Original title: Representations and ethical categories in supervised learning
Authors: Hvorecký, Juraj
Document type: Papers
Conference/Event: Cognition and Artificial Life /22./, Olomouc (CZ), 20240515
Year: 2024
Language: eng
Abstract: While successful in many domains, supervised networks have also been behind many serious blunders. These are usually associated with human faults in the design, training or usage of networks. We will utilize the literature on adversarial attacks to indicate that in tasks where robust categories play a crucial role, supervised networks might not offer the best and defensible solution. These tasks fall into sensitive areas of ethics, politics, aesthetics and maybe others. The paper could be seen as a contribution to a debate on the limits of usability of supervised networks across a variety of domains.
Keywords: adversarial attacks; categories; ethics; supervised learning
Project no.: GA20-14445S (CEP)
Funding provider: GA ČR
Host item entry: Cognition and Artificial Life 2024, ISBN 978-80-88123-36-1
Note: Související webová stránka: https://use.icaci.org/cognition-and-artificial-life-central-europe-conference-on-eye-tracking/

Institution: Institute of Philosophy AS ČR (web)
Document availability information: Fulltext is available at external website.
External URL: https://webcentrum.muni.cz/media/3673714/cal2024_proceedings_final.pdf
Original record: https://hdl.handle.net/11104/0357871

Permalink: http://www.nusl.cz/ntk/nusl-668302


The record appears in these collections:
Research > Institutes ASCR > Institute of Philosophy
Conference materials > Papers