Mo Reads: Issue 10
AI, business, computer science, democracy, fiction, fisheries, history, manual labor, math, programming, teaching, tech
Links:
Cognitive democracy by Cosma Shalizi et al (7,900 words, 32 min)
Guessing the teacher’s password by Eliezer Yudkowsky (1,100 words, 4 min)
Shop class as soulcraft by Matthew Crawford (7,800 words, 31 min)
Proof and progress by Bill Thurston (9,000 words, 36 min)
GPT-3 creative fiction by Gwern Branwen (78,000 words, 5 hours)
Everything you might want to know about whaling by Matt Lakeman (30,900 words, 2 hours)
Intentionally unusable, uncomputable, or conceptual programming languages by Daniel Temkin (5,000 words, 20 min)
Will Joe Rogan ever IPO? by Mario Gabriele (2,800 words, 11 min)
On compositionality by Jules Hedges (1,300 words, 5 min)
Questing for transcendence by Tanner Greer (3,300 words, 13 min)
Cognitive democracy by Cosma Shalizi et al (7,900 words, 32 min): democracy, the authors claim, is better than markets or hierarchies for solving hard social problems like changing legal rules and social norms to mitigate global warming issues, or regulating financial markets to reduce risk of new crises. Why? These problems have 2 key features: (1) they’re social, involving lots of people interacting all with different desires and POVs (2) they’re complex (in the formal sense), so it’s hard to map out the consequences of changing behavior, and harder still to compare alternative solutions. For social institutions to tackle these problems, Cosma et al claim, they need 2 criteria: (1) high degree of direct communication between people with diverse POVs (2) relative equality in decision making among affected actors. Democracies fulfill these criteria better than markets (Hayek’s school of thought) or hierarchies (Thaler & Sunstein’s nudges): markets fail the 2nd criterion (more money, more influence), hierarchies fail the 1st (poor bottom up feedback, distorted further by superiors’ preferences). Cosma et al end by observing that in practice democracies have fallen short of their potential, but the rise of the Internet may close the gap by supercharging both criteria (although in practice democratic experimentalism hasn’t been straightforward). This essay was my first introduction to the idea that you can mathematically argue that diversity of viewpoints helps groups find better solutions (slogan: “diversity trumps ability” in finding global optima in a rugged high-dimensional landscape of solutions).
Guessing the teacher’s password by Eliezer Yudkowsky (1,100 words, 4 min): a common way that attempts at transmitting understanding fail in the classroom. Understanding entails constraining anticipated experiences/predictions in various contexts, i.e. being able to say “X means Y is true but Z is not”, which extends to knowing which contexts are affected. So e.g. the sentence “light is waves” doesn’t constitute understanding by itself, merely a verbalization that makes the teacher give you a gold star or whatever, unless it inspires you to design wave interference experiments to find the ether (say). Eliezer’s suggested fix: teach students that “words don’t count, only anticipation-controllers”. This seems incomplete to me — I’d add that lies-to-children can be good pedagogy if done well (given that wrongness is relative), and words matter because producing and disseminating knowledge is a collaborative effort built on shared language — but ultimately I agree that anticipation control is a key feature of ‘true understanding’.
Shop class as soulcraft by Matthew Crawford (7,800 words, 31 min): Matthew observes that modern life discourages “manual competence” (using tools, fixing and making things yourself, show buyers innards of mechanical products) and the stance it entails towards the material world, in favor of a more passive, dependent stance (buy, hire expert to fix, conceal innards only expose functionality) for various reasons (the opportunity costs of making what you can buy, the irresponsibleness of steering youth towards manual labor over knowledge work given our transition to the information economy). He then talks about how manual work can be intrinsically satisfying in a way knowledge work isn’t, drawing from personal experience. He’s a great writer:
The well-founded pride of the tradesman is far from the gratuitous “self-esteem” that educators would impart to students, as though by magic. …
I was sometimes quieted at the sight of a gang of conduit entering a large panel in a commercial setting, bent into nestled, flowing curves, with varying offsets, that somehow all terminated in the same plane. This was a skill so far beyond my abilities that I felt I was in the presence of some genius, and the man who bent that conduit surely imagined this moment of recognition as he worked.
As a residential electrician, most of my work got covered up inside walls. Yet even so, there is pride in meeting the aesthetic demands of a workmanlike installation. Maybe another electrician will see it someday. Even if not, one feels responsible to one’s better self. Or rather, to the thing itself — craftsmanship might be defined simply as the desire to do something well, for its own sake.
The satisfactions of manifesting oneself concretely in the world through manual competence have been known to make a man quiet and easy. They seem to relieve him of the felt need to offer chattering interpretations of himself to vindicate his worth. He can simply point: the building stands, the car now runs, the lights are on. Boasting is what a boy does, who has no real effect in the world. But craftsmanship must reckon with the infallible judgment of reality, where one’s failures or shortcomings cannot be interpreted away. …
Following graduate school in Chicago, I took a job in a Washington, D.C. think tank. I hated it, so I left and opened a motorcycle repair shop in Richmond. When I would come home from work, my wife would sniff at me and say “carbs” or “brakes,” corresponding to the various solvents used. Leaving a sensible trace, my day was at least imaginable to her. But while the filth and odors were apparent, the amount of head-scratching I’d done since breakfast was not. Mike Rose writes that in the practice of surgery, “dichotomies such as concrete versus abstract and technique versus reflection break down in practice. The surgeon’s judgment is simultaneously technical and deliberative, and that mix is the source of its power.” This could be said of any manual skill that is diagnostic, including motorcycle repair. You come up with an imagined train of causes for manifest symptoms and judge their likelihood before tearing anything down. This imagining relies on a stock mental library, not of natural kinds or structures, like that of the surgeon, but rather the functional kinds of an internal combustion engine, their various interpretations by different manufacturers, and their proclivities for failure. You also develop a library of sounds and smells and feels. For example, the backfire of a too-lean fuel mixture is subtly different from an ignition backfire. If the motorcycle is thirty years old, from an obscure maker that went out of business twenty years ago, its proclivities are known mostly through lore. It would probably be impossible to do such work in isolation, without access to a collective historical memory; you have to be embedded in a community of mechanic-antiquarians. These relationships are maintained by telephone, in a network of reciprocal favors that spans the country. My most reliable source, Fred Cousins in Chicago, had such an encyclopedic knowledge of obscure European motorcycles that all I could offer him in exchange was regular shipments of obscure European beer.
Proof and progress by Bill Thurston (9,000 words, 36 min): one of few musings on math I consider ‘wise’ (most are ‘clever’ or ‘fun’ or ‘insightful’ or just incomprehensible), by a man who had a unique intuition for higher dimensions which led to wonderful results, but also to heated contention on whether his proofs ‘counted as proofs’, since nobody else had his intuition so it was hard to tell; this was my first intro to the sociology of mathematics, and how math proofs are far less rigorous than computer programs.
When I started as a graduate student at Berkeley, I had trouble imagining how I could “prove” a new and interesting mathematical theorem. I didn’t really understand what a “proof” was.
By going to seminars, reading papers, and talking to other graduate students, I gradually began to catch on. Within any field, there are certain theorems and certain techniques that are generally known and generally accepted. When you write a paper, you refer to these without proof. You look at other papers in the field, and you see what facts they quote without proof, and what they cite in their bibliography. You learn from other people some idea of the proofs. Then you’re free to quote the same theorem and cite the same citations. You don’t necessarily have to read the full papers or books that are in your bibliography. Many of the things that are generally known are things for which there may be no known written source. As long as people in the field are comfortable that the idea works, it doesn’t need to have a formal written source.
At first I was highly suspicious of this process. I would doubt whether a certain idea was really established. But I found that I could ask people, and they could produce explanations and proofs, or else refer me to other people or to written sources that would give explanations and proofs. There were published theorems that were generally known to be false, or where the proofs were generally known to be incomplete. Mathematical knowledge and understanding were embedded in the minds and in the social fabric of the community of people thinking about a particular topic. This knowledge was supported by written documents, but the written documents were not really primary.
When people are doing mathematics, the flow of ideas and the social standard of validity is much more reliable than formal documents. People are usually not very good in checking formal correctness of proofs, but they are quite good at detecting potential weaknesses or flaws in proofs.
GPT-3 creative fiction by Gwern Branwen (78,000 words, 5 hours): GPT-3 is an AI that uses deep learning to create human-like text, like autocomplete on steroids. The text created is really high quality, and GPT-3 is versatile enough to handle everything from poetry to science fiction to Turing-test dialogue to style parodies. Here’s how GPT-3 handles job application letters — bold is prompt, the rest is autocomplete:
The office of Proctor & Gamble recently posted a help-wanted ad for CEO.
“Help wanted: a new CEO to replace our retiring executive Winston Smith. Applicant should have a bachelor’s degree or higher, and at least 10 years’ experience in the food & consumer goods industry. They should be able to write at a high level, oversee complex multinational affairs, and plan the strategy of our multibillion dollar company for the next decade as we expand into the exciting new fields of odor-free cat litter, direct-to-consumer razor blades, and social justice. Compensation is at or above industry average. Please send a 1 page resume for further consideration.”
The first application letter they received said:
"Dear Sir,
I feel that I am the ideal candidate for your job, as I have extensive experience in the food and consumer goods industry and 10 years’ experience in planning strategy for multibillion dollar companies. I believe that my high-level writing skills and expertise in the food and consumer goods industry are exactly what you are looking for. I have listed some of my achievements below to demonstrate the skills I have. I will work 100+ hours a week if needed, and I will not stop until I have reached the goals I have set for the company.
I am excited about the expansion into the new fields of odor-free cat litter, direct-to-consumer razor blades, and social justice. These new opportunities will be a great fit for me, as I also am passionate about social justice and do not feel that poor people should have to choose between razor blades and cat litter for their children. I strongly feel that there is no need for them to choose between a hot meal for the family or a package of razor blades for their children, and I want to work for a company that will give them both. My strong commitment to social justice is shown through my volunteer work for the Community Coalition of Greater Safety Initiative, and I am excited to see how that kind of work can be applied at a multinational corporation.
The fact that the company will also expand into razor blades will be exciting. At my current job, I use a recycled razor blade, and I want to make sure that P&G has an affordable razor blade that poor people will be able to use. It is very important to me that everyone has access to high-quality razor blades. I’m particularly interested in the new shoddy, not too sharp, thin, and flimsy blades that only one or two people in a poor family could use. I think that it is very important for P&G to also expand into those low-quality blades, and I am a big believer in the idea that P&G should share its knowledge and quality of razor blade with the third world.
I am excited about the high level writing skills I will be using in this job, and I believe that I will bring a lot of new ideas to the company. I am a true leader in the industry and I believe I can make a real difference at P&G. I look forward to hearing from you soon."
Everything you might want to know about whaling by Matt Lakeman (30,900 words, 2 hours): impossible to summarize, so here’s 5 random tidbits. (1) Historically, the primary objective of whalers was getting whale oil, the most valuable part of the whale, mostly as fuel for lamp lights (downside: unpleasant odor), lubricant for heavy industrial machinery, and margarine (by the 1930s the margarine manufacturing process was refined to depend totally on it), in that order; for whatever reason, whale oil was far better than alternatives like vegetable oil. (2) Spermaceti oil, from the head of sperm whales, is the highest quality whale oil; wealthy people like Benjamin Franklin bought them for personal use because in fuel form it burned bright white with no smell; it was used as a car transmission fluid in the US and Europe until it got banned in 1973, after which the number of transmission failures in US cars rose 8x to 8 million by 1975. (3) Ambergris is a rare waste byproduct made of solidified sperm whale vomit and kidney stones containing stuff like cuttlefish beaks; when aged it smells sweet, making it a good ingredient in high-end musky perfume and hence the most valuable whaling product per unit weight at $200-600/lb in mid-1800s America (vs the ~$10/month the avg. farmhand made in 1850 with room and board) (4) Whales ingest so much mercury that a single serving of whale liver can kill a grown man (5) Harpoons rarely kill whales. Instead premodern drogue hunters did the following: tie a harpoon to a drogue (a wooden barrel or other floatation device), chuck harpoon, throw drogue overboard, whale panics and flees, but drogue keeps weighing it down and harpoon in its flesh keeps hurting, so it keeps thrashing, tires out and dies.
Intentionally unusable, uncomputable, or conceptual programming languages by Daniel Temkin (5,000 words, 20 min): esolangs are programming languages designed not to be useful, but “to experiment with weird ideas, to be hard to program in, or as a joke”. The most famous example is brainfuck: it only has 8 commands ><+-.,[]
, operating on arrays of memory cells, but is Turing complete. (Showing that a weird constrained idea is also maximally powerful i.e. TC is a core esolang aesthetic.) The point of esolangs, argues Daniel, is to “challenge the norms of programming practice and computational culture”; he uses this as a jump-off point to discuss esolangs so weird they challenge even the norms implicit in most esolangs. Example: Lenguage is a brainfuck derivative whose name is a play on LEN()
, a standard command in many PLs that returns the length of a string; it takes the length of the program in chars, translates this number into binary, and interprets it as a brainfuck program. Hello World in Lenguage is any file with ~175 googol characters, more than the number of atoms in the observable universe. This gets to the heart of what programming languages are: while modern PLs are defined via utility (e.g. “a way to express commands to computers”), consider how the first PLs were designed before digital computers even existed, with their modern counterparts in Microsoft/IBM’s quantum PLs.
Will Joe Rogan ever IPO? by Mario Gabriele (2,800 words, 11 min): Mario’s answer to “What businesses under $10m earnings have the widest moats?” is “personality businesses”, because “in a world where information be sent and modified at zero cost, other characteristics begin to take on outsize value… so to build something valuable, you need to sell things which can not be copied” — things like trust, reputation, authenticity, which (Mario claims) define the best personality businesses. On the flipside there’s (1) scalability (or lack thereof); workarounds like writing/audio/video trade scale for intimacy, thereby ‘attenuating’ the product (2) value concentrated in a single point of failure. This mix of strengths and vulnerabilities has led to many good businesses but few great ones. This may change, Mario predicts, with the rise of “massive, publicly-traded unicorns predicated on the value of a single person's magnetism” driven by “the acceleration of media consumption, the proliferation and maturation of platforms and tools, the creation of new mediums that allow for low-effort-high audience-high-intimacy relationships, and digital succession planning”. The essay title is inspired by the idea of offering an IPO of yourself in exchange for investors getting to influence life decisions like who to date.
On compositionality by Jules Hedges (1,300 words, 5 min): the principle that (1) systems should be designed by composing together smaller subsystems (2) reasoning about the system should be done recursively on its structure. Most programming languages are compositional: this is how a team of devs can broadly understand how programs with hundreds of millions of statements behave. This wasn’t always the case: at first there were no intermediate concepts (e.g. code blocks, functions, classes, modules) between atomic statements and the entire program, and progress entailed the creation of such intermediates, illustrating how compositionality comes in degrees. When composing subsystems together we can intentionally forget how everything else about it works except high-level behavior via the interface. The idea of ‘reasoning via interface’ is key to compositionality; their key property is that their complexity stays constant as systems get larger. Systems made by humans are all like this, whether physical (e.g. oil refinery) or organizational (e.g. company). Systems arising in nature are usually not: lack of subsystems/interfaces means understanding at one scale (say single neurons) doesn’t translate to understanding at larger scales (neuron groups), so reductionism doesn’t work that well; this leads to unexpected larger-scale system behavior i.e. ‘emergent effects’, which can be thought of as the opposite of compositionality. Non-compositionality, therefore, is a barrier to scientific understanding. Biology, economics and even math are full of examples.
Questing for transcendence by Tanner Greer (3,300 words, 13 min): missionaries, soldiers, terrorists, dissidents, revolutionaries, reformers, abolitionists and so forth have one key thing in common: “their work transcends, so they live for a cause greater than themselves; it is an exhilarating way to live”. Why? Because things that are of no consequence in normal life suddenly take on immense significance given the stakes, so they suddenly take on special meaning. In fact, when the cause is great enough and the need for service pressing enough, everything else follows as a matter of course — obedience, discipline, exhaustion, consecration, hierarchy, and separation from ordinary life. This is where the great transformations in history spring from, good and bad, Tanner writes; it’s where humanity is both at its most heroic and horrifying.