Researchers at the Courant Institute of Mathematical Sciences developed an artificial intelligence assistant aimed at curbing health disinformation by answering questions related to diets with a comprehensive summary of nutrition research and studies online.
The model, called DietNerd, launched last month and can answer users’ questions using a large language model — a type of artificial intelligence trained to recognize patterns in text — searches through biomedical research databases such as PubMed for related information. The site then compiles peer-reviewed research and presents easy-to-understand advice and a list of the research articles it referenced.
Courant professor Dennis Shasha, University of Illinois Urbana-Champaign student Zubair Yacub and Courant graduate student Shela Wu led the project. In an interview with WSN, Wu said she created the model to combat false or misconstrued information about nutrition online.
“There’s a lot of marketing tactics out there that might try to influence people to adopt a certain diet or lifestyle,” Wu said. “That’s why we wanted to leverage the power of large language models in this space because it’s very quick to be able to summarize things.”
Wu said she and Shasha have a shared interest in leveraging AI models responsibly to make research more accessible. She added that their goal was to design a system that prioritizes safety and bias mitigation when providing answers, implementing more expansive search methodologies to ensure accurate and unbiased results. Wu said that for highly contended topics among researchers, the system prioritizes exposing users to a diverse range of viewpoints that ensure a “diverse pool of perspectives.”
“The results are one that you might not get from another research assistant out there because of the unique article search strategy that we’ve implemented,” Wu said. “It helps open that door towards information that was previously inaccessible to the everyday person.”
The researchers are now looking to adjust DietNerd’s framework to help people understand other complex subjects. In a statement to WSN, Shasha said other AI assistants are currently in development, including ContractNerd, InvestorNerd and WirelessNerd.
“This is just opening up another avenue towards another resource and knowledge base that people might not always default to in the beginning,” Wu said.
Contact Liyana Illyas at [email protected].