Industrial AI Advances Require Focus on Safety and Governance

Industrial AI is entering a new phase, moving beyond predictive analytics to automate more complex tasks through advances in foundation models, physical AI, and agentic AI. Unlike purely digital AI, industrial AI can directly interact with physical systems, necessitating a strong focus on safety, reliability, and critical infrastructure. Arti Garg, chief technologist at AVEVA, highlights the challenge of leveraging these new capabilities while maintaining safe and reliable operations. The increasing capability of newer AI systems also makes them harder to predict and explain, amplifying the need for responsible deployment strategies. A key foundation for this transition is data integration. Industrial systems generate vast amounts of information across telemetry, service logs, and engineering documents. Newer AI technologies can connect and correlate this disparate data more rapidly, providing operators with real-time support for problem diagnosis. Furthermore, AI-powered robots could enhance safety by gathering information in hazardous environments, reducing the need for human entry. However, increased autonomy demands new governance approaches. AVEVA's framework for responsible AI prioritizes security, efficiency, and human safety and oversight. Garg advocates for AI to augment human decision-making rather than replace it entirely, with clearly defined guardrails specifying the boundaries of automated actions and the necessity of human supervision in critical decision loops. Sustainability is also a critical consideration. AI can play a role in managing complex power systems, especially with the growing integration of renewable energy sources. Concurrently, organizations must develop better methods to assess and understand the environmental footprint of AI technologies themselves. Garg is actively involved in an IEEE working group dedicated to developing standards and best practices for this evolving field, aiming to build a safer path toward autonomous industrial AI. The development of these advanced AI systems is occurring within a landscape where the potential impact on industrial operations is significant, requiring a proactive approach to risk management and ethical considerations. The integration of AI into physical industrial processes represents a paradigm shift, moving from observation and prediction to active intervention and control. This necessitates a robust understanding of potential failure modes and the implementation of fail-safe mechanisms. The complexity of these systems means that traditional oversight methods may not be sufficient, prompting the need for novel governance structures that can adapt to the dynamic nature of AI-driven operations. The emphasis on human-in-the-loop systems is a critical component of this strategy, ensuring that human judgment remains a central element in high-stakes decision-making processes. The ongoing work within organizations like AVEVA and standards bodies like IEEE underscores the industry's commitment to addressing these challenges proactively, aiming to unlock the benefits of industrial AI while mitigating its inherent risks.