Introduction to AI Safety

Catalog description

As we delegate more to artificial intelligence (AI) and integrate AI more in societal decision-making processes, we must find answers to how we can ensure AI systems are safe, follow ethical principles, and align with the creator's intent. Increasingly, many AI experts across academia and industry believe there is an urgent need for both technical and societal progress across AI alignment, ethics, and governance to understand and mitigate risks from increasingly capable AI systems and ensure that their contributions benefit society as a whole. Intro to AI Safety explores these questions in lectures with targeted readings, weekly quizzes, and group discussions. We are looking at the capabilities and limitations of current and future AI systems to understand why it is hard to ensure the reliability of existing AI systems. We will cover ongoing research efforts that tackle these questions, ranging from studies in reinforcement learning and computer vision to natural language processing. We will study work in interpretability, robustness, and governance of AI systems - to name a few. Basic knowledge about machine learning helps but is not required. View the full syllabus at http://tinyurl.com/42rb2sfv. Enrollment is by application only. Apply online at https://forms.gle/v8msM8nJ5FgeEHx1A by 9:00 PM PDT on Saturday, March 16, 2024.