AI and Adversarial AI in insurance : Background, examples, and future implications

This paper describes the rapid and dynamic pace of Artificial Intelligence (AI) and Machine Learning (ML) developments that have revolutionized the insurance sector, and discusses their interpretability. AI content-based processing of information includes image and video analysis, facial recognition and automated decision making for claims management and fraud detection.The paper focuses on adversarial AI, namely on the creation of input data slightly altered to mislead a machine learning system and make it produce incorrect predictions.It provides a case study of the impact of adversarial AI in health insurance. Not only can the model be fooled for detecting the malignancy of patients, but a higher level of perturbation can increase the success rate of the attack by lowering the accuracy of the system. We deem important, for insurance companies ready to adopt AI technologies, to be aware of the consequences of adversarial attacks. We conclude with policy recommendations, consistent with the current regulatory framework