Paper Abstract
"Artificial intelligence (AI) is increasingly being integrated into different areas of our lives. AI has the potential to increase productivity and relieve workload on staff in high-pressure jobs such as healthcare. However, most AI healthcare tools have failed. For AI to be effective, it is vital that users can understand how the system is processing data. Explainable AI (XAI) moves away from the traditional ‘black box’ approach, aiming to make the processes behind the system more transparent. This experimental study uses real healthcare data – and combines a computer science and psychological approach – to investigate user trust and understanding of three popular XAI algorithms (Decision Trees, Logistic Regression and Neural Networks). The results question the contribution of understanding towards user trust; Suggesting that understanding and explainability are not the only factors contributing to trust in AI. Users also show biases in trust and understanding – with a particular bias towards malignant results. This raises important issues around how humans can be encouraged to make more accurate judgements when using XAI systems. These findings have implications in relation to ethics, future XAI design, healthcare and further research."

The full paper is available through SpringerLink, provided that you have proper access rights.