Some interesting things to read the second weekend of February

Dear Friends,?

The most intense essay I’ve read in recent months is the story, in Esquire, of Patrick Fealey, a writer who is now homeless and living in Rhode Island. Fealey is a novelist and art critic who used to contribute to The Boston Globe. He’s been brought down by debilitating manic depression. Fealey tells his story in a straight-ahead way, bringing you into the details of his budget, his daily interactions with the police, and the way regular people pulling into the beach, where he parks many days, treat him. “They are so afraid. I know I look disheveled, but I don’t believe there’s anything wrong with me intellectually or spiritually. I know I could look better, but I just don’t see what the big deal is.”

The story isn’t designed, like others in this genre, to yank at your heartstrings. It’s just a cold, hard accounting of what it’s like to live with a dog in a car on Jif-and-Smucker’s-strawberry-jam sandwiches. The photography, by Philip Montgomery, is extraordinary.

[The Searching Smarter section below is sponsored by Elastic.]

My other favorite recent essay is rather different: it’s the Vatican’s “Note on the Relationship Between Artificial Intelligence and Human Intelligence.” AI is a tool created by human intelligence and the quest to understand it is, thus, a quest to understand humanity. That said, there are differences between performing tasks and thinking, and AI should be considered not “as an artificial form of human intelligence but as a product of it.” The authors describe the risk that AI could make the world less equal as well as the risk that it could be used to create false Gods.

The essay ends with a call for AI to be used for the betterment of all, which resonates whether one is Catholic, atheist, or of any belief:?

“Since a ‘person’s perfection is measured not by the information or knowledge they possess, but by the depth of their charity,’ how we incorporate AI ‘to include the least of our brothers and sisters, the vulnerable, and those most in need, will be the true measure of our humanity. The ‘wisdom of the heart’ can illuminate and guide the human-centered use of this technology to help promote the common good, care for our ‘common home,’ advance the search for the truth, foster integral human development, favor human solidarity and fraternity, and lead humanity to its ultimate goal: happiness and full communion with God.”

I also loved trying out Steven Johnson’s new form of interactive storytelling, in which AI is used to enhance the way we read and talk to his latest book of history. (Johnson has long been one of my favorite magazine writers and authors, and he recently made the jump into the tech industry, where he helped create Google’s remarkable NotebookLM.) I spoke at length on the questions of AI and journalism on the Grill Room podcast yesterday.?

Renee DiResta has written, in Noema Magazine, one of the best things I’ve read on how social media has fractured and splintered into ideological bubbles. Zeynep Tufekci has written smartly about how to think about DeepSeek, and you can’t go wrong listening to Jonathan Ross on this subject either. I also enjoyed reading about Ed Park’s discovery of an unpublished memoir he had written more than two decades before. Here’s Eric Lach on the strange Manhattan luxury condo building that was designed so poorly that it now sits unoccupied and tilting like a banana. And here’s the ever-marvelous Gary Shteyngart telling nearly the opposite story: his quest to find the perfect suit.

This article is sponsored by Elastic, the Search AI Company. Elastic makes AI possibilities a reality through the power of Search AI. With AI use cases evolving rapidly, there’s a lot to be excited about. Watch industry leaders share what’s most fascinating to them about generative AI models and their potential.

When I talk to people about AI right now, it’s usually not long before the subject turns to agents. These systems are models empowered to tackle complex problems out in the world that require multiple steps. Of course, bots have been trading securities (and sneakers, and restaurant reservations) for years. But an agent promises something more powerful: an AI that understands plain language requests and is empowered to act independently to achieve a goal—ideally one that you have given it.?

Say you’re shopping for a new car: Instead of driving out to spend all day haggling at a dealership, you could feed an AI your desired specs and price and let a large language model contact and negotiate with a range of sellers, who might have their own agents working too. When the salesman tells your AI he needs to go to the back in order to “talk to his manager,” the model will handle it with limitless patience while you go back to practicing guitar.

Now extrapolate that process to the untold number of negotiations and transactions that take place every day, and you can see why business leaders see agents as one of the next big things in AI. It’s still unclear exactly the shape agents will take, of course, but this white paper from the World Economic Forum provides a framework. Repetitive and tedious tasks will be taken on by agents; high-stakes situations where errors would be catastrophic will long need human help.

The potential of independent AI agents raises big questions about how an AI can (and should) act in the physical world. This piece from Jonathan Zittrain (whose name ChatGPT still won’t say) is a great introduction to the complexity of regulating agents. What happens, for example, if your car-selling agent lies or accidentally offers you the car for free? What if your agent gets hijacked or hacked? Zittrain also considers what happens when an agent is matched with an opposing agent, when there would be potential for a spiral of expensive and unproductive computing.?

The biggest risks come if agents get out of our control. This has been explained by one of the pioneers of the field, Yoshua Bengio. “The current state-of-the-art method seeking to evolve AIs into agents through reinforcement learning techniques involves creating systems that seek maximum positive rewards—thus opening up the possibility of the AI eventually being capable of overtaking the reward system,” he writes. To Bengio, we may need non-agentic AI systems to keep the agents in check.

Bengio and others are hard at work on this problem. They’ll need to move quickly because, whatever the risks, we are moving rapidly into a world in which AI agents are everywhere.

Cheers * N

I hope you enjoy this newsletter, which, thanks to your support, now has more than 475,000 subscribers.?Please continue to forward it to anyone else who might enjoy it. They can sign up here. And you can learn more about Elastic, this week's sponsor of the Searching Smarter section, here.