Sanmi Koyejo, an assistant professor of computer science at Stanford University, is a leading figure in the field of trustworthy AI research. His work focuses on the intersection of machine learning, scientific discovery, and the evaluation of AI systems. Koyejo's journey into AI began unexpectedly, as he initially pursued a career in electrical engineering, building electronics and working on control systems. However, during his graduate studies, he discovered the exciting world of machine learning and quickly shifted his focus. Today, his research group at Stanford explores three key areas: understanding AI systems, building trustworthy AI, and applying AI to real-world problems, particularly in science and healthcare.
One of Koyejo's areas of interest is astronomy, which he finds fascinating due to the unique challenges it presents for AI. Unlike other fields where AI excels, astronomy operates within a single universe, making it difficult to create large datasets for training machine learning models. Instead, astronomers rely on combining observations with physical understanding and scientific intuition. Koyejo's work in this area involves collaborating with researchers to enhance the efficiency of analysis methods while preserving the underlying physical principles.
A critical aspect of Koyejo's research is the distinction between passing a benchmark test and doing science. While benchmark tests are useful for comparing and tracking AI performance, Koyejo emphasizes that they should not be mistaken for scientific discovery. He argues that AI systems often agree with each other even when they are wrong, and this should not be interpreted as correctness. Instead, it highlights the need for rigorous evidence and standards, similar to those used by scientists.
Koyejo also stresses the importance of scientists' involvement in shaping AI tools. As AI becomes more integrated into scientific work, researchers should not be passive users but active participants in deciding the capabilities and limitations of these tools. Scientists possess a deep understanding of what constitutes evidence, the significance of mistakes, and the criteria for trustworthiness, which are essential in guiding AI development.
For students, Koyejo offers valuable advice. He encourages them to explore diverse fields and consider the impact they want to have. Astronomy, with its many open questions, provides exciting opportunities. However, he warns against chasing quick results and emphasizes the importance of deep learning, building context, and thoughtful tool usage. Koyejo's research aims to bridge the gap between impressions and headlines, moving towards evidence-based understanding.
In conclusion, Sanmi Koyejo's work in trustworthy AI research is a testament to the importance of careful evaluation and understanding in the field. His insights into the relationship between AI and science, as well as his advice for students, offer valuable perspectives on the future of AI and its role in advancing scientific discovery.
