Princeton computer scientist Arvind Narayanan says AI is moving toward a future where chatbots feel more trustworthy and may remove the tells that now reveal AI slop. He described a coming system in which more accurate chatbots could become what many people treat as a 'truth oracle.' Pew’s survey this year found that 10% of Americans are already doing this.
Narayanan delivered those comments at the Computation + Journalism Symposium at Northwestern’s McCormick Foundation Center. His keynote was titled 'AI and Journalism: Skating Where the Puck is Going.' He urged the audience to prepare for AI not as it is today, but as it will be five or 10 years from now.
He said his view comes from distinguishing between engineering problems that can be fixed and limits that are inherent. He also noted that he has never worked in a newsroom. 'It’s possible I don’t know what I’m talking about,' he cautioned, before laying out his case.
Narayanan split his argument into two parts. First, if chatbots become the default place to settle factual questions, then the newsroom loses one more piece of its bundle, and possibly the most basic one. He said 'virtually all' questions about who is running for political office are headed there. What remains, he argued, is work a machine cannot settle: contested questions, narrative, storytelling and being explicit about what he called 'positionality.'
He also said today’s chatbots are no longer simple next-word predictors. In his words, they are 'neurosymbolic systems' made up of a language model plus tools that search the web, retrieve documents and write code to analyze data. He said errors that were common a year ago, including miscounting the letters in 'strawberry' or Google’s AI Overviews suggesting glue on pizza, now rarely occur. He called the remaining problems 'gradually getting solved.'
Narayanan said his elementary-school-age children may never experience hallucination the way adults do, much as Wikipedia went from an untrusted reference to a 'much more authoritative' source. He added, 'I’m not saying it’s inevitable. I’m not saying it’s a good thing.' He estimated the wider shift at about a decade out.
He also said agenda-setting by the media has backfired. He called The Washington Post’s role in Watergate 'an unalloyed good,' but cited research from 'many communication scholars and economists' showing that similar efforts in the Trump era have fed polarization and hastened the collapse of trust.
For newsrooms, the market warning is clear: if machines absorb factual lookup, human outlets will have to justify their value through judgment and trust. Capital rewards clear rules, and audiences are already voting with their attention. The old bundle of reporting, fact-checking and distribution is no longer guaranteed to stay together.
The broader lesson is that free enterprise adapts faster than any media institution. If AI can answer routine questions at lower cost, the burden shifts to newsrooms to prove why their work deserves a premium. That is not a technology story alone. It is a story about leverage, credibility and which institutions still earn the public’s trust.
