author: niplav, created: 2020-12-30, modified: 2023-07-23, language: english, status: in progress, importance: 3, confidence: opinion
Reading a text is sometimes a big time investment, but people don't approach their reading in a structured manner, e.g. by keeping notes, making flashcards, doing exercises, writing reviews or making summaries. I have been taking notes on the books I read since mid 2017, but have neglected writing reviews or summaries that might be useful to others. This is my attempt at salvaging that oversight.
“Compassion, by the Pound” by the economists F. Bailey Norwood and Jayson L. Lusk is one of those books that are excellent in their first half, and somewhat (but not utterly) disappointing in their second half. The two economists, spurred by their own research and their perceived lack of good information on the topic of farm animal welfare, start off with a historical overview of animal agriculture and animal welfare activism, proceed to talking about the sentience of animals, give an enormous overview of standard animal agriculture praxis, push in two completely unnecessary chapters about how philosophers and economists see animal agriculture (the first one being massively oversimplifying, the second one being annoyed), present a model for consumers to use for deciding what to eat (although the links have fallen victim to linkrot), copy-paste one of their papers into the book, and finish with those kinds of general closing statements that are as so often too vacuous to be interesting.
At its best, the book is just a delicious heap of information about animal agriculture praxis. From detailed lists of surgeries performed on animals without anaesthetics (dehorning, beak trimming, castration, teeth clipping and tail docking) to the behaviour of cows in big pens (they huddle together in a corner, and don't use up all the space) to the hierarchical behavior in chickens (much more strongly than in cows, actually a major factor of injury in cage-free egg production), the book presents an industrial-scale mountain of interesting facts about animal agriculture. The best parts of the book pretty much scream to be flashcardized.
Norwood's & Lusk's judgement seems well informed and not particularly strongly clouded by bias, and presented in a empathetic, but also neutral tone (except in the case of them mentioning in a side remark that surgeries such as castration on animals are nearly always performed without anaesthetics, seemingly regarding this as completely acceptable).
However, not everything is golden under the sun. The chapter on philosophy is especially painful (or might this just be my Gell-Mann amnesia speaking?) – they seem dismissive of philosophers' arguments, present them in short and watered-down form, and even state in a footnote:
If there is one thing we have learned from reading the works of ethical philosophers; it is that no one ever, ever wins the debate
— F. Bailey Norwood/Jayson L. Lusk, “Compassion, by the Pound” p. 388, 2011
The chapter on Talking with Economists is better, but plagued with the eternal problem of economics: people don't like it, and the same debate about the very basic tenets of economics needs to be rehashed over and over again. As it happens here, much to my own disappointment (“Yes, sure, I agree that things have a price, that regulation is often nonsensical and consumers change their minds when presented with the same scenario, worded slightly differently. Can we get back to fascinating in-depth descriptions of animal agriculture now, please?”).
Chapter 9, Consumer Expressions, is not bad per se, but still sloppy: It is abundantly clear that the chapter is simply a paper by the two authors copy-pasted into the book. The experiments they perform are interesting and scientifically sophisticated, but the chapter is nonetheless jarring to the reader – clearly somewhat out of place in the rest of the book.
Two things stand out to me from this book:
“Compassion, by the Pound” is sometimes clearly a product of annoyance – an annoyance at animal advocates who allegedly spread misinformation about farming practices, annoyance at people who just don't understand economics (which I get, yes, it's frustrating), and yes, sometimes also annoyance at the horrifying conditions many farm animals have to live under. Hopefully both economists and animal advocates won't have to be annoyed as much in the future, but for the time being, we're still killing and eating animals.
“The Human Predicament” is a book about life philosophy, written by the pessimistic analytic philosopher David Benatar. In it, Benatar describes what he calls the human predicament (hence the title), which consists of the fact that human lives are usually bad, and much worse than people themselves think. In his view, human lives lack cosmic (and sometimes terrestrial) meaning, are bad because they're much shorter than they could be, much more filled with pain and discomfort than humans think, and full of ignorance, unfulfilled desires and physical deterioration during the course of one's lifetime.
However, according to Benatar, all alternatives are also bad: death, because it often deprives of life, and annihilates the person dying; and suicide, for much the same reasons, unless it annihilates a life that is awful enough to justify death. Life extension, under Benatar's view, is extremely unlikely, and even if achieved, would only prolong the misery of human existence.
The only positive option is to not come into existence at all–or at least not make others come into existence, even though one desires to. He alludes several times to one of his other books, Better Never To Have Been, in which he advocates for antinatalism.
Reading this book felt a little bit pointless to me. Since beliefs are for actions, and Benatar is just applying a linear transformation to all available options (if everything's bad, nothing is), you act exactly the same. Although I had a phase where I believed antinatalism quite strongly, and still don't plan on having kids (although I know that this attitude might change with increasing age), but overall antinatalism does not strike me as a pragmatic policy, me instead adopting an anti-pure-replicator strategy.
Especially the chapter on meaning felt irrelevant: I don't have an internal experience of meaning (or the lack thereof), and oscillate between believing it to be a subtype of high-valence qualia and believing it to be a mechanism for the mind to do things that are in themselves not enjoyable (a "second reward signal" next to pleasure).
Benatar mentions cryonics, life extension technology and transhumanism in general, and while his treatment of these topics is more respectful than most, he dismisses them fairly quickly. I disagree with his underlying skepticism on these the feasibility of radically altering the human condition through technology, given that it seems that humanity can expect to find itself in a period of hyperbolic economic growth (see also Roodman 2020).
I am also not a fan of the pessimism-optimism distinction. Benatar himself touches on this:
that a view is pessimistic should, in itself, neither count in its favor nor against it. (The same, of course, is true of an optimistic view.)
— David Benatar, “The Human Predicament” p. 225, 2017
It seems to me that humans can believe very bad things to be the case and still be happier than most other humans in their lives (I know this is at least true for one human, myself). This, combined with the fact that Benatar simply shifts the utility function downwards, makes me inclined to rejecting much of his worldview as simply a matter of emotional tone on the same facts everyone else also believes.
Finally, I want to accuse Benatar of insufficient pessimism (on his own criteria): The most likely outcome for humanity (and for life in general) seems not to be total extinction, but instead a universe filled with beings most capable of copying themselves, the whole cosmos teeming with subsistence-level beings with very boring conscious experiences until the stars go out. (Or even worse scenarios from anti-aligned artificial intelligences, see Tomasik 2019).
Overall, the book had some interesting points about suicide, the quality of life and meaning, but felt rather pointless.
Illustrates the theory-practice gap, but in the other direction: excellently practical first half (which helped me get into the first jhāna (briefly) during a long retreat (the hard part is getting the access concentration good enough, which the book doesn't spend enough time on, in my opinion—only a short appendix (at least there's recommendations for other books)). The anecdotes from his students and their problems with entering the jhānas are fascinating (pīti that doesn't go away? jhānas contraindicated with seizures?), as are his reports of deep concentration states on long retreats (the visual field turning white in the fourth jhāna, and reports about the the nimitta, make me wonder what goes on in the visual cortex during absorption meditation).
But Brasington just wants to believe that the Suttas are basically infallible, especially when they report what the Buddha said (Brasington has remarked on podcasts that we know that the Buddha knew what he was talking about, which I don't get—even if he was a great meditator and thinker, he could just have been wrong sometimes): Expecting the Suttas to accurately and coherently reflect reality in all its aspects is a bit too optimistic for me. But Brasington goes full memetic immune disorder on the Suttas, and the result is just…uninteresting?
Preceded by a superior book with the same topic; this one is sleeker, less filled with random interesting facts, less scientific, less exuberant in its prose. I enjoyed the introduction of the SPC framework (though that might be also a flaw, unlike with ITN I haven't even seen anyone else pay lip service to SPC…), found the alleged first popular introduction to population axiology cute, and liked the chapters on stagnation.
But honestly? I enjoyed the research that led to those chapters more than the chapters in the book themselves (especially Rodriguez 2019 and Rodriguez 2020), and I think the team that made The Precipice would've done a nicer job at exposition.
Similarly, I was not a huge fan of the chapter on risks from artificial intelligence. Too conservative, which might've been warranted before GPT-3, but mid-2022? Bad timing to be all "could be soon or bad, or both, or not, idk". (Although apparently other reviewers have the opposite issue, so perhaps a good compromise was struck).
I am unsure about the value lock-in frame. On the one hand, it's a very rough description of some of the danger with AI x-risk, but not all danger fits in that format: What if AI systems don't lock in any specific value, but kill off humanity and then go on to explore the space of all possible value? This framing also invites endless bickering about "who gets to control the AIs values" and "democracy" and "social solutions", and the completely separate issue of stable totalitarianism.
Finally: Who the hell decided this was a good way to do endnotes? In general the best policy is to under no circumstances use endnotes, ever, why. But WWOTF makes it 10x worse: I usually read endnotes, because I'm unusually curious and bad at priorization, but WWOTF only has ~25% substantive endnotes, with the rest being just incomplete references (which can be accessed on the website)—so I found myself flipping back and forth, only to be disappointed most of the time. Surely there must be a better way of distinguishing between citations and endnotes.
Maybe I should've avoided it: Pop philosophy that is already in my own groundwater.
If you're reading this site, read The Precipice instead. (Not a full condemnation of WWOTF).
Curiosity is the drug of the internet.
—Gloria Mark, “Attention Span” p. 114, 2023
Read this while researching attention spans, I did not find what I was looking for (remaining mostly unconvinced that the reported statistics are strong enough to justify the claim that attention spans have been declining). Otherwise acceptable; and in some parts genuinely novel to me, giving a plethora of ways of measuring attention (transcranial Doppler sonography, functional near-infrared spectroscopy, facial thermography to measure cognitive effort, blood velocity…), claims that the Pomodoro technique hasn't been experimentally tested (big if true!). Apparently people often self-interrupt while on a task, which I've noticed myself doing more & more. The Big 5 relate to how humans perform tasks:
Those who score high in Neuroticism in personality tests also tend to perform worse on selective attention tasks where they have to pay attention to some things and ignore distracting stimuli,²⁰ much like the Stroop task.
—Gloria Mark, “Attention Span” p. 154, 2023
We expected that conscientious people would be more likely to be continuous email checkers, and that is exactly what we found. In fact, it explained their email checking behavior to a striking extent […] we found that people who score higher on the personality trait of Openness perform better in environments with interruptions.
—Gloria Mark, “Attention Span” p. 156, 2023
This leads to conscientious people being more exhausted if possible low-effort interruptions are taken away from them, they just work continuously until exhaustion.
Mark's background in art gives some entertaining anecdotes and statistics, I especially enjoyed learning about dialectical montage and decreasing shot-lengths in movies, series and advertisements.
Apparently people want to use this as a self-help book‽ Bizarre.
It sounds odd to say that happiness should be an engineering discipline, but that seems to be the inevitable conclusion.
—Stuart Russell, “Human Compatible” p. 123, 2019
Another book with an orange cover, and another popularization of a thing I spend a lot of time thinking and reading about. But I like this one much more!
Thoroughly enjoyed the many tidbits from AI history, and the stories about semi-successful systems, as well as a preference-utilitarian definition of "sadism, envy, resentment and malice", a naive approach to meta-reasoning ("just reason about a thing if the expected value of reasoning is positive", without talking about the obvious boots-trapping problems…but still), learning about the Baldwin effect and the quotes about risks from artificial intelligence from Butler's Erewhon.
Skeptical about transformative AI soon, and about the scaling hypothesis, but probably for reasons I can't understand. Also this was written before GPT-3, so he might've changed his mind since then.
The book does assume that reward is the optimization target, and doesn't mention inner optimizers, but your popularization of alignment can only do so much. I should really read into the whole CIRL/corrigibility debate, any day now.
The book did have endnotes, which I hate, but less so than with What We Owe The Future—perhaps because I got to read the titles of the papers and not just a naked "Foo et al. 2010", perhaps because there was just more content per footnote.
The task is, fortunately, not the following: given a machine that possesses a high degree of intelligence, work out how to contol it. If that were the task, we would be toast. A machine viewed as a black box, a fait accompli, might as well have arrived from outer space. And our chances of controlling a superintelligent entity from outer space are roughly zero. Similar arguments apply to the methods of creating AI systems that guarantee we won't understand how they work; these methods include whole-brain emulation¹—creating souped-up electronic copies of human brains—as well as methods based on simulated evolutions of programs.² I won't say more about these proposals because they are so obviously a bad idea.
—Stuart Russell, “Human Compatible” p. 171, 2019
These were written for the 2019 LessWrong Review.
I read this post only half a year ago after seeing it being referenced in several different places, mostly as a newer, better alternative to the existing FOOM-type failure scenarios. I also didn't follow the comments on this post when it came out.
This post makes a lot of sense in Christiano's worldview, where we have a relatively continuous, somewhat multipolar takeoff which to a large extent inherits the problem in our current world. This is especially applies to part I: we already have many different instances of scenarios where humans follow measured incentives and produce unintended outcomes. Goodhart's law is a thing. Part I ties in especially well with Wei Dai's concern that
AI-powered memetic warfare makes all humans effectively insane.
While I haven't done research on this, I have a medium strength intuition that this is already happening. Many people I know are at least somewhat addicted to the internet, having lost a lot of attention due to having their motivational system hijacked, which is worrying because Attention is your scarcest resource. I believe investigating the amount to which attention has deteriorated (or has been monopolized by different actors) would be valuable, as well as thinking about which incentives will start when AI technologies become more powerful (Daniel Kokotajlo has been writing especially interesting essays on this kind of problem).
As for part II, I'm a bit more skeptical. I would summarize "going out with a bang" as a "collective treacherous turn", which would demand somewhat high levels of coordination between agents of various different levels of intelligence (agents would be incentivized to turn early because of first-mover-advantages, but this would increase the probability of humans doing something about it), as well as agents knowing very early that they want to perform a treacherous turn to influence-seeking behavior. I'd like to think about how the frequency of premature treacherous turns relates to the intelligence of agents. Would that be continuous or discontinuous? Unrelated to Christiano's post, this seems like an important consideration (maybe work has gone into this and I just haven't seen it yet).
Still, part II holds up pretty well, especially since we can expect AI systems to cooperate effectively via merging utility functions, and we can see systems in the real world that fail regularly, but not much is being done about them (especially social structures that sort-of work).
I have referenced this post numerous times, mostly in connection with a short explanation of how I think current attention-grabbing systems are a variant of what is described in part I. I think it's pretty good, and someone (not me) should flesh the idea out a bit more, perhaps connecting it to existing systems (I remember the story about the recommender system manipulating its users into political extremism to increase viewing time, but I can't find a link right now).
The one thing I would like to see improved is at least some links to prior existing work. Christiano writes that
(None of the concerns in this post are novel.)
but it isn't clear whether he is just summarizing things he has thought about, which are implicit knowledge in his social web, or whether he is summarizing existing texts. I think part I would have benefitted from a link to Goodhart's law (or an explanation why it is something different).
I believe this is an important gears-level addition to posts like hyperbolic growth, long-term growth as a sequence of exponential modes and an old Yudkowsky post I am unable to find at the moment.
I don't know how closely these texts are connected, but Modeling the Human Trajectory picks up one year later, creating two technical models: one stochastically fitting and extrapolating GDP growth; the other providing a deterministic outlook, considering labor, capital, human capital, technology and production (and, in one case, natural resources). Roodman arrives at somewhat similar conclusions, too: The industrial revolution was a very big deal, and something happened around 1960 that has slowed the previous strong growth (as far as I remember, it doesn't provide an explicit reason for this).
A point in this post that I found especially interesting was the speculation about the back plague being the spark that ignited the industrial revolution. The reason given is a good example of slack catapulting a system out of a local maximum, in this case a malthusian europe into the industrial revolution.
Interestingly, both this text and Roodman don't consider individual intelligence as an important factor in global productivity. Despite the well-known Flynn-Effect that has mostly continued since 1960 (caveat caveat), no extraordinary change in global productivity has occurred. This makes some sense: a rise of less than 1 standard deviation might be appreciable, but not groundbreaking. But the relation to artificial intelligence makes it interesting: the purported (economic) advantage of AI systems is that they can copy themselves, thereby making population growth not the most constraining variable in this growth model. I don't believe this is particularly anticipation-constraining, though: this could mean that either the post-singularity ("singularity") world is multipolar, or the singleton controlling everything has created many sub-agents.
I appreciate this post. I have referenced it a couple of times in conversations. Together with the investigation by OpenPhil it makes a solid case that the gods of straight lines have decided to throw us into the most important century of history. May the godess of everything else be merciful with us.
These were written for the 2020 LessWrong Review.
I read this post at the same time as reading Ascani 2019 and Ricón 2021 in an attempt to get clear about anti-aging research. Comparing these three texts against each other, I would classify Ascani 2019 as trying to figure out whether focusing on anti-aging research is a good idea, Ricón 2021 trying to give a gearsy overview of the field (objective unlocked: get Nintil posts cross-posted to LessWrong), and this text as showing what has already been accomplished.
In that regard it succeeds perfectly well: The structure of Part V is so clean I suspect that it sweeps a bunch of complexity and alternative approaches under the rug, and the results described seriously impressed me and some of the people I was reading this text with at the time (We can reverse arthritis and cataracts in mice‽ We can double their maximum lifespan‽). It is excellent science propaganda: Inspiring awe at what has been accomplished, desire to accomplish more, and hope that this is possible.
While the post shines in parts III, IV and V, I have some quibbles and complaints about the introduction, part I and part II. First, I disliked the jab against cryonics in the first paragraph without considering the costs and benefits, which rightly received some pushback in the comments (the strongest counter-observation being that barring some practical suggestions for slowing down aging right now, cryonics and anti-aging research occupy very different parts of the strategy for life-extension, and can be pursued in parallel). Part II disappointed me because it was pro-longevity advocacy under the veneer of a factual question: Has anybody actually tried to think through how a world without aging might actually look like, instead of re-treading the same pro-aging trance and anti-aging science arguments? That seems like a question that is both interesting and pretty relevant, even when you believe that ending aging is important enough that it should definitely be done, if just to prepare for weird second- and third-order effects.
(Part I felt like I was a choir being preached to, which isn't that bad, but still…)
I really liked learning a bunch of new facts about aging (as for example the list of species that don't age, that aging is responsible for 30% of lost DALYs, and distinction between gerontology, engineering and geriatrics). Factposts are underrated.
The comments on this post were often very good, and had some nice discussion about whether the advice in section VII was to be focused on.
I've been overly negative in this review, but overall I still like this post, and have voted a 4 on it (which I might change to a 1). The parts III-V are excellent, and I have only minor problems with the preceding parts. This is the kind of post I would give a science-interested skeptic of anti-aging research. I'd like to have this post in the review, because it represents something some part of the core to the LessWrong transhumanist aesthetic that often gets overlooked.
The problem outlined in this post results from two major concerns on lesswrong: risks from advanced AI systems and irrationality due to parasitic memes.
It presents the problem of persuasion tools as continuous with the problems humanity has had with virulent ideologies and sticky memes, exacerbated by the increasing capability of narrowly intelligent machine learning systems to exploit biases in human thought. It provides (but doesn't explore) two examples from history to support its hypothesis: the printing press as a partial cause of the 30 years war, and the radio as a partial cause of 20th century totalitarianism.
Especially those two concerns reminded me of Is Clickbait Destroying Our General Intelligence? (Eliezer Yudkowsky, 2018), which could be situated in this series of events:
I suspect some culturally transmitted parts of the general intelligence software got damaged by radio, television, and the Internet, with a key causal step being an increased hypercompetition of ideas compared to earlier years.
Kokotajlo also briefly considers the hypothesis that epistemic conditions might have become better through the internet, but rejects it (for reasons that are not spelled out, but the answers to Have epistemic conditions always been this bad? (Wei Dai, 2021) might be illuminating). (Survivorship bias probably plays a large role here: epistemically unsound information is less likely to survive long-term trials for truth, especially in an environment where memes on the less truth-oriented side of the spectrum are in a harsher competition than memes on the more truth-oriented side).
This post was written a year ago, and didn't make any concrete predictions (for a vignette of the future by the author, see What 2026 looks like (Daniel's Median Future) (Daniel Kokotajlo, 2021)). My personal implied predictions under this worldview are something like this:
I found the text quite relevant both to thinking about possible alternative stories about the way in which AI could go wrong, and also to my personal life.
In the domain of AI safety, I became more convinced of the importance of aligning recommender systems to human values (also mentioned in the post), if they pose larger risk than commonly assumed, and provide a good ground for experimentation on alignment techniques. Whether aligning recommender systems is more important than aligning large language models seems like an important crux here: Are the short-term/long-term risks higher for recommender systems (i.e. reinforcement learners) larger than the risks from large language models? Which route appears more fruitful when trying to align more generally capable systems? As far as I can see, the alignment community is more interested in attempts to align large language models, compared to recommender systems, probably due to recent progress in that area and because it's easier to test alignment in language models (?).
The scenarios in which AI powered memetic warfare significantly harm humanity can also be tied into research on the malicious use of AI, e.g. The Malicious Use of Artificial Intelligence: Forecasting, Prevention, and Mitigation (Brundage et al. 2018). Policy tools from diplomacy with regard to biological, chemical and nuclear warfare could be applied to memetic and psychologcial warfare.
The text explicitely positions the dangers of persuasion tools as a risk factor, but more speculatively, they might also pose an existential risk in themselves, in two different scenarios:
On the personal side, after being fooled by people using GPT-3 to generate tweets and seeing at least one instance of observing someone asking a commenter for the MD5 hashsum of a string to verify that the commenter was human (and the commenter failing that challenge), along with observing the increasingly negative effects of internet usage on my attention span, I decided to separate my place for sleeping & eating from the place where I use internet, with a ~10 minute commute between those two. I also decided to pay less attention to news stories/reddit/twitter, especially from sources affiliated with large governments, downloaded my favourite websites.
This post was relevant to my thoughts about alternative AI risk scenarios as well as drastic personal decisions, and I expect to give it a 1 or (more likely) a 4 in the final vote.