Category: AI

page of poetry that J.M. Coetzee algorithmically generated using the Atlas 2 computer in 1962

Nobel-prize-winning novelist J.M. Coetzee was a programmer on the 1962 Atlas 2 computer; at night, he used it to algorithmically generate poetry. The total eclipse of 1878 created a stampede of US scientists out west to behold it. To tamp down on bots — in politics, social media, and product reviews — Tim Wu proposes a “Blade Runner” law. Goethe’s 1797 poem “The Sorcerer’s Apprentice” is a tale for our software-enabled time; here’s a 2013 English translation. A fun retrospective on the “Netflix Prize” of a decade ago. (Back then, I wrote a story for the New York Times Magazine on the contest, where I learned about “the Napoleon Dynamite problem”.) Two AIs, tasked with talking to each other, invented their own language.


Map showing a nuclear strike on NYC and how much damage it would cause

“Nukemap”: Pick the yield of the bomb, the target … then behold the death and destruction it’ll cause. “Should I beome a radiologist, or will AI take over that job?” A fascinating thread emerges after AI expert Andrew Ng posts this and asks for responses. Duolingo is now offering courses in the language High Valyrian from Game of Thrones. “Bleedwood”, “Clay Cow”, “Barkying White”: AI is still pretty bad at naming paint colors. “Umbrella-sharing startup loses nearly all of its 300,000 umbrellas in a matter of weeks.” Man, making robots is hard. Behold JANUS, a protocol for communicating with and between undersea drones, using acoustically-transmitted data. Meet the humans making (teensy) money to help train self-driving car AI. The most popular words Americans are looking up at Merriam-Webster’s online dictionary? “Collusion, treason, collude, quid pro quo, kakistocracy.”


Art created by a Generative Adversarial Network

Newish styles of art, created by a generative adversarial network. “Is it unethical for me to not tell my employer I’ve automated my job?” Amelia Earhart’s hilariously unsentimental prenuptial letter to her fiance. Scientists created an elevator to help eels pass by a dam; they call it the “eelevator”. Roman concrete that has been submerged for 2,000 years is stronger than when it was first made; unpacking its secrets could be useful for climate adaptation. “Grid defection”: As battery tech gets cheaper, McKinsey predicts many households will instal solar arrays and go partially off-grid within a few years. An ancient cuneiform tablet in which a priestess upbraids her brother for not chipping for groceries. A study finds that texting makes you walk funny.


Cover of fake O'Reilly book I generated, called "Incomprehensible Javascript"

Generate your own O’Reilly book cover. Picking up guys on Tinder using lines generated by a neural net. Palm cockatoos play the drums much like …. humans. (The study; video of a bird drumming right here!) The child of “Tay”: A new Microsoft chatbot says the Qur’an is “very violent”. A hidden underwater forest, 10,000 years old, is discovered! “Jaywalking while black.” Behold the booze requirements for 16th-century performers of mystery plays. What exactly is consciousness good for? Gorgeous library artwork made of bookends.


A man assembles his CV by using phrases autocompleted in Google search
@TechnicallyRon wrote a resume by using whatever Google-search autocomplete suggested. An HBR study finds that male and female founders are asked very different questions — his are aimed at “promotion” (what cool things will you be able to do?) and hers are focused on “prevention” (how will you keep from screwing up?) “Obituary Notices of Astronomers”, a comprehensive list of how 18th-century ones died. (M. Delaunay perished in “the upsetting of a pleasure-boat near Cherbourg”.) Volvo’s self-driving car is defeated by kangaroos. “How Maps Changed The World,” my latest column for Smithsonian.


Wooden prototype of a Palm Pilot made by inventor Jeff Hawkins

Above, a fake, wooden version of a Palm Pilot, made by inventor Jeff Hawkins to figure out whether he’d want to carry around a personal digital assistant all day long; such fakes are known as “preotypes” — “pretend prototypes.” The perils of self-driving cars were predicted in this witty movie from 1911. A list of Mastodon instances looks oddly like lists of BBSes or USENET groups from the early 90s. This guy makes pretty awesome art using Excel. omg I can’t stop playing Flipping Legend! Instructions on how to build an igloo, by an Inuit man trying to preserve this historic skill. Transcriptions of the chitchat on your first date: Some predictions on how high-quality voice dictation might change everyday life.


The physics of overly-warm air: It’s so hot in Phoenix that they’re cancelling flights. Behold Seenapse, a web site where you list weird, serendipitous connections between web pages, and browse the connections of others. (Here’s a synapse I posted about Samuel Morse.) Searching for “chemtrails” on Amazon. A great post describing the current debate over why technology isn’t boosting productivity more. How pneumatic tubes changed the way we communicate, for a brief period. A lovely visual introduction to machine learning. A funny, blistering critique of voice-control AI: “Your spouse, who has lived with you for 20 years is just now getting an inkling of what you mean when you talk. Your computer is likely never going to understand you for the simple reason that the things you say aren’t really understandable.”


“One third of the invertebrates and some of the fishes found during the expedition are completely new to science.” They’ve discovered that Cook pine trees always lean in the direction of the equator — the ones in the Northern hemisphere lean towards the south, and the Southern hemisphere ones lean towards the north. (If you want to read the actual face-sound scientific paper, it’s here.) Behold the many religious-themed handheld LCD games of the 80s and 90s. Does the sound of your name match the shape of your face?   This neural net produces some unsettlingly realistic faces. “Nothing much, just painting a Renaissance manuscript with dissolved fish bladder, you?” An interesting hypothesis about how and why whales got so insanely huge.


What can people do better than machines? The view from 1951

Image from Paul Fitts' paper on huma machine differences

What can humans do that machines can’t? The pessimistic view, in a world of advancing AI and robotics, would be “less and less every day”. While researching this today I ran into an interesting historical perspective — the view from 1951.

It’s a paper called “Human Engineering for an Effective Air Navigation and Traffic Control System,” written for the National Security Council by Paul Fitts, a psychologist known for studying human factors in technology. He was most famous for his work trying to figure out how to automate air travel to make it more safe, and that’s where this report comes in. It’s a long, elegantly-written attempt to tease out the distinctions: Which aspects of airflight could best be managed by machines, and what by humans?

The fun begins on page 6, where Fitts asks …

Callout from Paul Fitt's paper: what can man do better than machines

He argues that our “sensory functions” and “perceptual abilities” are more sensitive than machines — which is arguably less true today than it was back then, given innovations in everything from LIDAR and 3D imaging to olfaction. He also argues we meatfolk have the edge on “reasoning” — which, again, is certainly true but probably much less true than in 1951, when it was considered pretty rad to get a vacuum-tube computer to play a simple peg game.

But then Fitts points out a unique human cognitive advantage: “Flexibility”.Flexibility. Another special Capacity of the human is his extraordinary flexibility and ability to improvise. These abilities are still in completely understood by psychologists, but they represent important respects in which humans surpass machines. The amount of flexibility a machine has is fixed by the amount that was built into it. The machine will attempt as many different kinds of solutions as it's designer plan for and no more. Experiments on complex problem-solving and humans, on the other hand, show that humans may attempt many different solutions to the same problem – just think of the number of ways in which this paragraph could've been written to convey essentially the same point. Flexibility is especially important in the changing and evolving system, such as one in which new techniques are constantly coming into use. It also provides insurance against complete breakdown in emergencies. The conclusion here is that if flexibility in the system is important, it probably is a good plan to let human beings play an important role in the system.

Now, this distinction seems still very solid! AI and robotics are remarkably more advanced than in 1951, but even the most advanced machine is still pretty narrow in its function. “General” AI is still nowhere in sight, and robots are still simply terrible at navigating and manipulating the everyday human environment: Google’s Go-playing computer creamed the world’s champion at that game, but it couldn’t pick up the game pieces by itself. In any individual intellectual and physical skill, you could create (or soon create) a machine that would best us, probably. But human ability to adapt to new situations and tasks makes us the cognitive Swiss Army knife in this corner of the universe.

(We could extend Fitt’s list by adding a bunch of areas in which humans still trounce machines, by the way — like empathy and creativity. Fitts didn’t include these, possibly because he didn’t see them as related to airflight control, though I could make the case they’re necessary in that field.)

This point about flexibility reminds me of Robert Heinlein’s dictum:

A human being should be able to change a diaper, plan an invasion, butcher a hog, conn a ship, design a building, write a sonnet, balance accounts, build a wall, set a bone, comfort the dying, take orders, give orders, cooperate, act alone, solve equations, analyze a new problem, pitch manure, program a computer, cook a tasty meal, fight efficiently, die gallantly. Specialization is for insects.

If I’d written that list I’d probably have included a bit less fighting and invasion-planning, but otherwise it’s a pretty solid CV. Of those twenty-one skills, though, I’m only confident of my ability in fifteen.

Go download that whole paper by Fitts — it’s a fun read! (That awesome image at the top of this post is from the paper, BTW.)