Book chapter
Towards unifying the descriptive and prescriptive for machine ethics
Trolley Crash, pp.69-88
Academic Press
2024
DOI: 10.1016/B978-0-44-315991-6.00011-X
Abstract
Due to the ebb and flow of our social norms, one must rely on the testimony and feedback of other social agents to stay current. However, there exist basic moral norms that an agent should grasp without relying on such evidence. How do we reconcile these two facts when building ethical artificial intelligence systems? Recent attempts at building such systems have been purely empirical, or in the realm of descriptive ethics, ignoring the need for a prescriptive basis. Thus, for the outputs of such models to be ethical they must then get lucky with ethical inputs. Here, I argue that we must minimize such reliance on luck by unifying prescriptive and descriptive ethics, i.e., by providing norm learning systems with moral guard rails. I further argue that testing such reliance on luck is a necessary next step for machine ethics research and I provide a potential framework for evaluating this by drawing upon research on the moral-conventional distinction.
Details
- Title: Subtitle
- Towards unifying the descriptive and prescriptive for machine ethics
- Creators
- Taylor Olson - Northwestern University
- Contributors
- Hsin-Fu Wu (Editor)Shannon Ellsworth (Editor)Michael Salpukas (Editor)Peggy Wu (Editor)
- Resource Type
- Book chapter
- Publication Details
- Trolley Crash, pp.69-88
- DOI
- 10.1016/B978-0-44-315991-6.00011-X
- Publisher
- Academic Press; San Diego, CA
- Language
- English
- Date published
- 2024
- Academic Unit
- Computer Science
- Record Identifier
- 9984948044202771
Metrics
1 Record Views