Ethical evaluation of the design and development of an AI service

September 17, 2020 12:30 pm CET

For this new Tech Talk, we propose to conduct an ethical evaluation exercise of an AI service, focusing on Machine Learning/Natural Language Processing and Business Process Management/Business Process Automation.

The first step is to look at the legal and moral framework (issues of transparency, bias, fairness, explicability, etc.): can it be modelled and encoded?
Otherwise, how can we ensure the compliance of the solution?
All of this must also be done with respect for the values of implementation: inclusion, agile methods, Devops …
Once the solution is produced, the question arises as to whether it correctly reflects the original ethical intent: what gap is acceptable, between the desirable and the technically feasible?
Efforts are also made to anticipate the use and consequences of AI service. This produces interesting reflections but the analysis here becomes complex and difficult to reconcile with the encoding/compliance approach: what happens, for example, if the most explainable AI solution is the one that consumes the most energy? Or if it seems less conducive to business?

Ultimately, because it is difficult to bring the thickness of reality into models and computer code, we will have to evoke other means of ethical co-construction of AI: in particular a practical approach based on the eco-system, anchored in reality, sustainable and which seeks justice in design.

Moderated by Arnaud Billion, Head of the “AI Ethics” initiative at the IBM France Lab.
Arnaud is also a lawyer and researcher in fields at the border of Law and IT.
The connection information will be communicated to you a few days before the date of this e-meetup.