Abstract:
Artificial Intelligence (AI) has become one of the most active areas of research in
the field of Computer Science and Engineering. However, aligning these AI
advancements with human needs and goals is not being investigated at the same
rate. To address this, the idea of Human-Centered Machine Learning (HCML) has
been proposed. ‘Human’ in HCML mainly targets the non-AI-expert stake-holders
of AI systems. However, understanding how the behavior of non-AI-experts affects
the HCML process is under-explored. My PhD research aims to contribute to this
understanding by investigating the effects of the role played by non-AI-experts in
the development of HCML systems. In this thesis, I first explored the current work
in the field of HCML through a systematic literature review to understand the
present state of the field, challenges and opportunities. It helped identify the
underrepresented domains in HCML, over-focus on AI developers rather than
actual end-users, and the ability to leverage existing powerful AI models to easily
investigate the human context of HCML applications.
In addition, I developed 5 HCML applications targeting different non-AI-expert
groups from different domains. With these, I investigate the effects of the role of
the non-AI-expert under 3 scenarios: 1) generic end user, 2) domain expert end
user, and 3) generic end user and domain expert collaborator. In my thesis, I
contribute with several artefacts and empirical studies and present the design
implications of developing HCML systems targeting non-AI-experts in various
domains.
