By akademiotoelektronik, 05/04/2023

TransAlgo: assessing the accountability and transparency of algorithmic systems

How was the TransAlgo project born?

Following the law for a digital republic, Axelle Lemaire, then Secretary of State for Digital and Innovation, ordered in 2016 to the General Council for the Economy (CGE) a report on the procedures for regulating content processing algorithms. One of the recommendations of this report was the establishment of a collaborative scientific platform intended to promote, on the one hand, the development of software tools and methods for testing algorithms and, on the other hand , promoting their use. We are working on setting up a platform called TransAlgo for the development of transparency and accountability of algorithmic systems, due to the duality of data and algorithms.

Inria has been entrusted with the role of operator of TransAlgo , with the scientific contributions of many academic players united around the challenges of TransAlgo , in particular the CNRS. In addition to scientific expertise, Inria will provide software development assistance.

This platform will be a first in Europe.

Why bother about this subject?

A simple question: are automatic recommendations for the consumption of goods and services (for example Netflix for audiovisual content or Amazon for everyday products) fair to the consumer or the service provider? Recommendation engines are becoming more and more prescriptive and their transparency now represents an important economic issue, for example for producers of cultural content. Is the consent to the use of personal data really respected? A recent study by Inria and the CNIL has pinpointed a well-known economic player. In question, its mobile application which exceeded the consent of the user by communicating his GPS position despite the refusal of the latter. In this case, the managers were not aware and had to initiate an internal investigation to understand where the problem came from. The lack of program loyalty is not necessarily intentional! Another example of unfair behavior is the volatile pricing you may see when the price of your plane ticket increases as you visit an e-commerce site. The purpose is not to slow down innovation or new business models but to support innovation by providing informed information to consumers, whether they are citizens (B2C) or businesses (B2B), and by traceability of automated decision-making. Transparency is an asset for the "en-capacitation" of the consumer but also a factor of economic competitiveness. In the case where the consumer of the service is a professional (B2B), a situation of unfair competition or not can be evoked.

There are also sorting mechanisms in search engines, mechanisms for recommending and selecting proposed content that does not currently appear transparently... All of this can have impacts that the Most people still don't measure up on granting bank loans, insurance, recruitment situations, and the like.

TransAlgo: Assessing Accountability and Transparency algorithmic systems

There are real issues, therefore, of information, neutrality, loyalty, fairness, non-discrimination, the fight against unfair competition, respect for consent and privacy, etc. A very important thing to understand however is that the scientific platform TransAlgo will in no way be in charge of the regulatory control of the algorithms or the use of the data. It will offer a range of studies, tools and services to all the players concerned.

What are the scientific challenges of TransAlgo?

The transparency of algorithmic systems is a real challenge for academic research. This calls on several disciplinary skills and many of the topics identified have not yet been sufficiently explored by academic research, hence the importance of multiplying the research effort. Two approaches will be developed by TransAlgo: the auditability of algorithms and the development of new generations of "transparent by construction" algorithms which facilitate the measurement of their transparency, their explanation and the traceability of their reasoning. We will also strive to develop so-called "responsible by construction" algorithms if they respect the laws, and if they comply with certain rules and values ​​of our societies.

An algorithm is transparent if one can easily verify its "responsibility", for example, if it opens its code, if it explains both the origin of the data it has used, and that which he produces, if he explains his results, or even if he publishes traces of his calculations. Note that we will also consider situations where the code is not open because there is no obligation to disclose it.

How are you going to proceed?

To be able to tackle this, it is necessary to first define what is called transparent, neutral, loyal, or fair software, concepts that are quite legal. This work involves checking the conformity between its specifications and its behavior, in other words the gap between what it is supposed to do and what it does. It will also shed light on its compliance with ethical and legal rules. The methods and technical tools of the transparency of algorithmic systems are a complex and multifaceted subject. The properties that we want to verify, for example non-discrimination or loyalty, include an important part of subjectivity, which depends on the use cases and the contexts. This makes their specification difficult. The scientific challenges are numerous and very little research work exists on the subject.

TransAlgo has a substantial educational role vis-à-vis the general public to explain the concepts used, some of which can be objectified and others not. The TransAlgo platform will offer a space for resources and participatory exchanges to the scientific community and beyond. You will find white papers, reports, articles, but also data sets and controlled test protocols. The platform will also be a space for sharing good practices at national and international levels, a training space with online courses. The objective is to increase collective awareness of the issues around data and information management and processing algorithms, to acquire algorithmic tools to empower citizens, public authorities and professionals.

TransAlgo will also be a platform for the animation of the scientific community dedicated to the challenges posed by questions of transparency and the responsibilities of algorithms. These questions require interdisciplinary expertise and bring together several academic players, in addition to Inria and the CNRS, such as Sciences Po, IMT, the University of Grenoble Alpes, the University of Paris-Sud, the University of Versailles-Saint -Quentin, the Pierre and Marie Curie University and the ENS.

We have created five working groups on the following topics:

How can we be sure to have a foothold in the real world?

We are attached to use cases from the lives of citizens and professionals. Associations also have a role to play in identifying and objectivizing the current situation of certain practices of platforms or services through a contributory mechanism (citizens and professionals).

To bring up very real use cases, we have exchanges with the general direction of competition, consumption and the repression of fraud (DGCCRF), the Superior council of audio-visual (CSA), the French Personal Data Protection Supervisory Authority (CNIL), the FING (New Generation Internet Foundation), in addition to the CERNA (Commission for reflection on the ethics of research in digital sciences and technologies of Allistene). We also intend to work on the basis of feedback from the expression of needs that will come from manufacturers and consumer associations.

Tags: