11-12 December 2017
Algorithms and software implementing them are now ubiquitous in our lives. Rankings, credit scores, recommendation systems, University admissions are just small examples where algorithms are widely used and having a real impact on our societies. This is all the more important since today algorithms (and software) are designed and implemented at an industrial level and most important are embedded in « automatic devices » having a certain decision autonomy.
There is increasing awareness and concern about the impact of this fact for our lives, our societies, our economies. Certainly, it is a topic of concern for computer scientists who actually design and implement algorithms and relevant software. However, it is also a topic of concern for those who use algorithms or make decisions based on them. Under such a perspective, we consider there is room for creating or enhancing a new line of research related to the design of « socially responsible algorithms ». However, despite looking politically correct it needs to be defined through topics:
- Can we consider algorithm design under the perspective of mechanise design theory? After all algorithms are formal procedures allocating resources to tasks and issues such as « fairness », « neutrality » etc. could be relevant.
- Can we audit algorithms? Many algorithms today can have an unpredictable behaviour this depending on the training set (for algorithms which « learn ») or have an unpredictable impact where they are applied, this depending upon the complexity of the situation where they are used. Should we (and how) audit algorithms before allowing their use (as for food and drugs)?
- Are the formal verification and requirements for engineering tools sufficient in order to handle such issues?
- What will be the impact of regulating the use of algorithms on our economies and our societies?
This workshop is organised by LAMSADE and DIMACS with the support of the House of Public Affairs, the Governance and Regulation Chair of Université Paris Dauphine, the GDR Policy Analytics and the GDRI Algorithmic Decision Theory of the CNRS.