Melissa Fleming and Jeremy Hymans say AI chatbots could reshape the battle over misinformation

At the SDG Media Zone hosted at the United Nations, Melissa Fleming, UN Under-Secretary-General for Global Communications, and Jeremy Hymans, founder of Purpose, discussed risks and opportunities posed by large language models (LLMs) and conversational AI for the information ecosystem.

Jeremy Hymans said the next phase of the information fight will center on what people ask and receive from their LLMs, which he described as people's emerging "digital significant others." "What the LLM says to you about that will be incredibly important," Hymans said, and he urged advocates to influence how AI developers set boundaries so models report reliably sourced information rather than "scraping up the junk and feeding it back to people."

The Nut Graf: Both speakers contrasted social media dynamics, which often reward sensational or polarizing content, with the one-on-one character of many LLM interactions. They argued that LLMs could either worsen misinformation if trained on poor sources or help depolarize discourse if developers prioritized trustworthy training data and guardrails.

Fleming illustrated the risk of AI hallucinations with a recurring informal experiment: she said she repeatedly asked ChatGPT who her husband is and the model returned different invented responses over time. "Each month it told me I had a different husband," she said, using the example to underline how conversational AI can invent plausible but false statements.

Both speakers said AI presents an advocacy priority. Hymans noted that while changing the policies of established platforms (Meta, X) is difficult, shaping the next widely used platforms and the way AI is built is still feasible. He suggested that if LLMs are grounded in reliable sources they may "depolarize and take the heat out of the information environment a little bit," but he cautioned that strong boundaries and truthful training data are required.

The exchange concluded with Fleming noting that the UN’s global risk report singled out misinformation and disinformation as the top risk people felt least prepared to address. The speakers framed AI as the next major front in that struggle and urged broader engagement by governments, platforms and civil-society actors to shape AI’s role in public information.

The session closed with the speakers calling for advocacy, improved platform and AI governance, and the continued use of people-powered communications strategies to complement technical solutions.