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 complex question of trust in AI systems. In this interview, we delve into his journey, research interests, and insights on the role of AI in astronomy.
A Serendipitous Path to AI
Koyejo's journey into AI was not a straightforward one. Initially, he envisioned a career in electrical engineering, building electronics and working on control systems. However, during his graduate studies, he stumbled upon machine learning, and it quickly became his main focus. He found the application of machine learning tools to solve complex problems in cognitive radio systems particularly exciting, leading him to abandon other areas of study.
The Exciting Intersection of AI and Astronomy
Koyejo's interest in astronomy emerged as a natural extension of his work. He was drawn to the unique challenges and opportunities presented by this field. Astronomy, unlike many other domains where AI excels, operates with limited data. Instead of vast datasets, astronomers deal with the vastness of the universe itself. This requires a different approach, combining observations with physical understanding and scientific intuition.
Koyejo's research group is exploring how AI can enhance astronomy by improving the efficiency of analysis methods while preserving the underlying physical principles. He emphasizes that the goal is not just to speed up processes but to gain a deeper understanding of the universe.
Beyond Benchmarks: The Science of AI
One of Koyejo's key contributions is his emphasis on the distinction between passing benchmarks and doing science. AI researchers often evaluate systems using standardized tests, but Koyejo argues that this should not be the sole measure of success. He believes that AI should be evaluated based on its ability to work in real-world, messy situations, not just controlled environments.
A surprising finding from his research is that AI systems often agree on incorrect answers. This raises questions about the interpretation of agreement in AI. Koyejo suggests that scientists should apply the same standards of evidence to AI as they do to their own work, ensuring that AI systems are held to the same rigorous criteria.
AI and the Future of Scientific Discovery
Koyejo also discusses the evolving nature of scientific work in the age of AI. With the ease of producing research papers and the pressure on review systems, there is a growing need for scientists to actively shape the development of AI tools. He believes that scientists should not be passive users but should contribute to the design and implementation of these tools, considering what counts as evidence and what makes a result trustworthy.
Mentorship and the Value of Deep Learning
In the context of AI, Koyejo highlights the importance of mentorship. With the rapid advancements in technology, producing work quickly is easier, but understanding it is crucial. He encourages students to explore diverse fields and find their impact points, especially in fields like astronomy, where many questions remain unanswered. However, he also advises against chasing quick results, emphasizing the need for deep learning and building context.
Looking Beyond the Hype
Koyejo's research aims to bridge the gap between the excitement of AI and the need for careful evaluation. He believes that scientists should ask better questions and move beyond impressions and headlines to evidence and understanding. This aligns with the goals of astronomers, who strive for a deeper comprehension of the universe.
Conclusion: The Future of AI in Astronomy
Sanmi Koyejo's work is a testament to the potential of AI in advancing scientific discovery, particularly in astronomy. His research not only pushes the boundaries of AI but also emphasizes the importance of trust, evaluation, and collaboration between AI and scientists. As AI continues to evolve, Koyejo's insights will undoubtedly shape the future of this exciting field.
