Workshop Addresses AI Ethics

On May 8, the National Science Foundation (NSF) Artificial Intelligence Planning Institute for Data-Driven Discovery in Physics and Carnegie Mellon University's Block Center for Technology and Society held the Responsible AI in the Natural Sciences workshop on CMU's campus and online.

"The use of AI in scientific research has exploded over the last three to five years," said Rupert Croft, professor of physics and one of the workshop's organizers. "It seems like almost every scientific paper has some sort of AI component to it, but people are often worried about how reliable it is."

Croft said the workshop brought together experts who focus on different facets of problems that arise from AI as well as over 600 participants, who attended in person and over Zoom. Some are scientists or researchers who want to specifically discuss how AI should be used in scientific research and publications. Others, like speaker Hoda Heidari, analyze the ethical use of AI in general.

"AI as a technology is still in infancy despite all the excitement around it," said Heidari, an assistant professor of computer science and a member of the Block Center. "We have made some progress in recent years in developing the necessary frameworks, tools and processes to promote the responsible use of this technology, but we have a long way to go. Especially with the rise of generative AI, we have entered relatively uncharted territory, and the biggest lesson we should carry forward is the need for strong governance and accountability mechanisms."

At the close of the workshop, the speakers addressed audience questions in a panel session. They explained the pluses and minuses of AI models and reassured that AI would not replace vital work done by researchers. Most saw the future of AI as a partnership between humans and technology.

"We and our machines, we should work together either as an assistive device or in a collaborative fashion," said Ahmed Tafti, director of the University of Pittsburgh Health and Explainable (HexAI) Research Laboratory. "People five years ago, six years ago, they believed that AI would replace radiologists, but after all these years with advances in AI components and computational methods, we can look where we are at. We can see we cannot replace radiologists with AI. AI and people can work collaboratively toward a better world."

To analyze whether humans could tell the difference between human-generated content and AI-generated content, the workshop held an AI vs. Human Contest. Participants were encouraged to send in both human-generated and AI-generated scientific data then judge for themselves to see if they could determine which were which. The winning participant was 72% correct, which showed how difficult it could be for people to judge AI and human content with the rise of generative AI models.

Though the daylong workshop could not address all the potential outcomes of AI technology in the sciences, the proceedings offered an optimistic view of a future where scientists use AI technologies as valuable tools to further their research.

"I hope AI will enable us to do things that we hadn't imagined were actually possible or we haven't got the imagination to think about," Croft said. "I hope it's something useful."

The workshop was funded through the NSF AI Planning Institute and the Block Center. The NSF AI Planning Institute is sponsored through a National Science Foundation grant. The Block Center was founded thanks to Carnegie Mellon alumnus Keith Block and his spouse, Suzanne Kelley.