How to make AI trustworthy?
Description
"Evan Miyazono, founder of Atlas Computing, a nonprofit, tackles the urgent issue of AI capabilities outpacing human review. In this insightful talk, Miyazono explores the separation of AI capability concerns from the fear of superintelligent AI or imbalanced power dynamics among AI users. He explains the limitations of current AI review mechanisms and proposes a future where review processes are robust and scalable.
Miyazono emphasizes the importance of moving beyond mere alignment and developing systems where human oversight remains integral. He highlights the need for specifications that define success and failure criteria, enabling AI systems to meet these benchmarks reliably. By examining various domains, such as formal verification and small molecule biochemistry, Miyazono illustrates practical steps toward achieving these goals."
Miyazono emphasizes the importance of moving beyond mere alignment and developing systems where human oversight remains integral. He highlights the need for specifications that define success and failure criteria, enabling AI systems to meet these benchmarks reliably. By examining various domains, such as formal verification and small molecule biochemistry, Miyazono illustrates practical steps toward achieving these goals."