Abstract
Recommendation engines have become one of the most impactful technological forces influencing public life but are virtually inaccessible to those who experience their effects firsthand. In this capstone research, two parallel approaches will be taken to tackle the problem of recommendation engine opacity: (1) an engineering approach through which recommendation algorithms will be made understandable to regular people, and (2) an STS-based approach examining the social consequences of keeping such technologies under wraps.
The technical component of the project involves the creation of a user-oriented interpretability approach which acts as an intermediate layer between a recommendation algorithm and the user interface itself, providing human-interpretable insights regarding the recommendations made. The framework includes three modes of explanation that correspond to varying labels of technical knowledge of the user: visualization of important features, decision tree approximation, and natural language explanation.
While the technical component addresses how recommendations can be made understandable, it leaves a prior question unanswered: why has opacity persisted and who benefits from it? The STS paper investigates TikTok’s For You Page as a case study in algorithmic power by using the concept of Actor- Network Theory (ANT), a framework that treats human and non-human entities as active participants or actors who shape social outcomes, to define who the users, creators, advertisers, and regulators are, how these groups interact, whether through engagement or monetarily, how the algorithm acts as a kind of social actor (e.g., it moderates the way users/creators and their content connect), and whether or not these actors are regulating the way the algorithm operates.
The main finding of the study is that the opaque function of TikTok’s algorithm helps maintain the asymmetrical power dynamics that exist between TikTok’s users, content creators, advertisers, and therefore reduces the ability of regulators to manage the functioning of the algorithm, allowing TikTok and its users to be unregulated and unaware of how they are being influenced.
The findings of these projects illustrate that, while explainability is essential, it is not sufficient to improve user experience. Regulation needs to improve the way in which users interact with and benefit from the algorithm. Similarly, regulation should also treat TikTok as a constantly changing and evolving algorithm. TikTok is simply a reflection of an entire economy based upon an opaque system of attention seeking. To turn this into a better system, the first step is to make the algorithm and the underlying economic model visible from a technical and moral perspective.
