University of Twente Student Theses
Epistemic and Ethical Issues in Machine Learning Based Recidivism Risk-Assessment: Lessons from Philosophy of Measurement
Peet, Luca van der (2023) Epistemic and Ethical Issues in Machine Learning Based Recidivism Risk-Assessment: Lessons from Philosophy of Measurement.
This is the latest version of this item.
|
PDF
584kB |
| Abstract: | Trade-offs between different definitions of fairness plague many machine learning applications, from mortgage lending, to hiring algorithms and recidivism prediction tools. However, the existing literature on fairness in machine learning resides predominantly in the computer science domain and struggles to offer solutions to these trade-offs. Fairness is mostly reflected in a formulaic and reductionist view and significant progress in statistical progress has stagnated for decades. In this thesis I revisit the discussion that erupted around one of the most notorious algorithms that was accused of persistent fairness issues towards black people: the COMPAS algorithm for recidivism prediction used by many courts in the US. I introduce a recent publication from philosophy of measurement which addresses fairness issues by abandoning the commonly accepted benchmark for accuracy in machine learning for a metrological conception of accuracy (Tal, 2023). I conclude that the latter may be a suitable candidate for rethinking statistical fairness in a more fundamental sense. |
| Item Type: | Essay (Master) |
| Faculty: | BMS: Behavioural, Management and Social Sciences |
| Subject: | 08 philosophy |
| Programme: | Philosophy of Science, Technology and Society MSc (60024) |
| Link to this item: | https://purl.utwente.nl/essays/97759 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page