Guidelines for ethical data annotation and release practices without exploiting workers and stealing data.
We can build useful AI systems without stealing data and exploiting labor. Instead of attempting to guzzle every available data on the internet, we advocate for curation practices that only collect the data needed to build specific modes. Instead of exploiting labor, we advocate for a collaborative approach that treats data workers as knowledgeable builders of AI systems. Instead of releasing datasets and models into the world without documentation, we advocate for the careful documentation practices that are widespread in other engineering disciplines.
Image: Janet Turra & Cambridge Diversity Fund / https://betterimagesofai.org / https://creativecommons.org/licenses/by/4.0/
Our paper details how researchers and practitioners in AI and related fields can more responsibly collect and release the data used to train and evaluate these systems (it has now been downloaded over 350 thousand times, one of the most downloaded articles of the Communications of ACM) the Association for Computing Machinery (ACM)’s flagship publication.
This article maps the legal issues that arise at different stages of the data lifecycle: the data collection, labeling and distribution processes, and how the data is used to build and evaluate models and applications.
Based on more than two years of research and field work, this paper describes a framework for not only documenting datasets but documenting data production process as a whole.
How can fairer working conditions for data workers be created? This Discussion Paper outlines guidelines for the use of data work in academic research.
This peer reviewed study by Karen Boyd “illustrates ways in which Datasheets for Datasets, an accountability intervention designed to help engineers explore unfamiliar training data, scaffolds the process of issue discovery, understanding, and ethical decision-making.”
“Article 50: Transparency Obligations for Providers and Deployers of Certain AI Systems”
The narratives around big data and data science are overwhelmingly white, male, and techno-heroic. In Data Feminism (MIT Press, 2020), Catherine D'Ignazio and Lauren Klein present a mode of thinking about data science and data ethics that is informed by intersectional feminist thought.
