‘AI Snake Oil’ author sees chatbots evolving into a ‘truth oracle’—and journalism heading somewhere it hasn’t been in 200 years

Arvind Narayanan thinks the noisy, sloppy AI era won’t last. The Princeton computer scientist told a journalism audience that chatbots are getting more trustworthy, the obvious tells are fading, and the day may come when people use them the way they once used a reference book: as a truth oracle. A Pew survey this year found 10% of Americans already do that.

At Northwestern’s Computation + Journalism Symposium, Narayanan asked the room to think five or 10 years ahead, not just about today’s mistakes. He said the old picture of chatbots as next-word guessers is already outdated. They now work more like “neurosymbolic systems,” with search, retrieval and code tools bolted on, which is why some of the embarrassing errors from a year ago are less common.

That matters because if machines become the default place for factual lookups, news loses one of its core jobs. Narayanan said “virtually all” questions about who’s running for office are headed there. What remains, in his view, is the work machines can’t settle: contested questions, narrative, storytelling, and the awkward human idea he calls positionality.

His bigger argument was about journalism itself. The modern newsroom, he said, is a bundle of reporting, analysis, fact-checking and distribution that only looks natural because history made it so. Earlier eras had penny papers, political patronage, wartime propaganda, hired reporters that were once an insult, and even the invention of interviews. Today, he said, those pieces are splitting apart again. Trade reporting, local civic work, investigations and analysis are already migrating to different kinds of outlets and funders.

The sharpest point was money. Narayanan said publishers need chatbot distribution more than AI firms need any one outlet, which leaves licensing deals weak. He wants journalists to push for democratic input into how AI platforms are tuned, and maybe even a tax on AI and social media companies. Otherwise, the industry may find out that the last 200 years were an unusually convenient arrangement.