June 10, 2025
By Louis DiPietro
Ask a large language model (LLM) – a generative artificial intelligence (AI) technology like ChatGPT – what a Muslim man looks like, and it may churn out photos of terrorists. Punch in a request for photos of New Delhi, India’s capital city with rich cultural diversity, and you might get slides of fiery slums. “People with disabilities” might yield images of people with glitchy, dysmorphic faces in wheelchairs – a tell-tale sign that AI is producing images without enough data to draw from.
To leading AI scholars and researchers at the “Thought Summit: LLMs and Society,” held May 19-21 in the ILR Conference Center, such errors aren’t harmless; they reveal a fundamental design problem and systemic issues, said Aditya Vashistha, assistant professor of information science in the Cornell Ann S. Bowers College of Computing and Information Science and the summit’s lead faculty organizer. AI tools are being designed by and for Western culture, which represents just 15% of the world’s population, but used around the globe, he said.
“One reason why this is problematic is because a large proportion of the world's population live in non-Western settings where their values, lived experiences, and needs are systematically deprioritized in current AI advances,” Vashistha told the crowd of roughly 40 scholars, many from Cornell. “This is even more concerning given the large infusion of AI technologies into high-stakes settings, in education, healthcare, and law, among others.”
Hosted by Cornell’s Global AI Initiative as part of the Thought Summit program organized by the Cornell Center for Data Science for Enterprise & Society, Thought Summit: LLMs and Society is the second of several summits slated this fiscal year at Cornell. The opening event brought together interdisciplinary AI scholars, researchers, and policymakers from industry, academia, and think tanks to consider how best to integrate LLMs into high-consequence areas. OpenAI, Meta, Google Research, Microsoft Research, and Infosys were some of the companies represented at the summit, which was also sponsored by the Cornell Center for Social Sciences. Data & Society, the Center for Democracy & Technology, and the Chilean National Center for AI (CENAI) were among the policy and AI think tanks in attendance.
Over the summit’s three days, researchers considered the role of LLMs in areas like health and medicine, environment and sustainability, and education; brainstormed how LLMs affect marginalized communities; and mulled tensions and lessons for future LLM design. From these discussions, summit participants will craft a white paper that outlines pathways toward more inclusive and accountable AI that works for a global audience.
“There is this opportunity for great change through all that AI can bring about, including new kinds of innovation, scientific discovery, creative tools and applications,” said Provost Kavita Bala, an AI researcher herself, in her opening remarks. “The work you’re doing – to think about the impact of AI from design, development, implementation, all the way to validation – is very critical and urgent now, exactly because all of us know that there is this global impact of AI. AI has to serve the whole world.”
In his remarks, David Shmoys, the Laibe/Acheson Professor of Business Management and Leadership Studies in Cornell Engineering and director of the Center for Data Science for Enterprise & Society, said his hope for this year’s Thought Summits is to spark ongoing collaboration among experts in industry, academia, government, and NGOs around pivotal topics like the societal impact of large language models.
“This summit is our opportunity to reflect, to connect across sectors and geographies and imagine how together we might shape a more responsible, equitable, and pluralistic future for AI,” Vashistha said.
The following faculty served as session leads during sessions:
-
Introductions -– Malte Jung, associate professor of information science in Cornell Bowers and the Nancy H. ’62 and Philip M. ’62 Young Sesquicentennial Faculty Fellow, and Duarte Santo, former lecturer in the College of Agriculture and Life Sciences (CALS).
-
“Metrics of Success” – Matthew Wilkens, associate professor of information science in Cornell Bowers, and Natalie Bazarova, associate vice provost in the Office of the Vice President for Research and Innovation and professor of communication in Cornell CALS.
-
“High-Stakes Domains” – Rene Kizilcec, associate professor of information science in Cornell Bowers, and Wilkens.
-
“Responsible LLMs & Issues” – Allison Koenecke, assistant professor of information science in Cornell Bowers, and Marten van Schijndel, assistant professor of linguistics in the College of Arts and Sciences.
-
“High-stakes Populations” – Cristobal Cheyre Forestier, assistant professor of information science in Cornell Bowers, and Maria Goula, professor of landscape architecture in Cornell CALS.
The next summit, called “Everyday AI and Mental Health – Navigating a Tipping Point,” will take place June 16-20. Attendance is by invitation only. Cornell faculty are invited to submit proposals by June 16 for a third summit on data science and AI.
The Cornell Global AI Initiative was launched in 2024 with support from Global Cornell as part of the Global Grand Challenge.
Louis DiPietro is a writer for the Cornell Ann S. Bowers College of Computing and Information Science.