Astral Codex Ten Podcast

Machine Alignment Monday 6/19/23 The face of Mt. Everest is gradual and continuous; for each point on the mountain, the points 1 mm away aren’t too much higher or lower. But you still wouldn’t want to ski down it. I thought about this when reading What A Compute-Centric Framework Says About Takeoff Speeds, by Tom Davidson. Davidson tries to model what some people (including me) have previously called “slow AI takeoff”. He thinks this is a misnomer. Like skiing down the side of Mount Everest, progress in AI capabilities can be simultaneously gradual, continuous, fast, and terrifying. Specifically, he predicts it will take about three years to go from AIs that can do 20% of all human jobs (weighted by economic value) to AIs that can do 100%, with significantly superhuman AIs within a year after that. As penance for my previous mistake, I’ll try to describe Davidson’s forecast in more depth. https://astralcodexten.substack.com/p/davidson-on-takeoff-speeds

Finalist #5 in the Book Review Contest   [This is one of the finalists in the 2023 book review contest, written by an ACX reader who will remain anonymous until after voting is done. I’ll be posting about one of these a week for several months. When you’ve read them all, I’ll ask you to vote for a favorite, so remember which ones you liked] I. I found Njal’s Saga hard to follow. Halfway through, a friend reassured me it wasn’t my fault. The medieval Icelanders had erred in releasing it as a book. It should have been the world’s wackiest Phoenix Wright: Ace Attorney spinoff. Remember, medieval Iceland was an early attempt at anarcho-capitalist utopia. When Harald Fairhair declared himself King of Norway, the Norwegians who refused to bend the knee fled west to build a makeshift seastead on a frozen volcanic island. No lords, no kings, no masters. Only lawsuits. So, so many lawsuits. Once a year, the Icelanders would meet at the Althing, a free-for-all open-air law court. There they would engage in that most Viking of pastimes - suing each other, ad nauseam, for every minor slight of the past six months. Offended parties would sell their rights to prosecute a case to the highest bidder, who would go around seeking fair arbitrators (or, in larger cases, defer to a panel chosen by chieftain-nobles called godi. Courts would propose a penalty for the losing side - usually money. There were no police, but if the losers refused to pay, the courts could declare them “outlaws” - in which case it was legal to kill them. If you wanted to be a Viking in medieval Iceland, you needed a good lawyer. And Njal was the greatest lawyer of all. https://astralcodexten.substack.com/p/your-book-review-njals-saga  

Unfortunately I hate many of you. Only the ones with Twitter accounts. If you don’t have one of those, you’re fine. But if you do have one, there’s a good chance you said something which horribly offended me. You said everyone who believed X was an idiot and a Nazi, and I believed X. You read the title but not the body of an article about some group I care about, and viciously insulted them based on your misunderstanding of their position. You spent five seconds thinking of a clever dunk on someone who happened to be a friend of mine trying really hard to make the world better, and ruined their day. https://astralcodexten.substack.com/p/your-incentives-are-not-the-same  

You know all the stuff we’ve been talking about here the past few years - mental mountains, trapped priors, relaxed beliefs under psychedelics? The new keyword for all of that is “canalization”. At least that’s what I gather from a giant paper recently published by some of the leading thinkers in computational psychiatry (Karl Friston, Robin Carhart-Harris, etc). A quick review: you can model the brain as an energy landscape . . . . . . with various peaks and valleys in some multidimensional space Situations and stimuli plant “you” at some point on the landscape, and then you “roll down” towards some local minimum. If you’re the sort of person who repeats “I hate myself, I hate myself” in a lot of different situations, then you can think of the action of saying “I hate myself” as an attractor - a particularly steep, deep valley which it’s easy to fall into and hard to get out of. Many situations are close to the slopes of the “I hate myself” valley, so it’s easy to roll down and get caught there. What are examples of valleys other than saying “I hate myself”? The authors suggest habits. If you always make the sign of the cross when passing a graveyard, there’s a steep slope from the situation of passing a graveyard to the action of signing the cross. We can be even broader: something really basic like edge-detection in the visual system is a valley. When you see a scene, you almost always want to automatically do edge-detection on it. Walking normally is a valley; there’s a certain correct sequence of muscle movements, and you don’t want to start rotating your ligaments in some weird direction halfway through. https://astralcodexten.substack.com/p/the-canal-papers tt

Finalist #4 in the Book Review Contest [This is one of the finalists in the 2023 book review contest, written by an ACX reader who will remain anonymous until after voting is done. I’ll be posting about one of these a week for several months. When you’ve read them all, I’ll ask you to vote for a favorite, so remember which ones you liked] https://astralcodexten.substack.com/p/your-book-review-mans-search-for  

Sometimes people do a study and find that a particular correlation is r = 0.2, or a particular effect size is d = 1.1. Then an article tries to “put this in context”. “The study found r = 0.2, which for context is about the same as the degree to which the number of spots on a dog affects its friskiness.” But there are many statistics that are much higher than you would intuitively think, and many other statistics that are much lower than you would intuitively think. A dishonest person can use one of these for “context”, and then you will incorrectly think the effect is very high or very low. https://astralcodexten.substack.com/p/attempts-to-put-statistics-in-context  

Original post: Why Is The Academic Job Market So Weird? Table Of Contents Comments With More Information On Academic Hiring 2. Comments About How Things Got This Way 3. Comparisons To The Programmer Job Market 4. Comparisons To Other Job Markets 5. Proposed Solutions 6. Comments With Practical Advice For New PhDs https://astralcodexten.substack.com/p/highlights-from-the-comments-on-the-bc8  

Finalist #3 in the Book Review Contest [This is one of the finalists in the 2023 book review contest, written by an ACX reader who will remain anonymous until after voting is done. I’ll be posting about one of these a week for several months. When you’ve read them all, I’ll ask you to vote for a favorite, so remember which ones you liked] I'll begin with a contentious but invariably true statement, which I've no interest in defending here: new books—at least new nonfiction books—are not meant to be read. In truth, a new book is a Schelling point for the transmission of ideas. So while the nominal purpose of a book review like this is to answer the question Should I read this book?, its real purpose is to answer Should I pick up these ideas? I set out to find the best book-length argument—one that really engages with the technical issues—against imminent, world-dooming, Skynet-and-Matrix-manifesting artificial intelligence. I arrived at Why Machines Will Never Rule the World by Jobst Landgrebe and Barry Smith, published by Routledge just last year. Landgrebe, an AI and biomedicine entrepreneur, and Smith, an eminent philosopher, are connected by their study of Edmund Husserl, and the influence of Husserl and phenomenology is clear throughout the book. (“Influence of Husserl” is usually a good enough reason to stop reading something.) Should you read Why Machines Will Never Rule the World? If you're an AI safety researcher or have a technical interest in the topic, then you might enjoy it. It's sweeping and impeccably researched, but it's also academic and at times demanding, and for long stretches the meat-to-shell ratio is poor. But should you pick up these ideas? My aim here isn’t to summarize the book, or marinate you in its technical details. ATU 325 is heady stuff. Rather, I simply want to give you a taste of the key arguments, enough to decide the question for yourself. https://astralcodexten.substack.com/p/your-book-review-why-machines-will  

https://astralcodexten.substack.com/p/links-for-may-2023 [Remember, I haven’t independently verified each link. On average, commenters will end up spotting evidence that around two or three of the links in each links post are wrong or misleading. I correct these as I see them, and will highlight important corrections later, but I can’t guarantee I will have caught them all by the time you read this.]

In this post, the author suggests that the standard metrics for assessing the efficacy of medications, especially antidepressants, may be flawed and restrictive, indicating that if these stringent standards were applied to other common medications, they too would be deemed 'clinically insignificant', despite widespread acceptance of their effectiveness​. https://astralcodexten.substack.com/p/all-medications-are-insignificant