Search for a command to run...
Abstract As automation advances toward fully autonomous operations, the need for the humans involved in the process to receive sufficient information becomes increasingly important. They must have a clear overview of what decisions the systems are making and the reasoning behind them. The required information varies depending on the user's role, ranging from operating users to an auditing personnel member. This paper analyzes the key risks, requirements and constraints linked to the enabling human oversight in complex automation systems for various end-users involved. Artificial Intelligence (AI) methods are becoming increasingly prevalent in complex automation systems, including drilling automation. Drilling is an intricate process where unexpected issues can arise at any time, making it essential to provide end users with the right level of information at all times. Any drilling automation system, along with its AI components and methods, must contend with challenges such as poor observability and limited predictability inherent in drilling operations. Additionally, these systems must be capable of effectively managing high-risk scenarios that could lead to highly hazardous situations and potentially significant damage. Ensuring that humans operators receive sufficient information is crucial for addressing such situations safely. This paper illustrates these risks, constraints and requirements through two scenarios. The first examines autonomous decision-making during drilling, while the second focuses on a managed pressure drilling operation. With the current advancements from automated to autonomous drilling systems, it increases also the necessity to ensure that relevant information is continuously provided to the human operators involved in the process. In drilling operations, multiple vendors are involved, making human oversight challenging due to the decentralized nature of information. The distributed structure of such multi-agent automation systems adds further complexity to the process. Given a certain level of interoperability in these complex automation systems, some information may be directly accessed. However, this may not be the case for all situations and other information may need to be inferred or estimated, adding a yet new level of complexity. At the same time, excessive information overload for end users should be avoided. The type of information required depends on the user's role—operational users may need insights into data uncertainties and model estimations, while monitoring users may require an understanding of dependencies between the different automation agents. Taking a holistic approach to enabling human oversight in complex drilling automation systems powered by AI can help ensure their safe adoption in practice. While this paper focuses on drilling related scenarios, the approach outlined is applicable to a wider range of AI-driven automation systems, including well planning, plug and abandonment (P&A), and intervention operations.