Artificial intelligence and autonomous functions are playing an increasingly important role in drilling operations. As systems become more and more capable of analysing situations and making their own decisions, it also becomes more important to understand how people can maintain an overview, retain control and intervene when something unexpected happens.

This is the subject of a new research article from NORCE.

“When we say ‘autonomous’, we mean that the system can make decisions on its own even in complex situations, not just follow predefined rules,” explains Rodica Mihai, senior researcher in the Energy Modelling and Automation research group at NORCE Energy.

In the article “A technical perspective to human oversight in complex drilling automation systems based on artificial intelligence methods”, she and her co-authors examine how human oversight can be ensured in complex drilling automation systems.

“Safety in autonomous drilling operations cannot be reduced to individual functions or isolated technical barriers,” she emphasises.

The need for human oversight and control

A greater degree of autonomy does not mean that humans will become irrelevant. On the contrary, the role often shifts from active management to monitoring, assessment and intervention as and when necessary. As a result, human oversight is becoming increasingly important.

These challenges are particularly evident in drilling operations, where critical decisions have to be made in real time and under uncertain conditions.

“Experience gained through the implementation of new drilling technology shows that digitalisation and automation can not only help improve safety, but also introduce new risks and vulnerabilities. When decision-making is transferred to more automated systems, it can become more difficult to detect errors, non-conformant sensor data or unexpected interactions between systems.

During a drilling operation, such conditions can develop rapidly and lead to serious incidents if the operators do not gain sufficient insight into what the system is doing, why it is doing it, and when it is necessary to intervene,” stresses Mihai.

“Drilling operations are complex and not everything can be measured directly. To support human oversight, the systems must be able to share relevant information”.

The information that is required depends on the user. Operational users need insight into uncertainty and actions, while those in more supervisory roles must understand the limitations of the systems concerned. At the same time, the amount of information must be limited so that it actually enhances situational awareness.

Technical solutions that support “the human in the loop”

A key goal of the research is to develop technical solutions that support humans in both expected and unexpected situations.

In automated drilling, for example, it is important to have solutions that will bring the system to a safe state when something unexpected happens and that allow the operator to take control.

Mihai highlights Safe Mode Management (SMM) as an important tool.

SMM is a method that can manage the transition from autonomous operation to manual control.

“With SMM, the system is switched to a temporary safe state, instead of shutting down abruptly or leaving everything up to the operator. The aim is to give the operator time to understand and assess the situation, while at the same time reducing the risk of errors,” she explains.

This type of safe transition is particularly important because autonomous systems must also be able to handle unforeseen situations where the basis for decision-making may be uncertain or incomplete.

To illustrate how risks and incidents can arise in practice, the research article uses three “what if” scenarios. The scenarios illustrate how AI-based solutions – including incomplete data and unclear human-system interfaces – can contribute to an incident occurring or developing in the wrong direction.

Scenarios and practical challenges

1. Autonomous drilling system in real-time operation 
This scenario describes a situation where a system makes decisions based on uncertain data. The scenario illustrates how a lack of insight can make it difficult for drillers to intervene in time.

2. AI-based MPD system with a supervisory operator role 
This scenario highlights the risk of impaired situational awareness if the basis for decision-making is not sufficiently visible.

3. Planning tools based on language models 
This scenario illustrates how errors in generated plans and a lack of understanding of the model’s limitations can lead to incorrect use.

Taken together, these scenarios illustrate the need for robust mechanisms for information sharing, information presentation, error management and a safe transition to manual control.

Human oversight in distributed systems

Going forward, developments in the industry will increasingly require systems from different suppliers to interact. This increases complexity and can make it more difficult to detect errors and non-conformities in a timely manner.

“For example, systems based on different data and models may produce different assessments in the same situation. The question then becomes how we can automatically detect such inconsistencies and flag them,” notes Mihai.

This is not just about decision-making logic, but also about mechanisms for highlighting uncertainty and non–conformities in a way that supports human oversight and control.

A clear message

“The increased use of autonomy is changing the role of humans and placing new demands on technology,” says Mihai.

“If autonomous drilling systems are to be used safely, they must be designed not only to allow for human control, but also to provide a genuine understanding of what the system is doing, how decisions are made, and what safety mechanisms support human operators.