They determine our news feed, our on-demand video catalog, our purchase suggestions, our internet searches… Algorithms are omnipresent on all platforms, yet their operation remains obscure, even for the most knowledgeable! Given the Platforms: towards new governance power acquired by the platforms, are they able to “use it well”? This is the question that four experts attempted to answer at the Internet Governance Forum . Analysis.
The use of algorithms by platforms
Eric Brousseau professor at Paris Dauphine University / PSL, scientific director of hotels and motels email list the “governance and regulation” chair and the regulators’ club, director of the PSL “governance analytics” research initiative, co-director of the IREN master’s degree (Network Industries and Digital Economy) distinguishes two types of algorithms used by platforms. “Classical ” programming and artificial intelligence (AI) . AI algorithms are “sophisticated statistical inference software at very high frequency and in great depth with a particular objective: to be decision-making” . In practice, algorithms are used by platforms for two use cases: targeting and filtering .
Targeting is used in a content curation logic
This is due to the fact that we have a limited attention span. We can ask machines to select madagascar by the first quarter relevant content,” explains Eric Brousseau . Filtering, on the other hand, allows “taking contextual protection measures while avoiding absolute censorship . ” “On the question of targeting and filtering, today targeting is more in the domain of artificial intelligence, while filtering is still programming .” Even if “in practice, we use both,” he concludes.
Laurène Beccucci , a doctoral student at the 480 Montagne Humanités doctoral email data school, highlights the limits of filtering carried out by algorithms. Particularly in content moderation , a practice which consists of monitoring “all content submitted and generated by users on sites. These are editorial choices based on undefined criteria .” Currently, algorithms have great difficulty distinguishing ironic or sarcastic remarks from others . This is why platforms call on content moderators in addition, who can arbitrate when the machine is unable to do so. There are approximately 100,000 of them worldwide.
Biases inherent in algorithms
“Be careful not to essentialize the algorithm,” warns Loïc Rivière , General Delegate of Tech in France. We tend to believe that the algorithm is a self-sufficient machine, but it remains a human creation. By definition, an algorithm, “is a Platforms: towards new governance process of discrimination, its purpose is to select, classify, order and therefore to discriminate,“ he adds. It is not neutral. This is also the meaning of the work of Aurélie Jean who emphasizes that cognitive biases are transposed to algorithmic biases.