Alignment by Default
A hypothesized pathway in which ordinary progress in machine learning and broadly standard training methods produce an AI that learns and robustly pursues human values or intended goals well enough to count as aligned, without a specially engineered alignment method being the main reason for success.
Metadata
- aliases
- en
- alignment-by-default
- concept_id
- alignment_by_default
- preferred
- en
- Alignment by Default
- ja
- アラインメント・バイ・デフォルト
- relations
- relations
- broader
- prerequisite
- status
- active