The Limits of Value Transparency in Machine Learning
View / Open Files
Authors
Journal Title
Philosophy of Science
ISSN
0031-8248
Publisher
Cambridge University Press (CUP)
Type
Article
This Version
AM
Metadata
Show full item recordCitation
Nyrup, R. (2022). The Limits of Value Transparency in Machine Learning. Philosophy of Science https://doi.org/10.1017/psa.2022.61
Abstract
<jats:title>Abstract</jats:title>
<jats:p>Transparency has been proposed as a way of handling value-ladenness in machine learning (ML). This paper highlights limits to this strategy. I distinguish three kinds of transparency: epistemic transparency, retrospective value transparency, and prospective value transparency. This corresponds to different approaches to transparency in ML, including so-called ‘Explainable AI’ and governance based on disclosing information about the design process. I discuss three sources of value-ladenness in ML—problem formulation, inductive risk, and specification gaming—and argue that retrospective value transparency is only well-suited for dealing with the first, while the third raises serious challenges even for prospective value transparency.</jats:p>
Sponsorship
This research was funded in whole, or in part, by the Wellcome Trust [Grant number 213660/Z/18/Z] and the Leverhulme Trust, through the Leverhulme Centre for the Future of Intelligence.
Funder references
Wellcome Trust (213660/Z/18/Z)
Leverhulme Trust (RC-2015-067)
Leverhulme Trust (RC-2015-067)
Identifiers
External DOI: https://doi.org/10.1017/psa.2022.61
This record's URL: https://www.repository.cam.ac.uk/handle/1810/336740
Statistics
Total file downloads (since January 2020). For more information on metrics see the
IRUS guide.