[1] State of AI: Applications. Safety requirements. Big ideas.
Framing of the problem, identification of potential big ideas, and setting technical context for more in-depth discussion in the subsequent sessions. Examples of topics that could be included: (1) Roles and capabilities of emerging AI-based systems, including human-AI interaction and interacting AI systems, (2) How AI capabilities are manifested now and how they might be manifested in the future, building on statistical machine learning and data-intensive approaches, (3) The nature and significance of the challenges of safety and control, from a technical and mission perspective, (4) How we might do better in addressing these challenges, (5) Other technical considerations, such as potential trade-offs among capability, adaptability, resiliency, and confidence.
[2] Approaches involving AI algorithms, including intrinsic safety, explanations, evaluations.
Identification and consideration of potential intrinsic approaches to AI safety, for example based on creation of explanations, safety features incorporated into learning algorithms, analytic and predictive models, and so on. This could include consideration of trade-offs among capability, adaptability, resiliency, and confidence.
[3] Approaches involving mathematical models and reasoning.
Identification and consideration of mathematics-based approaches to modeling and reasoning about safety attributes for systems that include AI and learning functionality. This includes how safety properties might be expressed and analyzed, architecture and design models, engineering practices and tools, evaluation practices, and so on. It could also include consideration of trade-offs among capability, adaptability, resiliency, and confidence.
[4] Approaches based on systems and safety engineering.
Identification and consideration of systems-level approaches to assuring safety of systems that embed AI and learning functionality. This includes how safety properties might be expressed and analyzed, architecture and design concepts, engineering practices, evaluation practices, and so on. It could also include consideration of trade-offs among capability, adaptability, resiliency, and confidence.
