Panel Session | Managing the Risks of Ungoverned AI in the Subsurface
SESSION OVERVIEW
As subsurface workflows evolve from simple automated classifiers to complex foundation and generative models, AI is rapidly moving from an assisting tool to sitting directly between raw seismic data and multi-billion-dollar operational decisions. However, in resource industries where decisions carry immense, irreversible consequences, capability has outrun governance.
With major regulatory deadlines taking effect globally—such as the EU AI Act enforcement phases starting in 2026—implementing robust governance is no longer optional. This panel explores how operators, commercial developers, academics, and geoscientists can navigate the fine line between risk and reward. We will dive into how auditable, probabilistic generative AI can reduce human bias and subjectivity, how evolving data sovereignty requirements shape cloud versus local model deployments, and how university programs must adapt to prepare the next generation of geoscientists for an AI-integrated subsurface.
KEY DISCUSSION THEMES
Operator Risk vs. Reward: How E&P companies evaluate the business case for subsurface AI, balancing operational efficiency against the financial, safety, and reputational risks of improper decisions.
Auditability, Probabilistics, and the Death of Human Bias: Leveraging generative AI to produce transparent, statistically reasoned outcomes with verifiable audit trails—reducing over-reliance on individual seniority, urgency, or subjective interpretation bias.
Evolving Skillsets & Geoscience Education: Preparing future geoscientists to master advanced statistical models, physics-informed AI, and explainable tools—and how academic institutions are restructuring their curricula to meet this demand.
Commercial AI Development Without Guardrails: How technology developers manage risk, embed accountability, and establish safety protocols when creating commercial subsurface AI tools without strict regulatory boundaries.
Data Sovereignty, Firewalls, and the Infrastructure Arms Race: Navigating strict national data residence requirements (e.g., in jurisdictions like Indonesia) by evaluating private, firewall-confined models against public LLMs/foundation models—and avoiding vendor lock-in amid the global data center arms race.
Grounding AI in Physics and Geology: How core geoscientific principles—such as uncertainty quantification, explainable AI (XAI), and physics-informed machine learning (PINNs)—natively fulfill regulatory requirements for transparent, accountable AI systems
.
