in the original]. We...
    12 min
  • LW - Empowerment is All We Need by jacob cannell
    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: Empowerment is All We Need, published by jacob cannell on October 23, 2022 on LessWrong.
    Intro
    What/who would you like to become in a thousand subjective years? or a million?
    Perhaps, like me, you wish to become posthuman: to transcend mortality and biology, to become a substrate independent mind, to wear new bodies like clothes, to grow more intelligent, wise, wealthy, and connected, to explore the multiverse, perhaps eventually to split, merge, and change - to vasten.
    Regardless of who you are now or what specific values you endorse today, I suspect you too would at least desire these possibilities as options. Absent some culture specific social stigmas, who would not like more wealth, health, and power? more future optionality?
    As biological creatures, our fundamental evolutionary imperative is to be fruitful and multiply, so our core innate high level value should be inclusive genetic fitness. But for intelligent long lived animals like ourselves, reproduction is a terminal goal in the impossibly distant future: on the order of around 1e11 neural clock cycles from birth[1], to be more precise. Explicit optimization of inclusive genetic fitness through simulation and planning over such vast time horizons is simply implausible - especially for a mere 20 watt irreversible computer such as the human brain, no matter how efficient.
    Fortunately there exists an accessible common goal which is ultimately instrumentally convergent for nearly all final goals: power-seeking, or simply: empowerment.
    Omohundro proposed an early version of the instrumental convergence hypothesis as applied to AI in his 2008 paper the Basic AI Drives, however the same principle was already recognized by Klyubin et al in their 2005 paper "Empowerment: A Universal Agent-Centric Measure of Control"[2]:
    Our central hypothesis is that there exist a local and universal utility function which may help individuals survive and hence speed up evolution by making the fitness landscape smoother. The function is local in the sense that it doesn’t rely on infinitely long history of past experience, does not require global knowledge about the world, and that it provides localized feedback to the individual.
    To a sugar-feeding bacterium, high sugar concentration means longer survival time and hence more possibilities of moving to different places for reproduction, to a chimpanzee higher social status means more mating choice and interaction, to a person more money means more opportunities and more options. The common feature of the above examples is the striving for situations with more options, with more potential for control or influence. To capture this notion quantitatively, as a proper utility function, we need to quantify how much control or influence an animal or human (an agent from now on) has.
    Salge et al later summarized these arguments into the Behavioral Empowerment Hypothesis[3]:
    The adaptation brought about by natural evolution produced organisms that in absence of specific goals behave as if they were maximizing their empowerment.
    Empowerment provides a succinct unifying explanation for much of the apparent complexity of human values: our drives for power, knowledge, self-actualization, social status/influence, curiosity and even fun[4] can all be derived as instrumental subgoals or manifestations of empowerment. Of course empowerment alone can not be the only value or organisms would never mate: sexual attraction is the principle deviation later in life (after sexual maturity), along with the related cooperative empathy/love/altruism mechanisms to align individuals with family and allies (forming loose hierarchical agents which empowerment also serves).
    The key central lesson that modern neuroscience gifted machine learning is that the vast apparent complexity of the adult human brain, with al...
    30 min
  • AF - A Barebones Guide to Mechanistic Interpretability Prerequisites by Neel Nanda
    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 Barebones Guide to Mechanistic Interpretability Prerequisites, published by Neel Nanda on October 24, 2022 on The AI Alignment Forum.
    Co-authored by Neel Nanda and Jess Smith
    Why does this exist?
    People often get intimidated when trying to get into AI or AI Alignment research. People often think that the gulf between where they are and where they need to be is huge. This presents practical concerns for people trying to change fields: we all have limited time and energy. And for the most part, people wildly overestimate the actual core skills required.
    This guide is our take on the essential skills required to understand, write code and ideally contribute useful research to mechanistic interpretability. We hope that it’s useful and unintimidating. :)
    Core Skills:
    Maths:
    Linear Algebra: 3Blue1Brown or Linear Algebra Done Right
    Core goals - to deeply & intuitively understand these concepts:
    Basis
    Change of basis
    That a vector space is a geometric object that doesn’t necessarily have a canonical basis
    That a matrix is a linear map between two vector spaces (or from a vector space to itself)
    Bonus things that it’s useful to understand:
    What’s singular value decomposition? Why is it useful?
    What are orthogonal/orthonormal matrices, and how is changing to an orthonormal basis importantly different from just any change of basis?
    What are eigenvalues and eigenvectors, and what do these tell you about a linear map?
    Probability basics
    Basics of distributions: expected value, standard deviation, normal distributions
    Log likelihood
    Maximum value estimators
    Random variables
    Central limit theorem
    Calculus basics
    Gradients
    The chain rule
    The intuition for what backprop is - in particular, grokking the idea that backprop is just the chain rule on multivariate functions
    Coding:
    Python Basics
    The “how to learn coding” market is pretty saturated - there’s a lot of good stuff out there! And not really a clear best one.
    Zac Hatfield-Dodds recommends Al Sweigart's Automate the Boring Stuff and then Beyond the Basic Stuff (both readable for free on inventwithpython.com, or purchasable in books); he's also written some books of exercises. If you prefer a more traditional textbook, Think Python 2e is excellent and also available freely online.
    NumPy Basics
    Try to do the first ~third of these:. Bonus points for doing them in pytorch on tensors :)
    ML:
    Rough grounding in ML.
    fast.ai is a good intro, but a fair bit more effort than is necessary. For an 80/20, focus on Andrej Karpathy’s new video explaining neural nets:
    PyTorch basics
    Don’t go overboard here. You’ll pick up what you need over time - learning to google things when you get confused or stuck is most of the real skill in programming.
    One goal: build linear regression that runs in Google Colab on a GPU.
    Transformers - probably the biggest way mechanistic interpretability differs from normal ML is that it’s really important to deeply understand the architectures of the models you use, all of the moving parts inside of them, and how they fit together. In this case, the main architecture that matters is a transformer! (This is useful in normal ML too, but you can often get away with treating the model as a black box)
    Check out the illustrated transformer
    Note that you can pretty much ignore the stuff on encoder vs decoder transformers - we mostly care about autoregressive decoder-only transformers like GPT-2, which means that each token can only see tokens before it, and they learn to predict the next token
    Good (but hard) exercise: Code your own tiny GPT-2 and train it. If you can do this, I’d say that you basically fully understand the transformer architecture.
    Example of basic training boilerplate and train script
    The EasyTransformer codebase is probably good to riff off of here
    An alternate framing that may help give different intui...
    6 min
  • AF - POWERplay: An open-source toolchain to study AI power-seeking by Edouard Harris
    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: POWERplay: An open-source toolchain to study AI power-seeking, published by Edouard Harris on October 24, 2022 on The AI Alignment Forum.
    We're open-sourcing POWERplay, a research toolchain you can use to study power-seeking behavior in reinforcement learning agents. POWERplay was developed by Gladstone AI for internal research.
    POWERplay's main use is to estimate the instrumental value that a reinforcement learning agent can get from a state in an MDP. Its implementation is based on a definition of instrumental value (or "POWER") first proposed by Alex Turner et al. We've extended this definition to cover certain tractable multi-agent RL settings, and built an implementation behind a simple Python API.
    We've used POWERplay previously to obtain some suggestive early results in single-agent and multi-agent power-seeking. But we think there may be more low-hanging fruit to be found in this area.
    Beyond our own ideas about what to do next, we've also received some interesting conceptual questions in connection with this work. A major reason we're open-sourcing POWERplay is to lower the cost of converting these conceptual questions into real experiments with concrete outcomes, that can support or falsify our intuitions about instrumental convergence.
    Ramp-up
    We've designed POWERplay to make it as easy as possible for you to get started with it. Follow the installation and quickstart instructions to get moving quickly. Use the replication API to trivially reproduce any figure from any post in our instrumental convergence sequence. Design single-agent and multi-agent MDPs and policies, launch experiments on your local machine, and visualize results with clear figures and animations.
    POWERplay comes with "batteries included", meaning all the code samples in the documentation should just work out-of-the-box if it's been installed successfully. It also comes with pre-run examples of experimental results, so you can understand what "normal" output is supposed to look like. While this does make the repo weigh in at about 500 MB, it's worth the benefits of letting you immediately start playing around with visualizations on preexisting data.
    If we've done our job right, a smart and curious grad student (with a bit of Python experience) should be able to start reproducing our previous experiments within an hour, and to have some new — and hopefully interesting! — results within a week.
    We're looking forward to seeing what people do with this. If you have any questions or comments about POWERplay, feel free to reach out to Edouard at [email protected].
    Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org.
    3 min
  • EA - Lord Martin Rees: an appreciation by HaydnBelfield
    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: Lord Martin Rees: an appreciation, published by HaydnBelfield on October 24, 2022 on The Effective Altruism Forum.
    Other titles I considered:
    Lord Martin Rees’ new existential risk book published: "If Science is to Save Us"
    Lord Martin Rees should be on more existential risk introductions
    Lord Martin Rees is a boss
    [Disclaimer: I work at CSER, which Martin co-founded (in some sense he’s my boss), and I know and like him, so I’m biased towards him. However, I’ve mostly referred to objective, verifiable evidence, and others can corroborate.]
    Summary:
    Lord Martin Rees is one of the UK’s, and perhaps the world’s, most eminent scientists, has been an advocate of existential risk since at least 2003 (and arguably since the 1970s), and is a charismatic speaker and engaging writer. I’ll argue he should be featured in more ‘introductions to existential risk’ and should be turned to as a powerful advocate for existential risk reduction - like e.g. Will MacAskill and Toby Ord. In this short piece I’ll give a quick bio and describe some of his work. In part, this post is just an appreciation post – he’s just recently had his 80th birthday, after all.
    Photo by Hanna-Katrina Jędrosz for the New Statesman
    New book
    The occasion for this post is that his new book has just been published: If Science is to Save Us. Summary:
    There has never been a time when ‘following the science’ has been more important for humanity. At no other point in history have we had such advanced knowledge and technology at our fingertips, nor had such astonishing capacity to determine the future of our planet.
    But the decisions we must make on how science is applied belong outside the lab and should be the outcome of wide public debate. For that to happen, science needs to become part of our common culture. Science is not just for scientists: if it were, it could never save us from the multiple crises we face. For science can save us, if its innovations mesh carefully into society and its applications are channelled for the common good.
    As Martin Rees argues in this expert and personal analysis of the scientific endeavour on which we all depend, we need to think globally, we need to think rationally and we need to think long-term, empowered by twenty-first-century technology but guided by values that science alone cannot provide.
    Coverage:
    The Telegraph.
    New Statesman: Martin Rees: “This could be our last century on Earth”
    The Economist: How science can save the world
    Quick bio
    Martin Rees is a cosmologist and astrophysicist who’s done leading research (500+ papers) on black holes, quasars and the multiverse. He knew Stephen Hawking well (and wrote a very nice obituary for him).
    He’s been the Astronomer Royal (previous holders, Halley of ‘Halley’s Comet’ fame) since 1995. He was the 60th President of the Royal Society, 2005-2010 (previous holders Wren, Pepys, Newton, Rutherford, etc). The Royal Society is the UK’s national academy of sciences, and one of the most preeminent in the world. He was Master of Trinity College, Cambridge 2004-2012. He was made a Lord, specifically a non-party-political (‘crossbench’) member of the House of Lords, in 2005. He’s published 10+ books including:
    From Here to Infinity: Scientific Horizons (UK) / From Here to Infinity: A Vision for the Future of Science
    Just Six Numbers
    Our Cosmic Habitat
    Before the Beginning: Our Universe and Others
    Cosmic Coincidences: Dark matter, mankind and anthropic cosmology
    Gravity’s Fatal Attraction: Black Holes in the Universe
    New Perspectives in Astrophysical Cosmology
    This is all to say he’s one of the UK’s (and perhaps the world) leading scientists.
    Good introductory materials
    Rees has given two TED talks, which have together been watched 4-5 million times. I think they’re great introductions to the subject.
    Can we prevent the end of the world?...
    9 min
  • EA - Minimalist extended very repugnant conclusions are the least repugnant by Teo Ajantaival
    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: Minimalist extended very repugnant conclusions are the least repugnant, published by Teo Ajantaival on October 24, 2022 on The Effective Altruism Forum.
    This is part four of a series on minimalist axiologies (i.e. axiologies that essentially say “the less this, the better”).
    Every part of this series builds on the previous parts, but can also be read independently.
    Summary
    Population axiology matters greatly for our priorities. Recently, it has been claimed that all plausible axiological views imply certain “very repugnant conclusions” (defined below). In this response, I argue that minimalist views avoid these “very repugnant conclusions”, and that they face less repugnant conclusions than do contrasting offsetting views.
    1. Are repugnant implications inevitable?
    In population axiology, certain offsetting views, according to which independent bads can be offset by a sufficient amount of independent goods, face the Very Repugnant Conclusion (VRC):
    A population of arbitrarily many lives with arbitrarily high welfare is worse than a population of arbitrarily many arbitrarily negative lives plus sufficiently many ε-lives that each have an arbitrarily small quantity of positive welfare (Figure 1).
    Offsetting views also allow the ε-lives in the VRC to be rollercoaster lives that all contain unbearable suffering (purportedly counterbalanced by a sufficient amount of bliss).
    In particular, symmetric classical utilitarianism implies interchangeability between a non-suffering ε-life and the rollercoaster life illustrated in Figure 2 (provided that the “overall welfare” of the rollercoaster life equals ε).
    Additionally, one may replace each non-suffering ε-life in the original VRC with an intrapersonal VRC life (Figure 3).
    Recently, Budolfson and Spears (2018) have argued that all plausible views in population ethics imply similarly repugnant conclusions, namely that they imply either the VRC or a closely analogous Extended VRC (XVRC), which I illustrate shortly at the beginning of Section 2.
    The purpose of this essay is to argue that this claim does not apply to minimalist views. In a nutshell: minimalist views avoid the VRC, can avoid repugnant XVRCs, and, at any rate, face XVRCs that are less repugnant than are the comparable conclusions faced by offsetting views.
    Three claims
    Budolfson and Spears (2018, pp. 31–32) make the following three claims:
    Claim 1: No leading welfarist axiology can avoid the VRC.
    Claim 2: No other welfarist axiology in the literature can avoid the XVRC.
    Claim 3: The XVRC is just as repugnant as the VRC.
    The authors conclude that:
    Repugnant implications are an inevitable feature of any plausible axiology. If repugnance cannot be avoided, then it should not be. We believe this should be among the guiding insights for the next generation of work in value theory.
    Claim 1 does not apply to minimalist axiologies
    The scope of Claim 1 (“No leading welfarist axiology can avoid the VRC”) is limited to ‘leading’ welfarist axiologies, that is to views that, according to the authors, are commonly-held in the axiological literature (p. 8).
    These do not cover minimalist axiologies, although axiologies that are essentially minimalist have been defended, for instance, by Schopenhauer (1818/1819, 1851), Wolf (1996, 1997, 2004), Fehige (1998), Breyer (2015), and Knutsson (2021b, “axiological claim”).
    To the extent that the VRC seems repugnant, it is worth noting that all minimalist axiologies do avoid the VRC, and can do so neatly without relying on arbitrary or ad hoc assumptions.
    Claim 2 requires that we extend the XVRC
    Claim 2 (“No other welfarist axiology in the literature can avoid the XVRC”) is not straightforward to evaluate, because the original XVRC, as the authors define it, applies strictly only to views that make the assumption of independently aggregable ...
    43 min