What It Takes to Trust AI in Real-World Seismic Applications

“Deep learning is ubiquitous in data processing. The question is whether we have the courage to change the way we work.” Yangkang Chen discusses how deep learning has moved from experimentation to production in seismic processing and earthquake monitoring. Drawing on a decade-long effort to build an operational AI-driven monitoring system, he explains why tasks like first-arrival picking, velocity analysis, denoising, and reconstruction are especially well suited for deep learning. Yangkang emphasizes that success depends not just on algorithms, but on benchmarks, stability, teamwork, and trust. He also highlights how open and reproducible research lowers barriers for adoption and helps geophysicists apply AI confidently in real workflows. KEY TAKEAWAYS > Deep learning excels at repetitive, label-intensive seismic tasks that are slow and inconsistent using traditional methods. > Operational AI requires trust, built through benchmarks, validation, and a clear understanding of model behavior. > Open and reproducible workflows accelerate adoption, collaboration, and innovation across the geophysics community. Register for his course, Deep learning for revolutionizing seismic data processing, on March 24-25, 2026 at https://seg.org/shop/product/?id+=product&id=5c3b6821-549d-f011-b41b-7c1e521913ef. ABOUT SEISMIC SOUNDOFF Seismic Soundoff showcases conversations addressing the challenges of energy, water, and climate. Produced by the Society of Exploration Geophysicists (SEG) and hosted by Andrew Geary of 51 features, these episodes celebrate and inspire the geophysicists of today and tomorrow. Three new episodes monthly. See the full archive at https://seg.org/resources/podcast/.