New TCRG research explores how AI can be used to combat tobacco-related misinformation

at

A new paper from the Tobacco Control Research Group explores how recent developments in AI can be leveraged to combat misinformation in the tobacco domain. The proliferation of tobacco-related misinformation poses significant public health risks. Manual fact-checking of tobacco-related claims is resource-intensive and often cannot match the pace of misinformation spread.

The researchers set out to develop a proof-of-concept multi-agent AI approach to evaluating the credibility of tobacco-related claims in near real-time. The approach consists of three sequential agents: a Content Analyzer agent, to parse and categorise claims; a Scientific Verifier agent, to search trusted sources of information (WHO, CDC, Cochrane, PubMed) in a consistent way and return structured results related to the parsed and categorised claim; and an Evidence Assessor agent to assess the credibility of the claim against these structured results, with basic output of the credibility assessment on a five-point scale from Highly Unlikely to Highly Likely.

Initial results on a test set of 20 tobacco-related claims showed excellent agreement between the credibility assessments of the multi-agent AI system and those of tobacco control experts, with the AI system outputting results in under 10 seconds.

This work builds on other work from the team on leveraging Large Language Models in the tobacco control domain.

Co-author John Mehegan explains:

“A tool to quickly and accurately assess tobacco-related claims against widely trusted sources of information is potentially very valuable to public health advocates, policy makers and the general public. This proof-of-concept study shows such a tool is possible.”

Read the paper:

And the earlier research on which it expands:

Print Friendly, PDF & Email