Supported by three standalone yet complimentary essays, this thesis investigates the development of
service personalization that has been mediated by technologies characterized as having elements of
Artificial Intelligence (AI), including prediction, natural language processing, and machine learning.
The aim of this work is to expand our understanding of the role emerging technologies play in
affording personalization, and personalization’s relationship with systems increasingly capable of
mediating experiences directly with users.
Data was collected from participant observation of an AI development company over two and a half
years and comprised of a detailed mapping of the technologies as well as development documents,
chats, meetings, and interviews with developers and key users.
We found that the implementation of deeper forms of personalization over time led to the adoption
of emerging technologies like AI. In the context of a government agency, these algorithms changed
the way employees are screened and selected. We also found that requests for personalization led to
increasingly opaque systems where interpretation about how algorithms work emerges in place of an
explanation of how they work. Building upon these findings, a framework was developed to investigate
34 discrete cases of personalization across dimensions of ease of design and ease of understanding.
We found that the pursuit of deeper personalization leads to the adoption of tools that make
increasingly social decisions. That is, we utilize social technologies despite their complexity because
they make faster and deeper decisions about individuals from social data than can be done without
them. To accomplish this, various strategies are employed to help increase user tolerance for a lack
of understanding of their inner workings and to ensure they operate within bounds acceptable to
users and the designers of the systems. As these systems gain increasing autonomy, issues of bias
amplification, privacy, and an increasingly inexplainable logic behind decision-making remain
persistent and this has implications for theory and practice