A student at the School of Social and Political Science has received funding from Edinburgh Futures Institute (EFI) to investigate the cultural and societal challenges involved in the curation of facial recognition datasets.
The project, Hidden Humans in the Loop: Unpacking Societal Challenges in the Curation of Facial Recognition Datasets will be led by Benedetta Catanzariti, a PhD student in Science, Technology and Innovation Studies, with Co-Investigator Sarah Bennett, a PhD student at Edinburgh College of Art.
The funding will allow the students to engage AI practitioners in open and equitable discussion about methods to anticipate and mitigate data-related societal harm.
It is one of ten interdisciplinary and innovative projects across the University selected for the EFI funding.
Benedetta described the project:
“Over the last decade, applications of Artificial Intelligence (AI) facial processing technology have been implemented in a wide range of sensitive domains. These applications are often predicated upon machine learning algorithms trained on large-scale datasets consisting of labelled images categorised according to specific, pre-selected features.
“In addition to work conducted to improve the labelling quality, it is important to consider the outsourced nature of many face datasets. Considering the role of the gig economy in dataset curation is crucial to adequately accounting for the factors shaping the interpretation of facial features.
“In this project, we are not looking to evaluate forms of technical accuracy nor to de-bias datasets; rather, our aim is to foster critical understandings of the value-laden decisions that lead to certain classifications. In doing so, we promote insight into the working practices, challenges and constraints faced by human labellers that are often outside the purview of AI practitioners, and thus critical reflection into their own practices”.
Exploring real-world data curation with AI practitioners
The project will use online participatory workshops with AI practitioners to explore the real-world practice of data curation work.
It will examine:
- the contexts within which much of facial recognition dataset labelling occurs
- how the practices and structures which shape common pipelines of data curation might infuse the interpretation of data
- the agency of gig economy workers within crowd-sourcing approaches
The project builds on a set of three online workshops conducted in 2020 with non-AI experts to explore dynamics of crowdsourced labelling practices.
More about the EFI student research awards
The EFI student awards support student projects with a focus on innovative, multi-disciplinary research for the public good. The selected projects contribute to knowledge and capacity building in Data Driven Innovation across the City of Edinburgh and beyond.
EFI Acting Director of Research, Dr Ewa Luger said:
“EFI is delighted to be able to support these highly talented students in their individual research journeys. We are again excited and proud of the imagination, creativity and innovation arising from Edinburgh’s research students as they apply interdisciplinary responses to a range of compelling challenges. We look forward to seeing the results and sharing them with our community as they report."