Writers Propose 'Seed Bank' Models for Ethical AI Data Governance

Researchers from a human-computer interaction team conducted four workshops with over 100 creative writers to tackle a pressing issue: how to govern language models used for creative writing. Current AI models often limit writer consent, participation, and control. Participants generated more than 200 metaphors—objects, places, processes, and groups—to reason about alternative governance structures. Key themes emerged: the importance of consent in data use, defining community boundaries for sharing, giving proper recognition to contributors, and balancing scale against community needs. Metaphors like 'seed bank' (preserving diverse data), 'co-op' (shared ownership), and 'stoop swap' (informal exchange) illustrated different governance possibilities.

The findings point toward smaller, open models that encode group values rather than monolithic corporate systems. Writers envisioned models where data is treated like a community garden—tended collectively for mutual benefit. The paper discusses concrete ways to make these community language models a reality, such as decentralized data trusts and contributor-led curation. This research challenges the prevailing top-down AI development model, offering a blueprint for writer-centric tools that respect creative autonomy. For professionals building or using generative AI in creative fields, it suggests a future where governance structures can be as diverse and participatory as the writing communities they serve.