SESSION OVERVIEW

Rapid commercilization and widespread usage of AI based tools bring many practical issues forefront together with several exciting benefits. Researches are no longer interested in only efficiency , accuracy or scalability related improvements of methodologies, but also deployment issues are of significant concern as well. Among these issues, two closely interrelated concepts are leading the practical solicitude: interpretibility and fairness of results. Many of the AI algorithms act as a black-box in the sense that it is not clear how and why they arrive at a particular decision. Being transparent about the training sets does not solve the obscurity problem since it is not enough to come up with full interpretation of the mechanism. This is raising undestandable trust, accordingly acceptibility problems from the users and the society perspective.

In the same vein, fairness is viewed as a significant element in artificial intelligent ethics, which refers to “absence of any prejudice or favouritism toward an individual or a group based on their inherent or acquired characteristics” [1]. However it is obvious that there is long way in front of the scientific society to traverse as it can be deduced from the recent efforts that have been put together with a goal of standardizing related metrics.

This session aims at providing general framework behind the issues together with clear representations of the relevant problems as well as possible technical solutions.

In this regard, the scope might include but not limited to:

• – Sample experiments/cases that reveal the practical issues raised by abcence of fairness and explainibility
• – Experiment results that sets and clarifies the relationship among the fairness and explainibility on a common case or by a theoretical proof
• – Interesting implementations OR reviews of existing explainibility and fairness assessment methods
• – Proposals for new metrics applicable to new or existing algorithms together with sample cases
• – New algorithms/approaches to improve fairness and explainibility of results