2019/04/30 by Andrea Aler Tubella, Tubella, Andrea Aler, Andreas Theodorou +5
Social Sciences · Computer Science · #Ethics and Social Impacts of AI #Adversarial Robustness in Machine Learning #Blockchain Technology Applications and Security
paper · pdf · doi:10.48550/arxiv.1905.04994
Artificial Intelligence (AI) applications are being used to predict and\nassess behaviour in multiple domains, such as criminal justice and consumer\nfinance, which directly affect human well-being. However, if AI is to improve\npeople's lives, then people must be able to trust AI, which means being able to\nunderstand what the system is doing and why. Even though transparency is often\nseen as the requirement in this case, realistically it might not always be\npossible or desirable, whereas the need to ensure that the system operates\nwithin set moral bounds remains. In this paper, we present an approach to\nevaluate the moral bounds of an AI system based on the monitoring of its inputs\nand outputs. We place a "glass box" around the system by mapping moral values\ninto explicit verifiable norms that constrain inputs and outputs, in such a way\nthat if these remain within the box we can guarantee that the system adheres to\nthe value. The focus on inputs and outputs allows for the verification and\ncomparison of vastly different intelligent systems; from deep neural networks\nto agent-based systems. The explicit transformation of abstract moral values\ninto concrete norms brings great benefits in terms of explainability;\nstakeholders know exactly how the system is interpreting and employing relevant\nabstract moral human values and calibrate their trust accordingly. Moreover, by\noperating at a higher level we can check the compliance of the system with\ndifferent interpretations of the same value. These advantages will have an\nimpact on the well-being of AI systems users at large, building their trust and\nproviding them with concrete knowledge on how systems adhere to moral values.\n