Will A.I. Help Humanity? A Journalist Looks for Answers

In “AI for Good,” Josh Tyrangiel travels to the classrooms, hospitals and research labs where people are using artificial intelligence that might benefit society.

Even for skeptics, it’s getting harder and harder to argue that the obsession with artificial intelligence is pure hype. A.I. advances are fueling a historic stock market boom that may bust some time soon, but the responses that come from the latest versions of ChatGPT and Claude still often feel like magic. Pretty much everyone in the business world thinks what we’ve seen so far is just a taste of what’s to come — that artificial intelligence advances will inevitably lead to far better products and far lower labor costs.

That promise lies in the future, but the reality is that A.I. is here and already changing society. In “AI for Good” Josh Tyrangiel charts the present-day impact of this technological revolution with a series of case studies in which ambitious researchers, educators and entrepreneurs, mostly from outside Silicon Valley, familiarize themselves with the hot technology of the moment and try, movingly at times, to do something positive with it.

“AI for Good” grew out of a regular column Tyrangiel wrote for The Washington Post in 2023 and 2024, and it is useful at a moment when workers are being asked (and sometimes forced) to adopt and master Claude, ChatGPT and similar tools. The result is the kind of book designed to be enthusiastically recommended by chief innovation officers and management consultants. For better or worse, your boss will love it.

Tyrangiel wisely sidesteps the technical debates within the industry about what A.I. is. (ChatGPT is, of course, A.I., but so is the technology that transcribes your voice mail messages and lets you deposit a check through the bank app on your phone.) Tyrangiel also smartly avoids the bong-hit intellectualism that sometimes seeps into discussions of the technology’s risks. (Google “paperclip problem” if you want to go down that rabbit hole. Or, sure, ask a chatbot.)

It’s also refreshing to read an A.I. book that doesn’t concern itself with one of a handful of morally dubious masters of the universe, and instead lingers, for instance, on an Indiana school superintendent named Peggy Buffington, who sounds far more cleareyed about the challenges A.I. companies face than Sam Altman or Jeff Bezos.

“Some kids were just bullying the heck out of it,” she tells Tyrangiel, describing her school district’s experiments with a tutoring chatbot known as Khanmigo. “Cussing at it. I mean, you can figure out where that went.”

The first quarter of the book covers the development of Khanmigo, created by OpenAI and the nonprofit Khan Academy. The reception is mixed. At one point, after quoting a parade of adults who are awed, in one way or another, by the potential of A.I., Tyrangiel meets actual eighth graders who are, he reports, “wildly unimpressed.” One calls the chatbot “not helpful” for complex questions and “not close to the same as an actual teacher.”

This seems troubling, though the number of school districts using Khanmigo is growing quickly anyway. “The marketplace has rendered its own verdict,” Tyrangiel writes.

Next we’re off to the Cleveland Clinic, where the hospital’s C.E.O., Tomislav Mihaljevic, hopes that A.I. will help save lives. A new machine learning system designed to help doctors flag patients at highest risk for sepsis coincides with a lower rate of sepsis deaths, likely catching some cases that doctors would have missed.

These results are promising, if a little muddy. Despite improvements, the software continues to produce a number of false positives and it’s not clear if the A.I. deserves much credit for the drop in fatalities. As Tyrangiel notes, it’s possible Cleveland Clinic employees were simply on higher alert and caught more cases than usual because they knew their work was being matched against the A.I. Still, Mihaljevic frames this, in Tyrangiel’s words, as a “proof of concept” that suggests “A.I. does not need to be perfect to be useful.”

The book’s strongest section involves the efforts of an M.I.T. researcher named Kristy Johnson, who uses A.I. to make sense of the vocalizations of her developmentally disabled son. Johnson developed her tool by crowdsourcing audio recordings of nonverbal children along with their parents’ judgments about what those sounds mean. It’s a thrilling account, as an expression of both technical audacity and parental affection.

“AI for Good” is structured like a travelogue and the author’s prose is breezy, concise and well observed. But travelers don’t always come home with a nuanced understanding of what they’ve seen, especially if they don’t stray far from the double-decker tour bus.

Author: Max Chafkin, The New York Times

Read more...