How can trust in algorithms be restored? From technical transparency to explainability requirements

In October 2020, Eglantine Schmitt, CITiO Product Director, participated in a conference on the issue of technical transparency in artificial intelligence (AI). The conference set out to reexamine the need for transparency in AI and proved such a success that it lead to the writing of a book, which has just been published.

Chapter: ”Restoring trust in algorithms through explainability”

Eglantine wrote the chapter ‘Restoring trust in algorithms through explainability’ for the book. This chapter looks at the growing call for transparency in the way algorithms work and how this relates to user trust. Eglantine explains that this requirement can be difficult to apply to artificial intelligence technologies, which can be opaque by nature. She therefore proposes a new type of explainability to restore user trust, by exploring the way the algorithm is designed and creating a dialogue between those involved. In other words, rather than trying to understand the algorithm itself, the challenge is to explain how it works, how it was designed and in what organisational and socio-political context. Among other things, she refers to the economic and technological risks related to the need for transparency, which can compromise algorithm security.

How does CITiO apply these principles?

At CITiO, we deploy various algorithms to correct data. And we are keen to explain why and how our algorithms work. This is why we have set up various initiatives to explain our processes in layman’s terms, and why we attend conferences to explain our work and our approach.

Here are some of the methods we have used to explain this issue to the general public:

Some examples of conferences in which we have participated:

The book as a whole and the challenges it addresses

The book is divided into several chapters on a variety of subjects related to artificial intelligence and new technologies. There are some people who believe that the growing opacity in such areas is dangerous and are calling for greater transparency in order to better manage their impact. The engineers and platforms that develop these systems are looking for a different kind of transparency where they would be invisible to users. Despite the word transparency being in common use, it still needs to be agreed exactly what it means and why it is important and necessary for technical devices. The contributions in this book examine the relationship between transparency and technology, especially in the field of intelligent algorithms.

Where can you find this book?

The book is available for purchase to gain a more in-depth understanding of these important challenges for the future, via this link: https://www.vrin.fr/livre/9782711684267/intelligence-artificielle

Audrey Fleurot