Ongoing advances in Artificial Intelligence (AI) have spurred innovative approaches to nuclear engineering challenges. AI methods have been proposed for applications including reactor monitoring and control, core loading optimization, reduced-order transport, nuclear data evaluation, and detecting bias within computational results. While early results entice further exploration in some cases, the extent to which AI methods have the capacity to displace traditional numerical techniques is unclear, especially as AI methods come with their own unique challenges including explainability, uncertainty quantification, data availability, reproducibility, and potential vulnerability to adversarial reprogramming.

This panel discussion will bring together experts from AI and nuclear engineering to discuss the limitations of AI for nuclear applications. The panel will address questions such as:

  • Will AI methods find success in nuclear engineering? Which applications show the most promise, and where is it being misapplied?
  • Have attitudes towards AI been overly optimistic within the computational sciences in general?
  • Are nuclear applications uniquely ill-suited for AI methods due to safety constraints?

Panelists

  • Alessandro Fanfarillo (AMD)
  • Kelli Humbird (LLNL)
  • Mengnan Li (INL)
  • Vladimir Sobes (Univ. Tennessee, Knoxville)

Resources

To access session resources, you must be logged in and registered for the meeting.

Register NowLog In

Discussion

To join the conversation, you must be logged in and registered for the meeting.

Register NowLog In