Forum 42: How can an AI system ‘break’ ethically and socially?

On Tuesday 26th April, Shalaleh Rismani gave an online webinar for Forum 42, ‘How can an AI system ‘break’ ethically and socially?’. Ms Rismani is a PhD student of Electrical and Computer Engineering at McGill University in Canada, and is executive director at the Open Roboethics Institute (ORI).

SFC1 (Year 12) student Jamie attended the talk and wrote the following:

“Ms Rismani took us through an introduction to Artificial Intelligence (AI) principles and ethical issues in the field. She emphasised that the aim of AI is to benefit humanity in a safe and equitable manner, at which a risk assessment program is crucial to identify the potential hazards. In the talk she explained the thought process behind an AI risk assessment with a holistic approach. This includes supporting the wellbeing of people, respecting human rights, transparency, security, and accountability of AI to ensure the impact is as positive as initially intended.

Another idea she emphasised was the ambiguity of AI failures in a social context. The example she employed was COMPAS, a criminal scoring algorithm that plays a role in sentencing. With such an enormous impact on criminals’ lives, controversies arose as the system appeared to have prejudiced people of colour. She suggested how this reflects an imminent need for an assessment to characterise, identify and mitigate AI failures in a social setting, as well as recruiting a more inclusive community in the process of designing AI.

She then talked about the existing challenges in the field such as the difficulty to achieve impartiality across different social norms and the tension between human autonomy and the system itself. It was fascinating to explore the overlap between computational science and sociology in a broader view of AI application.

Finally, she introduced her research project focusing on a system of assessment of social failures. The key components she outlined were the identification of the social norms and the impacts on the stakeholders within each subsystem. She employed the social failure mode to analyse any possible outcomes and the likelihood of occurrence. This project is yet to overcome the diversity of social norms, and hopefully in the foreseeable future, this can be used to detect AI failures in the refinement of the system.”

Jamie, SFC1 (Year 12)

Deadlines and Reminders