Meeting Register Page

Meeting banner
CeADAR Tech Talk - Trustworthy AI in practice – from high-level principles to data science practice - Tjerk Timan
While to topic of Trustworthy AI has been appearing on many policy-and corporate agendas over the last 3 to 5 years, translating often rather abstract goals and terminology into data science practice and organisational day-to-day reality, is not an easy task (see . Reminiscent of the early days of privacy engineering in the context of web 2.0 and what is called the ‘social web’ (Boyd & Ellison, 2007) when the first privacy harms became apparent as a result of mass data collection and centralisation of data services (and servers). With the rise of privacy-preserving technologies and privacy-by-design approaches (see Cavoukian, Hoepman), the approach taken back then was to ‘educate’ computer scientists in ethics in an attempt create more awareness and versatility among the developers and codes behind many digital innovations (See f.i. the work by Friedman on value-sensitive design). Whereas this in itself has had its effects on for example professional ethics, codes of conduct and declarations from the field itself over time (such as the Montreal declaration on responsible AI[1]), the challenges faces by actually developing meaningful AI-based systems and to so responsibly should go beyond the responsibility of the computer scientist or developer alone.

In this talk I will delve into recent cases in which in a multi-disciplinary setting, we have aimed to actually put Trustworthy AI principles into practice during the development of an AI-based service. With many elements to still be tackled, we started off by looking at oversight and auditability of such a system and the different roles, responsibilities and (technical) tools to assess and ameliorate AI-based systems towards a higher level of trustworthiness. I will highlight 2 use cases in which we investigated bias, model robustness and explainability, and I will show parts of a handbook-in-the making on how to set up AI experiments within organisations based on our experiences in these- and other use cases.

Sep 15, 2022 11:00 AM in Dublin

Meeting logo
Meeting is over, you can not register now. If you have any questions, please contact Meeting host: Quan Le.