NeurIPS 2019 Expo Talk
Dec. 8, 2019
Trust and AI – addressing AI risks and governance
Nigel Duffy (EY)
While there are conversations around the dangers of unethical AI, there has not been enough progress made in translating these concerns to an actionable approach. The ecosystem is still focused on “what” the ethics are, and not making enough progress on “how” to put checks and balances in place to know if an AI system is abiding by those ethics and can be trusted.
If AI is to be trusted by the stakeholders it interacts with, then clear and comprehensive oversight procedures need to be developed. Regulatory compliance testing, performance auditing, quality management standards, and trusted third-party assessment need to be established.
Developing these requires an interdisciplinary approach, bringing together technical, legal, assurance and audit expertise. EY is proactively addressing this ‘how’ by working with a wide range of stakeholders to define practical guidance and strategies for building technologies that will advance society.
This talk will cover our latest data on the topic of ethical AI, thinking on addressing what questions need to be asked and methodologies put in place for AI practitioners to design and implement a trusted system, and how to monitor if it is acting ethically.