Author proposes ethical AI without unconsented data

peterbloem.nl44 min read

The author explores whether modern generative AI models can be built ethically without relying on unconsented copyrighted data or destructive book scanning. Focusing on code review as a motivating use case, the essay evaluates legal doctrines like fair use and examines the Common Pile as a potential open dataset. The author argues that combining open data, synthetic generation, and reinforcement learning could enable the creation of capable models without violating copyright. Open article →