Yan Jia

Yan Jia studied communications and electronics engineering in China, gaining a Masters’ from the Chinese Academy of Sciences in 2016. In 2017 she completed a Masters’ in Engineering Management at the University of York. Now she is a research associate in York and is in the final stages of a PhD in conjunction with Bradford Teaching Hospitals. Her work focuses on safety management in healthcare, unifying and integrating work in traditional safety engineering and in machine learning. Initial work focused on medication management for patients at risk of atrial fibrillation. More recently she has worked on the treatment of sepsis, using Reinforcement Learning (RL) in recommending treatments, showing how the use of established safety methods can inform the development of the RL elements of the system and support production of a safety case. Recent work has investigated the role of explainability in safety assurance.

Yan_Jia21.jpg

INVITED TALK: Safety of Artificial Intelligence: A Collaborative Model

Achieving and assuring the safety of systems that use artificial intelligence (AI), especially machine learning (ML), pose some specific challenges that require unique solutions. However, that does not mean that good safety and software engineering practices are no longer relevant. This paper shows how the issues associated with AI and ML can be tackled by integrating with established safety and software engineering practices. It sets out a three-layer model, going from top to bottom: system safety/functional safety; ``AI/ML safety''; and safety-critical software engineering. This model gives both a basis for achieving and assuring safety and a structure for collaboration between safety engineers and AI/ML specialists. The model is illustrated with a healthcare example which uses deep reinforcement learning for treating sepsis patients. It is argued that this model is general and that it should underpin future standards and guidelines for safety of this class of system which employ ML, particularly because the model can facilitate collaboration between the different communities.