From traditional AI explanations to Human-centered design in XAI for manufacturing applications

As the manufacturing industry continues to integrate AI and ML to improve operations, it is becoming increasingly important to design human-centered XAI for manufacturing applications. The traditional XAI approach prioritises technical aspects and transparency, but the human-centered approach focuses on what, when, and how to explain AI decisions to human end-users, by iteratively involving them in the development process.

The human-centered approach involves techniques such as interviews, hypothetical scenarios, focus groups, and questionnaires to uncover what information is understandable and useful to humans. By involving users in the development process, we can design AI systems that prioritise transparency, interpretability, and user-centeredness. This approach prioritizes the user's needs and preferences and ensures that the AI system is transparent and interpretable to them. Additionally, the use of scenarios early in the system development process to identify user needs for explanations, which can then serve as a basis for further development of explanations. By using this approach, manufacturers can ensure that AI systems are designed to meet the needs of human workers, resulting in safer, more efficient, and more effective operations.

Another area where human-centered design XAI is important in manufacturing is collaborative robotics or cobots. Cobots are designed to work alongside humans, often in close proximity. Human-centered XAI design considers not only the technical aspects of cobot design but also the social and cultural factors that influence human-robot interactions. This approach helps ensure that cobots are used in a way that is safe, efficient, and effective for humans.

The STAR ICT-38-2020 project aims at leveraging human-centered approaches to XAI, focusing on uncovering what, when, and how to explain AI decisions to human end-users, by involving them in the development process. By prioritising transparency, interpretability, and user-centeredness, human-centered XAI can help to improve safety, reduce the risk of errors and accidents, and increase the efficiency of operations in the manufacturing industry.

By Georgios Makridis, Panagiotis Koulouris, Dimitris Dardanis, Spyros Theodoropoulos, Dimosthenis Kyriazis / UNIVERSITY OF PIRAEUS RESEARCH CENTER