Join Puck to listen to this article
In just the past three weeks, three of the men who would be king of the A.I. revolution—Sam Altman, Dario Amodei, and Elon Musk—have moved to take their companies public at valuations so stratospheric that they’re less likely to make Croesus blush than to suffer a myocardial infarction. So it’s no surprise that the voluminous and typically breathless coverage of A.I. has lately tended to focus on those dudes: their aims and aspirations, peccadilloes and foibles, and whether any of them can be trusted with developing a technology that, ahem, just might kill us all.
The journalist and author Josh Tyrangiel, however, has been focused on a different, more fundamental, and arguably more relevant set of queries: What exactly is A.I. good for? What problems do we, as a society, collectively wish to address that the technology might help us solve? And what do we need to do right now to make sure A.I. winds up serving our collective interests rather than trampling over us?
Tyrangiel, a former editor of Bloomberg Businessweek and for several years the author of a regular Washington Post column on A.I., has just published a book addressing those very questions. Entitled A.I. for Good: How Real People Are Using Artificial Intelligence to Fix Things That Matter, it’s one of the least hyperbolic and most reasonable discussions of A.I. you’re likely to find, and thus a highly salutary corrective to the hyperventilation of both A.I. boomers and A.I. doomers that has dominated the discourse.
For a recent episode of my Impolitic podcast, Josh and I talked about how the media’s obsession with the dreams and schemes of A.I.’s poster boys misses the larger, more important story unfolding before our eyes; about places, from the Cleveland Clinic to the I.R.S., where A.I. is being deployed by doctors, researchers, and government officials to do real work; about why the mere mention of Palantir curdles the blood of progressives; and what it will take for political leaders to catch up to the changes A.I. will soon unleash on all of us. As always, this excerpt has been edited for clarity and concision, but you can listen to the whole thing here.
The Palantir Paradox
John Heilemann: Most A.I. coverage focuses obsessively on a handful of names—Elon Musk, Sam Altman, Mark Zuckerberg, Dario Amodei, Demis Hassabis. Your book mentions Sam Altman on four pages and Elon Musk on five. How do you resist that pull?
Josh Tyrangiel: The coverage has reached a point of oversaturation. At a certain point, I know what they’re after, right? They want to win, and that’s interesting. But to me that’s a short-term story. The biggest question is, What is this stuff going to do to make our lives better for generations to come? What I found was, nobody was really reporting on that. We were getting a kind of F12-level response. What good could it do? It could cure cancer, it could mitigate climate change. What bad could it do? It could cause our extinction or drain our existence of all meaning.
My bias inherently is toward complexity. I want to know what’s actually going on here. I understand it’s going to make some people a lot of money. I understand it’s going to revolutionize the interface of the web. But what about the problems we actually want to solve in our society? Can it be applied to those? I just couldn’t get answers, in part because these people are racing—the last thing they’re going to do is stop and say, “Let me work up some use cases for you.” I also couldn’t get answers because the type of people who run A.I. labs, by definition, are very, very good at understanding how to make models and tune models, and that is a very particular kind of intelligence. And it’s not mine, to be clear—it’s probably not yours. So the question I wanted to answer is, All right, what is this good for?
Right at the very beginning, literally the first page, you describe having what you call your A.I. awakening—when you’re sitting up late one night cruising YouTube, and you come across a Palantir video featuring a former four-star army general. You write that it made you feel optimism. Explain the epiphany, and how it set you on the path to write this book.
The video is basically just a retired general named Gus Perna, whom you’ll remember for having overseen Operation Warp Speed—the effort to distribute all the [Covid] vaccines. He describes being called in and told, “We’ve got to get all of these vaccines distributed to every single place in America simultaneously.” We needed to track our ability to do it so we could know when we achieved a certain rate of vaccination, and he basically had nothing at his disposal.
The natural flow of government took over. A bunch of consultants came in, somebody pitched a medical blockchain, somebody pitched a piece of hardware in every doctor’s office, in every CVS. And he’s just like, This is not the problem. Then a couple of people from Palantir came in and pitched him. They understood the problem; they didn’t overhype the solution. They understood it was going to be data that was going to do this—they were going to create machine learning algorithms within the data and give him a dashboard. He said, “Great, you’re hired.”
It turns out this is kind of what got us through the pandemic. It is the least glamorous form of A.I.—like, this is down in the pipes—and yet it did it. During the pandemic, I was reading the news just like everybody else, and I did not understand that that’s what got us there. So I started thinking, all right—if we can shore up our institutions using A.I., what will that mean for our relationship to government?
Palantir features prominently in the book, which is going to trigger a lot of readers. You even write that the very mention of Palantir—which was founded by Peter Thiel, one of Trump’s biggest backers, and Alex Karp, who was a socialist for a long time and backed Kamala Harris and Hillary Clinton—“curdles the blood of progressives.” What is it that you find so fascinating about Palantir? Why is the freak-out misplaced?
I get why people are attracted to and hate Palantir. The company was founded in part from funding through In-Q-Tel, which is part of the C.I.A. They also do things like supply the government with surveillance software for our enemies and domestically.
They also do things like make software that works, and the government has a history of buying software that doesn’t work. It’s not a perfect company, because there aren’t perfect companies. But I’m less interested in what people feel about Palantir than what the technology they make can do, and what it can do is bring order to a certain kind of systemic chaos. That systemic chaos—when you believe the government doesn’t work and will never work, when you believe healthcare can’t work—is actually just as damaging, if not more damaging, than providing software to ICE or others. The book is about how this stuff works, how we can use it the way we want to use it.
The book is really about laying out a bunch of examples where interesting and valuable stuff is being done, or starting to be done, with A.I. One of them is at the Cleveland Clinic—and Palantir is involved in the work there.
The Cleveland Clinic is one of the three best healthcare systems in America, which makes it one of the best in the world, and it runs margins of about 2 percent—that’s as good as it gets. The problem with being a hospital is that you’re basically a hotel—you have guests, you have staff, you have rooms, you have linens, food, all of it. Hotels make profits because they know when people are coming and they know when they’re leaving, and hospitals don’t.
So the Cleveland Clinic basically has this idea to use A.I. [to] help get closer to being a hotel than a hospital. What Palantir did—which is actually pretty dull work from a technology standpoint—is get all of the data that could lead one to more predictability, lay it out in a row, and clean it. Now the hospital has all that information—everything from understanding, in an electronic health record, when a nurse puts in that a patient is progressing well, to when a doctor says orally into their recorder, “We’re going to dismiss tomorrow.” So they can schedule transfers, make sure they’re staffed adequately. They reduced wait times in the emergency room by 90 minutes. This is not glamorous stuff, but that system works a lot better. We all touch healthcare—we should be demanding that we get the same results pretty much everywhere.
Regulating After the Catastrophe
You also write about the I.R.S., which you’d think would be the least likely institution to be doing anything interesting with A.I.
The agency is the most abused and most neglected—abused because everybody hates it, neglected because nobody wants to adequately fund it. And yet if it doesn’t work, we have no country. A guy named Danny Werfel, who was the commissioner under Biden, basically said, All right, we have to restore faith in our ability to do the job. Looking at the Individual Master File—this massive program that has not just your tax record, but every change you have ever made to your tax record—he realized you could do a lot of interesting things with that, including maybe giving you some predictive idea of what you may owe this year. You and I just went through tax season; we have to guess what we owe the I.R.S. under penalty of prosecution.
So, very quietly for the past eight years, the I.R.S. has been modernizing the I.M.F. so it can begin to do all of those things, not just with A.I. and machine learning, but by bringing people along. They got pretty close to implementing all these changes in the I.M.F. They are using these tools to actually go after tax cheats. The sad irony of the whole chapter is that the very first consequence of Trump being elected was DOGE. The intent was never to understand the system—it was to destroy the system. And guess what? The system won.
Dario Amodei has been saying publicly that A.I. could produce high G.D.P. growth alongside high unemployment—a combination we’ve basically never seen before. Nobody in politics seems to be talking about this at a scale commensurate with the challenge. What do we need to demand from political leaders?
I did a cover for The Atlantic a couple months ago about A.I. and the future, and one of the stunning things was just how few people in the middle of politics are thinking about it, talking about it, acting on it. But there are a couple of people who are thinking very deeply about it, and those people have a couple things in common: They lead political movements, and they have unbelievable instincts for fury, for knowing when their people are angry. And those people are Bernie Sanders and Steve Bannon.
Bernie has put out a series of policies—we need to nationalize some of this, we need to create four-day work weeks, we need to get way far ahead of it. And when I went to visit Bannon, he was like, I absolutely fucking agree—not only do I agree, but we need to go farther. These labs are being run by people who don’t care about the republic. And Bannon even said, Look, I’m an anarchist, but if you don’t have a regulatory apparatus for this stuff, just don’t bother. Just tear it all down. So I think those two people and their knowledge of politics indicate to me that this is moving far faster than most politicians understand. If I had to bet, we regulate after the catastrophe—that’s just what we do.