Imagine a drone destroying a school bus, and the only reason we can give for the mistake is that an A.I. system directed it there. Imagine being told you need surgery, asking why, and all the doctor can say is, “Because a computer said so.” What if the computer is wrong? A.I. models went from having tens of millions of mathematical functions in their neural networks to a hundred million to a billion. It has been estimated that the latest versions of Google Gemini and OpenAI’s GPT-5 contain trillions of mathematical functions. But one cost of that improvement has been transparency. As a model’s neural net gets bigger, it becomes even more difficult to understand how it works. A growing field of computer science known as interpretability embodies the conceit that in order to narrow or even bridge the expanding knowledge gap between A.I. models and humans, we need to treat A.I. more like a natural phenomenon than a human invention. “The natural world is, after all, full of complex structures arising from unknown rules,” Oliver Whang writes. “Galaxies and starfish and cancer cells are all black boxes, in a sense.” FEATURES MEA CULPAS Editor’s NoteUsually, the Sunday Crossword that runs in The New York Times Magazine is difficult. This week, it’s impossible. Because of a production error, this week’s puzzle contains a grid that does not match the clues. We sincerely apologize for the confusion and aggravation this may cause. The mistake was discovered too late to fix in the magazine, but the correct version of the puzzle can be found in the news section of Sunday’s edition of the daily New York Times. THIS WEEK’S COVER
COLUMNS
FROM THE ARCHIVES 25 Years of LenaLena Dunham, who was recently a guest on The Interview and the subject of an essay from Amanda Hess, has been mentioned in the pages of The Times hundreds of times. However, the earliest mention of Dunham was actually in The New York Times Magazine all the way back in 2001. As part of a special issue about the future of New York City in the aftermath of the attacks on Sept. 11, the magazine ran a series called “Let Us Count the Ways” with short statements from New Yorkers about why they love the city. Dunham, who is identified as a student at St. Ann’s School in Brooklyn, said: The U.T.B. spot — Under the Bridge — has been passed down through generations of girls in my school. I have a friend who’s 36 who went to St. Ann’s, and she used to go to U.T.B. We go there after a school dance or a party or whatever. You walk past all these big, cavernous industrial buildings, and then there’s this natural spot you would have never expected. There are slats in the bottom of the bridge, and at night the lights from the cars going over the bridge flash through the openings like a strobe light. And the water is striped with the lights of the buildings, and the trash floating by makes it feel like parts of the city are alive and moving on the water. You can’t see any stars, but the water is like a reverse sky. COMMENT OF THE WEEK Does Artificial Intelligence ‘Think’?From Ben Goren on Oliver Whang’s story about how we don’t really know how A.I. works: To those who would dismiss the notion of A.I. ever thinking for whatever reason, I would counter that this is as counterproductive as claiming that airplanes aren’t really flying because they don’t flap their wings, they don’t have feathers, they don’t lay eggs, or whatever essential property you think birds have that makes them birds. A.I. is waaaay past the point of these sorts of naïve philosophical arguments. It really does think, really is conscious, really does have judgment, and all the rest. And it’s as different from humans in the way it does that as a 737 flies differently from a sparrow. This article is spot on in identifying that it’s pointless to argue such philosophy; what matters is to get to the point where we can gain as much understanding of and trust in an A.I.’s judgment as we do with fellow humans. And let’s not forget how opaque our own minds are to ourselves, let alone others; it’s not like this problem is unique to A.I. or to this moment in history. The scientific method, after all, is our best attempt yet at a general solution, and points to one obvious answer: don’t just trust the A.I., but have it provide proof and subject said proof to peer review. At which point, who cares what it’s “really” thinking? That’s all for this week. Email us at magazine@nytimes.com with your thoughts, questions and feedback. Stay in touch: Like this email? Forward it to a friend and help us grow. Loved a story? Hated it? Write us a letter at magazine@nytimes.com. Did a friend forward this to you? Sign up here to get the magazine newsletter.
|
