I write this list each year for two reasons. First, I want to note the books that went beyond leaving a mark on me, to haunt me. These are the works I can’t get out of my mind. I often think of the people, events, and ideas in them, and they linger like ghosts around me. Second, I need to hold myself accountable, for if I don’t read books, I can fall into the dark depths of the internet with news websites, blogs, social media, or worse… “Welcome to the internet.”
My list this year was heavy on biographies and memoirs, along with machine learning papers (it was a burning hot year for advances in AI and ML); it’s where the fire of my curiosity burned.
BEST BOOK OF 2022
Dyson, Invention: A Life. Making things is hard, especially if you come from a place that looks down upon it, like the UK does for manufacturing. Dyson is a dogged inventor and entrepreneur whose company makes the best vacuum cleaners, hair dryers, air purifiers, and more. This memoir details his life story and process to build products and a company, focusing on first-principles thinking to identify human needs and engineer products in unconventional, innovative ways. I know how hard it is to manufacture machines, to build in the world of atoms, so I give Dyson a salute for sharing this story.
Schmidt and Rosenberg, How Google Works. This is a 2014 book that details the chaos of the older Google, when it was a powerhouse of creative engineers with the triumvirate of Schmidt, Brin, and Page leading it. The authors lay out their unique views on strategy, culture, talent, decision-making, communication, and innovation. Unfortunately, Google in 2022 seems to be a very different company. Other noteworthy tech histories I read this year included “No Filter (Instagram)” and “The Attention Factory (TikTok).”
Schlender and Tetzeli, Becoming Steve Jobs. Jobs was a great CEO who shone as a product genius and team builder with Cook, Jony, Johnson, Fadell, Schiller, Forstall, and more. They made the world’s most valuable company (Apple) and devices (iPhones and MacBooks). While the Isaacson biography gives a broader view of Jobs as a person, Schlender knew Jobs for much longer and gives the best story of his evolution from a bratty, man-child to a seasoned product creator, CEO, and Mensch. Other notable Apple biographies were Kahney’s “Tim Cook” and “Jony Ive”, both of who don’t get enough attention but are stars in their own right.
Slootman, Amp It Up. Slootman may be the best tech CEO after Nadella, Cook, and Zuck. This Dutchman has an unconventional background and style, but is the king of business software and has some great insights on running a top-notch company, with a focus on execution. Strong runner-ups for the CEO bios this year were Iger’s “The Ride of a Lifetime” (so good, he came back) and Dell’s “Play Nice But Win” (the rare founder who matures into a solid CEO).
Higgins, Power Play: Tesla, Elon Musk, and the Bet of the Century. Tesla, as an electric car company, is so technically difficult and ambitious that it shouldn’t exist. Musk took a whacky idea and made one of the most iconic and loved products of the last decade (despite his recent efforts at Twitter tarnishing Tesla and Musk). It’s a complex story of the car guys fighting with the software guys to build products that will one day roam on the moon and Mars.
Jackson and Delahanty, Eleven Rings: The Soul of Success. Phil Jackson is the former coach of the Chicago Bulls and the LA Lakers. He was a minister’s son who got pulled away into Eastern mysticism and major league basketball, who as a coach won the most NBA championships and had the 5th most wins when he retired. He also had difficult and prickly star players who fought often (Jordan and Pippen/Rodman, Kobe and Shaq), and he had to fuse them into high-performing teams.
Kolchin, American slavery, 1619-1877. This was the most painful book I read this year. The two greatest crimes of the American republic were the genocide against Native Americans and the systems of oppression via slavery, whose aftereffects still touch all American life. While I love the United States as a group of people, a culture, a constitutional republic, and a metaphysical entity with a set of evolving values, everyone needs to know the evil facts of its past, which this book presents well, unlike the historically inaccurate NYTimes 1619 project.
Roll, George Marshall: Defender of the Republic. Marshall was the Chief of Staff and General of the Army that oversaw the Allied coalition from DC in WW2, as the boss of General Eisenhower. He was a person of the highest ability and character, a leader who I place among Washington, Lincoln, and FDR as the best that the US can produce. His final great act as Secretary of State was the Marshall Plan to rebuild the European continent that he helped vanquish.
Ai Wei Wei, 1000 Years of Joys and Sorrows: A Memoir. Ai tells the story of his incredible life, from growing up in a camp during the Cultural Revolution to making art in Beijing and New York, and fighting the Chinese authorities. A powerful passage: “Despite the ever-present threat of government intervention, I was now more than ready to resume my role as a provocateur. My war with power was a bit like an online game: each time I died, I came to life again. Power might use all kinds of tactics to attack me or monitor me, but I could turn those tactics into an advantage, through public activity and through creative ripostes, maintaining the role that they least wanted me to play—that of a mass-oriented activist and artist. Freedom of expression became a central meaning of my art, for personal freedom is the highest value that we can know.”
Whipple, The Gatekeepers: How the White House Chiefs of Staff Define Every Presidency. Most people don’t realize the second most powerful person in the US is generally the White House Chief of Staff, who I consider to be a co-consul running the republic with the President like in the old Roman days. Yes, the Speaker of the House, Senate Majority Leader, and the Chief Justice have official roles and their own rings of power, but the Chiefs act like COOs for the country, and are worth paying attention to. Another notable memoir for a civil servant was “Duty” by Bob Gates.
Perlroth, This Is How They Tell Me the World Ends: The Cyberweapons Arms Race. Our lives are all digital, and unfortunately, that world is not secure and is completely fragile in ways that few understand (unless you’ve taken courses in Operating Systems, Networking, Systems Security, etc., and studied hacking). Perlroth’s story of that underworld, and the cyber-arms race between world governments, reads like a dystopian novel. But it’s real, and it shows why we need projects like Bitcoin, Ethereum, Urbit, IPFS, and Uniswap as the testing grounds for something better.
Goldberg, The Path of Modern Yoga: The History of an Embodied Spiritual Practice. This is is the best history of modern global yoga, being generally balanced and thoughtful. It’s estimated that 300mm people in the world practice it, which traces its 20th-century heritage to a small group of people. The book pairs well with Singleton’s “Gurus of Yoga” and Mohan’s book “Krishnamacharya: His Life and Teachings”, about the pivotal teacher of Iyengar, Desikachar, Indra Devi, and Jois. For older yoga history, see Mallinson’s “Roots of Yoga” and Bryant’s “Yoga Sutras of Patañjali.”
Odell, Anna: The Biography. Wintour is arguably one of the most powerful people in global media, and certainly in fashion. She excelled due to her strong work ethic, impeccable taste, and skill at leading creative teams. She’s intensely loyal to her crew, picking and growing young talent with a fashion fund that backs new designers, along with the Met’s Costume Institute. Her weakness is that she’s made some mistakes over time: backing fit and leather; not having enough diversity in her models and staff, and; most importantly not taking the digital side seriously enough for a long time, and failing to prepare fast enough for a post-print world as Condé Nast fades away.
Van Der Kolk, The Body Keeps the Score: Brain, Mind, and Body in the Healing of Trauma. I was reading this at a cafe when a Berkeley undergrad struck up a conversation and told me it was one of the most read books of her class. It’s much more than a book by a psychiatrist and somehow goes deeper into the mysteries of human nature, describing how overwhelming experiences affect the development of the brain, mind, and body awareness, all of which are intertwined. We are all creatures forged on trauma, and some become stronger, others weaker.
Srinivasan, The Network State: How To Start a New Country. This is a wild book, at least comparable to Davidson’s “The Sovereign Individual” or Scott’s “Seeing like a State”, if not to older thinkers like Locke and Rousseau. The last 2 chapters are the best, and go into detail about the main idea: “A network state is a social network with a moral innovation, a sense of national consciousness, a recognized founder, a capacity for collective action, an in-person level of civility, an integrated cryptocurrency, a consensual government limited by a social smart contract, an archipelago of crowdfunded physical territories, a virtual capital, and an on-chain census that proves a large enough population, income, and real-estate footprint to attain a measure of diplomatic recognition.” My deeper thoughts on the Leviathan are here.
Hodges, Alan Turing: The Enigma. I read this book as a teenager in the 1990s when it seemed relevant in the PC Age, and loved it – Turing was a hero of mine. Revisiting it 25 years later, it’s even more relevant in the AI age, from the “On Computable Numbers” paper to his work on ACE and general computers, and his early work on AI. There’s also the controversy over his death: was it an accident, suicide, or murder? It pairs well with Copeland’s book “The Essential Turing” (which has the best intro and explanation to the “Computable Numbers” paper, ACE, and some of his AI papers and talks). The ACE work pairs well with reading Von Neumann’s EDVAC report, as together they laid the basis of modern computer engineering.
Gross-Loh and Puett, The Path: What Chinese Philosophers Can Teach Us About the Good Life. This book is based on one of the most popular classes at Harvard College. It provides sketches on Confucius, Mencius, Laozi, Zhuangzi, Xunzi, and others on relationships, decisions, influence, vitality, spontaneity, humanity, and more. It is concise and yet deep, with many wise views that are not often considered in the West.
Blattman, Why We Fight: The Roots of War and the Paths to Peace. War is hell, and it’s also deeply woven into the human psyche and society. So what does that say about us? Blattman delves into the root causes of war (unchecked interests, intangible incentives, uncertainty, commitment problems, misperception) and some paths to peace (interdependence, checks and balances, rules and enforcement, interventions). This book pairs well with Rudd’s sensible new book “The Avoidable War: The Dangers of a Catastrophic Conflict between the US and Xi Jinping’s China.”
Hargittal, Martians of Science: Five Physicists Who Changed the Twentieth Century. “The Martians” were a group of prominent Hungarian scientists of Jewish descent who emigrated from Europe to the United States around WWII. Leo Szilard was asked why there is no evidence of intelligent life beyond Earth despite the high probability of it existing, and he responded: “They are already here among us – they just call themselves Hungarians.” The main Martians were Von Karman, Von Neumann, Szilard, Teller, and Wigner, who gave the world modern aeronautical engineering, the modern computer and simulations, atomic fission theory, hydrogen bombs, and the theory of the atomic nucleus and the elementary particles.
Tchaikovsky, Children of Time. A dark sci-fi novel about the remnants of humanity in ark ships traveling through space, who encounter a strange arachnid species after an experiment went wrong on a distant planet. The main plot twist comes from the winding passage of time, as the author plays it out on civilization and ark-ship scale over centuries.
Oliver, New and Selected Poems. Oliver is an under-appreciated New England poet, but beloved in the Bay Area yoga community for her poems on nature with imagery from her daily walks near her home. Some of her great poems are “Wild Geese”, “The Swan”, “The Summer Day”, “Invitation”, “Don’t Hesitate”, and “When Death Comes.” One poem starts: “Who made the world? / Who made the swan, and the black bear? / Who made the grasshopper? / This grasshopper, I mean— / the one who has flung herself out of the grass, / the one who is eating sugar out of my hand, / who is moving her jaws back and forth instead of up and down— / who is gazing around with her enormous and complicated eyes.”
Chiang, Stories of Your Life. Exhalation. Some thoughtful short stories that leave a chill. “Story of Your Life” gave us the movie “Arrival” – both are remarkable works of art, parallel and illuminating each other.
Ording, 99 Variations on a Proof. This is a strange book. The author picks a mundane cubic equation and finds 99 ways to prove it. Some are standard proofs (geometric, inductive, topological, algorithmic), but many are whacky and creative (blog, antiquity, tea, dialogue, hand-waving, screenplay, social media, etc). Besides showing how broad a mathematical proof could be, it tests the artificial line between fiction and non-fiction.
Devlin, The Millennium Problems: The Seven Greatest Unsolved Mathematical Puzzles Of Our Time. These are the seven most consequential math problems to be solved, per a committee in 2000 consisting of Alain Connes, Arthur Jaffe, Andrew Wiles, and Edward Witten, after consultation with other leading mathematicians. Only one has been solved, and six more would revolutionize their fields if they were (e.g. P=NP, the Riemann Hypothesis, Navier-Stokes, etc). It’s hard to explain them with relatively simple math and logic, and Devlin does that admirably. For the technical problem descriptions, see the CMI/AMS publication by Carlson, Jaffe, and Wiles called “The Millennium Prize Problems.”
Wells, Prime Numbers: The Most Mysterious Figures in Math. I’m obsessed with primes, the building blocks of all numbers and math. My sense is we only know some small percent of what we can know about them, and this book organizes centuries of prime research alphabetically, which gives it an irreverent randomness. Favorite sections were on abundant and amicable numbers, Andrica’s conjecture, formulae for primes, Gilbreath and Goldbach’s conjectures, permutable primes, and more.
Gauss, Disquisitiones Arithmeticae (trans. Clarke). The Prince of Mathematics wrote this book in Latin in 1801, and it’s one of the foundational texts of number theory. I’d call it one of the hardest books I’ve ever tried to read, though to be honest, it slayed me. I could follow the beginning of every chapter, and then like a wizard old Carl would take it up one or two notches and just flatten me. This book pairs well with Dunnington’s bio, “Carl Friedrich Gauss: Titan Of Science.” Disquisitiones is only for mathematical masochists, or if you want to feel extremely stupid.
Patterson and Hennessy, Computer Organization and Design. I never studied computer engineering and basics like instruction sets, arithmetic, CPUs, memory hierarchies, storage and I/O, etc. The closest I got were some popular books from Petzold and Danny Hillis, so I went straight to the best teachers to learn more. My favorite chapter was the one on GPUs.
Reimann, Manifolds on the hypotheses (Ueber die Hypothesen, welche der Geometrie zu Grunde liegen) (1851). This is a mind-bending paper that crosses the line between math and philosophy, has no equations, and yet extends Gauss’s theory of 2d surfaces to n-d space, where notions of distance, angles, and curvature can still be measured. It’s relevant to gravity, general relativity, and modern machine learning (where neural networks learn manifolds in n-d space).
Metropolis and Ulam, The Monte Carlo Method (1949). A paper created to help model nuclear explosions that became important to the quantitative finance industry for decades (I learned the methods in school but never read the original paper or its history). Ulam was also the co-inventor of the Hydrogen bomb and [Ulam spiral](https://en.wikipedia.org/wiki/Ulam_spiral).
Alonso and KOE. Zero to Monero (Whitepaper)(2018). Monero may be the most interesting project in all of crypto, competing with Bitcoin and Ethereum, and a few others. It’s basically Bitcoin with owner and transactional privacy. Monero is a standard one-dimensional distributed acyclic graph (DAG) cryptocurrency blockchain where transactions are based on elliptic curve cryptography using curve Ed25519.
Brynjolfsson et al. What Can Machines Learn and What Does It Mean for Occupations and the Economy? (2018). The authors investigate how machine learning (ML) will transform numerous occupations and industries at the task level. They apply a rubric evaluating task potential for ML and find that (“i) ML affects different occupations than earlier automation waves; (ii) most occupations include at least some SML tasks; (iii) few occupations are fully automatable using ML; and (iv) realizing the potential of ML usually requires a redesign of job task content.”
Aghion et al. Robots and automation (2021). A fascinating paper that uses a dataset of firms in France that suggests instead of destroying jobs, “automation has a positive direct effect on employment at the firm level” and could help workers, as firms with automation become much more productive, lower their quality-adjusted prices, and so benefit by higher market size and more employment.
Acemoglu, Harms of AI (2021). One of the world’s best research economists asking the right questions about where AI could go wrong, but with frustratingly simple and naive models to back up his hypotheses. I would like to see him take the brilliant intuition in this paper and gather datasets to prove or falsify them.
Lecun, A Path Towards Autonomous Machine Intelligence (2022). A paper by a leading ML researcher that aims to figure out what research directions could get us to the answer “How could machines learn as efficiently as humans and animals?”. He follows down the Minksy path of suggesting they need specific sub-models like a configurator, perception model, world model, cost model, actor model, and short-term memory (except Minsky suggested thousands of sub-models may be needed). This paper describes a roadmap for developing machines whose behavior is intrinsic rather than hard-wired or requiring supervision/rewards, with the key being self-supervised learning. My sense is that LeCun is right in the direction but wrong on the details, with reputations and fortunes to be made on getting the right details.
Habgood-Coote et al. Commentary on “Can a good philosophical contribution be made just by asking a question?”(2022). The paper itself is just the question, but thankfully the commentary goes into detail. It’s a good reminder that great questions are worth much more than their purported answers.
Manning, Human Language Understanding & Reasoning (2022). A clear and concise explanation about the “surprising breakthroughs in natural language processing through the use of simple artificial neural network computations, replicated on a very large scale and trained over exceedingly large amounts of data.” Basically the early stirrings of artificial general intelligence and foundation models that can connect sensory experiences, language, thinking, and feeling.
Hoffman et al. Training Compute-Optimal Large Language Models (the “Chinchilla’ paper) (2022). A stunning result from Google AI on how to better scale large language models. Keeping FLOPS constant, they find the ideal model size, # of tokens, and hints at dataset size. A huge accomplishment for massive model scaling.
Anil et al. On the Factory Floor: ML Engineering for Industrial-Scale Ads Recommendation Models (2022). The most important large-scale AI models in production in the world today are ads models, followed by search and content ranking models. Rarely do the ads engineers publish their thoughts, and this is a fun paper looking into what they do in this $500bn industry.
Fawzi et al. Discovering novel algorithms with AlphaTensor (2022). Matrix multiplication (matmul) is the heart of modern AI and ML – this paper uses AI to discover new matmul algorithms that are more efficient, and is an early landmark in AI pushing science while also making AI better (AI improving AI is the holy grail). A simple explanation.
Brown, Meta AI, et al. Human-level play in the game of Diplomacy by combining language models with strategic reasoning (2022). A simple yet also hard paper on how to combine a language model that negotiates with humans, a strategy module that makes decisions, and a filtering system that negates bad dialogue.
Chowdery et al. PaLM: Scaling Language Modeling with Pathways (2022). Google made a 540-billion parameter, densely activated, Transformer language model on 6144 TPU v4 chips using Pathways, a new ML system that enables highly efficient training across multiple TPU pods with near 100% utilization. Wow. This infrastructure achievement showed the benefits of model scaling by getting SOTA few-shot learning results on hundreds of language understanding and generation benchmarks. Basically it brings us one step closer to linguistic general intelligence. PaLM uses many advances like: SwiGLU Activation; Parallel formulations in transformer blocks; multi-query attention RoPE; Shared Input-output embeddings, and; SentencePiece vocabulary.
And a final quote to end a tough year:
No single thing abides; but all things flow.
Fragment to fragment clings–the things thus grow
Until we know and name them. By degrees
They melt, and are no more the things we know.
Globed from the atoms falling slow or swift
I see the suns, I see the systems lift
Their forms; and even the systems and the suns
Shall go back slowly to the eternal drift.