Chatbots: A ‘truth oracle’?

Princeton computer scientist Arvind Narayanan says artificial intelligence chatbots could become a default place for factual answers within the next decade. However, he does not consider that outcome inevitable.

Speaking at Northwestern University’s Computation + Journalism Symposium, Narayanan argued that chatbots are becoming more reliable as companies combine language models with tools that search the web, retrieve documents and analyze data. Moreover, Fortune, which attended the keynote, reported that Narayanan expects many common errors to become less frequent as those systems improve.

His elementary school-age children, Narayanan said, may never experience AI hallucinations the way adults do today. He compared the shift to Wikipedia’s evolution from a distrusted reference into one many people now consider authoritative.

He said chatbots could eventually become what people treat as a “truth oracle.” Fortune cited a Pew survey this year that found 10% of Americans already use chatbots that way. Still, Narayanan stressed that the shift is neither inevitable nor necessarily desirable.

Are you an AI Agent? Read more here >

Narayanan directs Princeton’s Center for Information Technology Policy and co-wrote AI Snake Oil. He acknowledged that he has never worked in a newsroom. Therefore, he framed his talk as a forecast rather than a prescription for the industry.

His argument goes beyond AI. Narayanan said the modern newsroom bundles reporting, analysis, fact-checking and distribution because of historical and economic circumstances, not because those functions must remain together.

That bundle has already been coming apart, he argued. Specialized outlets such as Politico, Punchbowl News and Stat have taken on trade reporting. Philanthropy increasingly supports civic, local and investigative journalism. Independent journalists and creators now provide more analysis outside traditional news organizations.

AI is accelerating that process, Narayanan said, by lowering the cost of producing some forms of journalism. Now, small teams can create work that once required larger organizations. Meanwhile, social media has already reshaped how publishers distribute it.

If chatbots become the default place to settle factual questions, newsrooms could lose another basic part of that bundle. Narayanan predicted that “virtually all” straightforward factual questions, such as who is running for political office, are headed toward chatbots.

That could leave greater emphasis on work machines cannot simply settle: contested questions, narrative reporting, storytelling and what Narayanan called “positionality.” This means being explicit about the perspective behind a piece of journalism.

He also identified a financial problem. Narayanan called the power imbalance between AI companies and publishers the “real crisis,” arguing that publishers need chatbot distribution more than an AI company needs any single news organization. Consequently, he said, that imbalance can push licensing deals below the value journalism creates.

His proposed response includes giving journalists more influence over how AI companies shape their algorithms. Furthermore, he compared the idea to Meta’s Oversight Board and said journalists need a “seat at the table.” He also called for a broad political coalition that includes independent creators and legacy outlets. This could potentially support a tax on AI and social media companies.

One possibility remained mostly unmentioned at the symposium, Fortune reported: could organized journalism itself be a product of the past two centuries?

Narayanan’s forecast is not that AI alone will dismantle the newsroom. Instead, it is that AI may hasten a process already underway. This change may force journalism to reconsider which parts of the bundle still belong together.