Regulation of algorithms: transparency, fairness, moderation

Questions of knowledge

In the article "For a regulation of algorithms"  are addressed the questions of the part of human intervention in an algorithm, how the regulation of these algorithms is practiced and  the transparency or rather opacity of them.

The human programs the algorithms, feeds them with certain data, all with a specific intention and then checks the effects in order to regulate the whole.

In this dynamic field, questioning is frequent and tradition or case law is of little help as the phenomena are so new. The paradoxes between possibilities and desires, between intentions and actual practices lead researchers to define ways of supervising algorithms.

Between intention and result

For example, older algorithms may no longer correspond to reality and yet continue to screen calls coming from landlines; other algorithms have biases established to maximize profits rather than customer satisfaction since it is known that only a small minority will challenge. Not to mention the biases established by an artificial intelligence based on data it has and for which we cannot explain how it arrived at its conclusions; when we talk about transparency, the material is sometimes quite opaque.

This leads to unfair situations for which no one seems to be able to take responsibility and which are difficult to reverse.  For example, fraudsters continue to advertise on the networks without being worried by Facebook, others distribute insanities proposed by their members without assuming responsibility finally others systematically censor everything that could affect their power. Between censorship and laissez-faire, it is not easy to establish a balanced algorithm or to supervise real intentions.

Also upstream approaches, at the level of programming, and downstream, at the level of monitoring, are developing.

"We think that academic research has a lot to contribute on these topics. Instead, it has so far contributed to building great software libraries for predicting and influencing customer behavior, and now aspires to build artificial intelligences that are explainable, frugal, and loyal. Arming the regulator against algorithmic behemoths poses other, non-elementary questions that require a duty of scientific rigor, a discourse of proof, or at least a solution parallel to the method of surveillance."

For the full article: For a Regulation of Algorithms

Visit inria.fr