Applying AI in Smart Refineries – Enhancing the Impact of AI on Downstream Operations
This session examined how industrial organisations are moving beyond single model approaches to AI, instead combining machine learning, physics based models, and rule driven logic with deep domain expertise. Panellists consistently emphasised that intelligence only becomes useful when it reflects how assets operate in the real world, making hybrid approaches essential in complex, safety critical environments.
Human expertise emerged as a defining factor in building trust. While AI systems can process vast volumes of operational data and surface patterns at speed, accountability for decisions remains firmly with people. Speakers stressed that keeping operators, engineers, and subject matter experts in the loop is critical, not only for safety and compliance, but also for ensuring AI recommendations are interpretable, actionable, and aligned with operational reality.
The discussion also addressed persistent challenges around data quality, interoperability, and context. Rather than striving for perfect data across every system, panellists advocated focusing on data fitness for purpose, supported by strong governance, clear ownership, and semantic layers that capture industrial language and asset hierarchies. This contextual grounding was seen as particularly important for generative and agent based systems, where accuracy and relevance are non negotiable.
Large language models were positioned as a powerful interface between people and complex data landscapes, enabling natural language access to structured and unstructured information at scale. However, speakers cautioned that these capabilities must be deployed with clear guardrails and an understanding of existing workflows to avoid repeating the high failure rates of poorly scoped AI initiatives. Success, the panel concluded, depends on grounding AI in real operational problems, designing around how work is actually done, and measuring value through better decisions rather than technology adoption alone.
Speakers
- Thep Smith, Opportunity Shaping Lead for Operation and Maintenance, Chevron
- Shan Pin Koh, Senior Architect, Dow Chemicals
- Ahmed Farag, Business Development, Commercial, Gecko Robotics
- Gulshan Oshan, Director, Consult Partner, Utilities & AI, Kyndryl
- Adam Williams, Senior Manager, Engineering Design Technology at Motiva
- Diti Sood, Co-Founder & CEO, Bolo-AI
Moderator
