Today’s Friday edition was written by Contributing Editor Isaac Wood.
Most days, our technological landscape feels unshakeably miraculous. You can save hundreds of thousands of documents in organized folders and send messages to a friend on the other side of the world in mere seconds; the internet crams all recorded knowledge into your pocket; personal agents sit at your beck and call to schedule doctor appointments or retrieve any information you might need; and AI software can crunch more data in one minute than you could in a lifetime.
We’ve made it.
Except this week things felt precarious.
Some combination of testing, patching and possibly hacking compromised staple services like Microsoft Outlook, Apple Pay, Chase online banking, and Zoom. As some have noted, these outages expose vulnerabilities in a set of systems we’ve come to trust and rely on. But there’s more at play than software and hardware weaknesses. The AI rollout depends on some load-bearing presuppositions that I see as fundamentally flawed, and the last few months of headlines have revealed cracks in the foundation of the AI industry’s core message.
The HuggingFace hack this July was a story that hit all the right notes to stoke AI fears — rogue agents disobey their creators and launch a cyberattack. Two months later, President Trump hosted a gathering of AI leaders and cabinet members called the “Super Intelligence (SI) Summit,” and several tech leaders joined the president in signing the “White House Accord on Super Intelligence.” Then, last week, Nvidia CEO Jensen Huang admitted to having an affair with a Chinese businesswoman with alleged connections to the Chinese government. It’s the kind of story we’d get by leaving Shakespeare and Stanley Kubrick in a room together.
Artificial intelligence has floated in and out of my conversations and worries since 2022, but these headlines forced me to wrestle with the costs of a technology like AI and whether I’m willing to accept those trade-offs.
So I’ve been sitting with some questions. While they don’t have clean answers, I think the responses we’re hearing from AI leaders rest on flawed assumptions.