“I’ve spent my life trying to undo habits—especially habits of thinking. They narrow your interaction with the world. They’re the phrases that come easily to your mind, like: ‘I know what I think,’ or ‘I know what I like,’ or ‘I know what’s going to happen today.’ If you just replace ‘know’ with ‘don’t know,’ then you start to move into the unknown. And that’s where the interesting stuff happens.” — a quote from Humans of New York
These are the ones that showed me what I did not know:
On Poetry
Elizabeth Bishop, “Questions of Travel.” Bishop asks whether she had any right to come and look. “Should we have stayed at home and thought of here?” The poem does not answer and does not let you off.
Seamus Heaney, “Digging.” His father cutting turf, his grandfather before that, and Heaney at the window holding a pen instead of a spade. He decides the pen is a tool as well. Whether he has convinced himself is the poem.
Philip Levine, “What Work Is.” Levine worked the Detroit auto plants and wrote about it for fifty years without once making it noble. The poems know what a shift does to a body, and what the body thinks about while it happens.
W. S. Merwin. Merwin dropped punctuation from his poems, and spent nearly forty years turning a ruined pineapple plantation on Maui back into palm forest, planting trees he would not see grown. Both are the same decision.
Bashō, “The Narrow Road to the Deep North.” A travel journal with poems in it, walked in 1689. Most of it is walking. The poems arrive when something stops him.
Richard Brautigan, “All Watched Over by Machines of Loving Grace,” 1967. A “cybernetic meadow where mammals and computers live together in mutually programming harmony.” Whether Brautigan means it or is mocking it has been argued for almost sixty years. I read it as both, which I think is the only way it holds.
Agha Shahid Ali. Ali brought the ghazal into English properly, refusing the loosened version American poets had been writing under the name. Kashmir is in nearly all of it. He is very good on the difficulty of writing about a place from somewhere else.
Kabir. A fifteenth-century weaver who dictated his verses because he could not write them. Plain speech, no ornament, and consistently rude to priests and scholars alike.
Miroslav Holub. An immunologist who wrote poems, or a poet who ran a laboratory, depending on who is asking. The poems are clinical, cold, and occasionally devastating.
On Intelligence
Alan Turing, “Computing Machinery and Intelligence,” 1950. Turing declines to answer whether machines can think and asks instead whether we could tell the difference. Seventy-five years on, the objections he anticipated in the same paper are still the objections being raised.
John McCarthy, Marvin Minsky, Nathaniel Rochester and Claude Shannon, “A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence,” 1955. The document that named the field. It proposes to make significant progress on machine intelligence over two months, with ten minds. Worth reading for the confidence alone.
John Searle, “Minds, Brains, and Programs,” 1980. Searle puts a man in a room with a rulebook for manipulating Chinese characters he cannot read, and asks what exactly is different about the computer. Forty-five years of rebuttals have not dislodged the argument, and none of them have satisfied Searle either.
Thomas Nagel, “What Is It Like to Be a Bat?,” 1974. You can know everything about echolocation and still not know what it is like. Nagel’s claim is that subjective experience is a fact about the world no third-person description reaches. Anyone confident about machine consciousness, in either direction, should read it first.
J. C. R. Licklider, “Man-Computer Symbiosis,” 1960. Written when a computer filled a room, and arguing that the goal was never to replace human thinking but to couple it to something that does not tire. Almost everything I work on is a footnote to this paper.
Douglas Engelbart, “Augmenting Human Intellect: A Conceptual Framework,” 1962. Engelbart set out to raise the collective intelligence of people facing problems too complicated to solve, and invented the mouse and hypertext along the way as incidentals. The framework was the point. The incidentals are what we kept.
Richard Sutton, “The Bitter Lesson,” 2019. Two pages. Seventy years of AI research reduced to a single finding: methods that scale with computation beat methods that encode what humans know, every time, and researchers keep declining to believe it. Bitter is the right word.
David Marr, “Vision,” 1982. Marr insists you say which of three questions you are answering: what the system computes, what algorithm it uses, or how the hardware implements it. Confusing the levels is still the most common way for two people to argue past each other about intelligence.
Norbert Wiener, “The Human Use of Human Beings,” 1950. Wiener’s account of cybernetics for people who were never going to read the mathematics, and the book where he works out what feedback and control mean for a society rather than a servomechanism. The title is the argument: the danger he is describing is not machines becoming human but humans being used as though they were parts. He had already written to the United Automobile Workers offering to help them prepare for what automation would do to labor. That was 1949.
On Computer Engineering
Alan Turing, “On Computable Numbers, with an Application to the Entscheidungsproblem,” 1936. Before there were computers, Turing described one, and then proved there are things it will never do. The machine was hypothetical. The limits were not.
Ken Thompson, “Reflections on Trusting Trust,” Turing Award lecture, 1984. Thompson shows you can hide a backdoor in a compiler so that it appears in neither the compiler’s source nor the program’s. The conclusion is that you cannot trust code you did not write yourself. He then notes you cannot entirely trust that either.
Claude Shannon, “A Mathematical Theory of Communication,” 1948. Shannon defines information as the resolution of uncertainty, and most of the twentieth century follows. It is also a pleasure to read, which is not something one usually says about the founding document of a field.
Peter Naur, “Programming as Theory Building,” 1985. Naur argues that a program is not the code but the theory held in the minds of the people who built it, and that when those people disperse, the theory dies even though every line survives. Rebuilding the theory from documentation does not work, because what was lost was never written down. The most unsettling paper here, and the one that has aged best.
Fred Brooks, “No Silver Bullet: Essence and Accident in Software Engineering,” 1986. Brooks divides the difficulty of software into the essential, which is the complexity of the problem itself, and the accidental, which is everything our tools impose on top. Better tools keep removing the accidental. The essential does not move. Four decades of methodologies have been sold on forgetting this.
Leslie Lamport, “Time, Clocks, and the Ordering of Events in a Distributed System,” 1978. With no shared clock, “at the same time” stops meaning anything, and Lamport builds an ordering out of nothing but messages. He took the intuition from special relativity. The most cited paper in distributed computing and still the clearest.
Melvin Conway, “How Do Committees Invent?,” 1968. A system’s structure will come to mirror the communication structure of the organisation that built it, whether or not anyone intends it. Usually quoted as a warning. Better read as an instrument: to change the architecture, change who talks to whom.
Edsger Dijkstra, “The Humble Programmer,” Turing Award lecture, 1972. Dijkstra’s claim is that the binding constraint in programming was never the machine but the size of a human skull, and that every genuine advance has come from arranging to need less of it in one place at one time.
Jim Waldo, Geoff Wyant, Ann Wollrath and Sam Kendall, “A Note on Distributed Computing,” 1994. You cannot paper over the difference between a local call and a remote one, because latency, partial failure, and concurrency are not implementation details that can be hidden. Thirty years of frameworks have tried anyway.
“Principles behind the Agile Manifesto,” 2001. Twelve sentences. Worth rereading against whatever process is currently being run in their name.
On Robotics
Karel Čapek, “R.U.R.,” 1920. The play that gave us the word, taken from robota, meaning forced labor. Čapek’s robots are not machines but manufactured people, and the plot turns on their being cheaper than workers. A century of engineering has borrowed his noun but declined his argument.
Rudolf Kálmán, “A New Approach to Linear Filtering and Prediction Problems,” 1960. How to estimate where something is when every measurement you have is wrong. Almost nobody was interested at first. Nine years later it was running in the computer that brought the lunar module down, and nearly every robot since has carried some version of it.
Valentino Braitenberg, “Vehicles: Experiments in Synthetic Psychology,” 1984. Braitenberg imagines machines with two sensors and two motors, wires them slightly differently, and watches observers call them timid, aggressive, or in love. The lesson is about the observer, not the vehicle. The shortest cure I know for over-reading behavior.
Hans Moravec, “Mind Children,” 1988. Where the paradox is stated: what we find hard, like chess and calculus, is easy for machines, and what any child does without thinking, like seeing and walking, is the hard part. Thirty-five years have confirmed it in both directions and explained it in neither.
Rodney Brooks, “Intelligence Without Representation,” 1991. Brooks argues that a mind does not need an internal model of the world, because the world is right there and can be inspected, and then builds robots that walk without one. The strongest case I know that intelligence requires a body.
Masahiro Mori, “The Uncanny Valley,” 1970. Mori proposed that our affinity for a machine climbs as it grows more humanlike and then falls off a cliff just short of the resemblance being complete, and that movement deepens the drop. His examples were prosthetic hands, bunraku puppets, and corpses. He wrote it as practical advice, telling designers to stop at the first peak rather than chase the second, and said later that he had never meant it as science. Fifty years of studies have not settled whether the valley exists. It continues to shape the field anyway.
On Physics
My father gave me “The Feynman Lectures on Physics” when I was in middle school in India, which is how physics arrived for me: as pleasure rather than syllabus. “QED: The Strange Theory of Light and Matter” and “The Character of Physical Law” came later. When my younger brother was eight, he saved his money and bought me “Surely You’re Joking, Mr. Feynman!” for my eleventh birthday. We still read it, and “What Do You Care What Other People Think?”, and we still find them funny. What Feynman teaches before any physics is that you are allowed to approach hard things lightly.
Richard Feynman, “Personal Observations on the Reliability of the Shuttle,” Appendix F to the Rogers Commission report, 1986. Feynman’s dissent, appended at the back because the commission would not put it in the body. He works out that management was estimating failure at one in a hundred thousand while their own engineers said one in a hundred, and asks how both numbers survived inside the same organization. It ends: “reality must take precedence over public relations, for nature cannot be fooled.” The same man as the books above, and he does not sound like him at all.
Edward Purcell, “Life at low Reynolds number,” 1977. At the scale of a bacterium, water behaves like tar and coasting is impossible: stop swimming and you stop instantly. Purcell works out what it is like to live there. The clearest demonstration I know that physics does not merely scale, it changes.
J. E. Gordon, “Structures: Or Why Things Don’t Fall Down,” 1978. Gordon explains why things break, which turns out to be the better way to understand why anything stands. Written by an engineer who spent the war on wooden aircraft, and funnier than a book about stress and strain has any right to be. If you want to know what a material will do before you put heat to it, start here.
D’Arcy Wentworth Thompson, “On Growth and Form,” 1917. Thompson argues that the shapes of living things are set less by descent than by physics: surface tension, load, the arithmetic of scale. An insect cannot simply be enlarged into an elephant. A century of biology has qualified him without replacing him.
Philip Anderson, “More Is Different,” 1972. Anderson’s case against reductionism: knowing the rules at one scale tells you very little about what appears at the next. Chemistry does not follow from particle physics, and psychology does not follow from chemistry. Written about condensed matter, read now mostly by people arguing about machine intelligence.
Eugene Wigner, “The Unreasonable Effectiveness of Mathematics in the Natural Sciences,” 1960. Wigner finds it genuinely strange that mathematics invented for its own sake keeps turning out to describe the world, and then declines to explain it. He calls it a gift we neither understand nor deserve. The title has been borrowed so often that the essay itself tends to go unread.
On Research
Richard Hamming, “You and Your Research,” 1986 lecture. Hamming spent a career at Bell Labs watching who did important work and who did not, and this is what he concluded. “If you don’t work on important problems, it’s not likely that you’ll do important work.” He intends it as an accusation.
Martin A. Schwartz, “The Importance of Stupidity in Scientific Research,” 2008. Two pages arguing that feeling stupid is not a sign you are failing at research but a sign you are doing it. The discomfort never goes away. My doctoral adviser, Takeo Kanade, once told me: “You don’t look miserable enough to be doing meaningful research.”
Marc Levoy, “Where do disruptive ideas come from?,” 2012 commencement speech. Mostly they come from people willing to be wrong in public. “Know your fears, know also what you really want, weigh the odds, and occasionally, make a run for it.”
Daniel Kahneman and Amos Tversky, “Judgment under Uncertainty: Heuristics and Biases,” 1974. The paper that catalogued how reliably people misjudge probability. The findings are common knowledge now and still do not help. Knowing about a bias does not remove it.
George E. P. Box, “Science and Statistics,” 1976. Where “all models are wrong, some are useful” comes from. The essay around the line is better than the line, and far less quoted.
On Reading
Srinivasan Keshav, “How to Read a Paper,” 2007. Three passes: skim, grasp, evaluate. The permission to stop after the first pass is what changed how much I read.
Edward Hirsch, “How to Read a Poem,” 1999. The nearest thing poetry has to a manual, and it refuses to be one. Hirsch treats the poem as a message in a bottle thrown by someone who could not know who would find it, which makes reading it an obligation rather than a skill.
John Ciardi, “How Does a Poem Mean?,” 1959. The contribution is the title. Not what a poem means but how it means, which moves the question from decoding to mechanism. He took it from Brooks and Warren by reversing them.
Cleanth Brooks and Robert Penn Warren, “Understanding Poetry,” 1938. The textbook that taught American departments to read a poem as a system rather than a biography. It won so completely that its assumptions now read as common sense.
Billy Collins, “Introduction to Poetry.” A poem about how his students read poems: they want to “torture a confession out of it.” The shortest argument against method on this list, and the one I would put last.
On Writing
Simon Peyton Jones, “How to Write a Great Research Paper.” Peyton Jones’s argument is that a paper exists to convey one idea, and that most papers bury theirs out of a misplaced sense of modesty. Start writing before the research is finished, because you will not know what the idea is until you try to state it. Either liberating or heretical, depending on the day.
George Whitesides, “Writing a Paper,” 1989. Whitesides has his group keep a running outline from the first day of a project, revised continuously, treated as the real record of the work. The outline is not preparation for writing. It is the instrument that shows which experiments are missing. Written for chemists, useful to anyone.
Henning Schulzrinne, “Writing Technical Articles.” A list of the specific ways technical prose goes wrong. Unglamorous, immediately applicable.
George Gopen and Judith Swan, “The Science of Scientific Writing,” 1990. They take unreadable paragraphs from real journals, rewrite them, and name the rule behind each repair. Their claim is that readers have fixed expectations about where in a sentence information should arrive, and that most bad scientific prose is not confused thinking but correct thinking loaded in the wrong order. The most immediately useful thing in this section.
George Orwell, “Politics and the English Language,” 1946. Six rules, the last of which is to break any of the others rather than say something barbarous. The essay is not really about style. It is about how vague language makes dishonesty easy and eventually makes it comfortable, which is why it has outlived every example in it.
Richard Hugo, “The Triggering Town,” 1979. Hugo separates the triggering subject, the thing that started the poem, from the real subject, which you discover later and which is always yourself. His instruction is to abandon the triggering town the moment it has done its work, because loyalty to what actually happened there will kill the poem. I write mostly in transit, and nothing has been more useful to me about what to do with a place after leaving it.
Rainer Maria Rilke, “Letters to a Young Poet,” 1929. Ten letters to a nineteen-year-old military cadet who wrote asking whether his poems were any good. Rilke never answers the question. He tells him instead to find out whether he would have to write if he were forbidden to. Patient, severe, and not in any way encouraging, which is why it works.
Ezra Pound, “A Few Don’ts by an Imagiste,” 1913. Two pages of prohibitions. Use no word that does not contribute. Do not put into mediocre verse what is already good prose. Go in fear of abstractions. It is Schulzrinne’s list for poems, written seventy years earlier and considerably ruder.
Mary Oliver, “A Poetry Handbook,” 1994. The mechanics: the line, the sound, the meter, what a single syllable changes. Written by someone who clearly thought the mechanics were the least interesting part of poetry and taught them thoroughly anyway.
Robert Frost, “The Figure a Poem Makes,” 1939. Frost’s two-page preface to his collected poems. A poem begins in delight and ends in wisdom, by which he means the poet must not know the ending at the start. “No surprise for the writer, no surprise for the reader.”
On Leadership
“Solitude and Leadership.” William Deresiewicz delivered to the plebe class at West Point in 2009. He argues that leadership begins in the capacity to be alone with your own mind. He was saying this to a room of people who had arrived there by excelling at everything asked of them, and telling them that was the problem.
“Crucibles of Leadership.” HBR, 2002. On the experiences that remake a person: not the ones that test you, but the ones after which you are someone else. Most people can name theirs.
Eliyahu Goldratt, “The Goal.” A novel about a factory, which is a strange way to teach the Theory of Constraints and an effective one. Every system has one bottleneck. Work anywhere else and you have accomplished nothing. Jeff Wilke, then CEO of Amazon’s consumer business, recommended it to me, and I read it while leading research teams for Amazon Fulfillment Technologies and Amazon Logistics, where the lesson was not abstract.
“Crucial Conversations.” How to keep talking once a conversation has become expensive. Mechanical in places, which is the point: under pressure nobody improvises well.
Michael Watkins, “The First 90 Days.” What to do in a role before you understand it. The useful part is the permission to spend those weeks learning rather than proving.