Overview
This lecture examines the often-overlooked role of maintenance in AI systems, arguing that ongoing care, respect, and guardianship are essential for ethical, effective, and equitable technology.
The Importance of Maintenance in AI
- Maintenance is as crucial as design and invention but often neglected in technology and AI.
- Neglected maintenance can have immediate and long-term, unequally distributed consequences.
- AI ethics discourse focuses more on design and catastrophic failures than ongoing maintenance.
What Maintenance Involves
- Maintenance ensures AI systems operate as intended by updating code, fixing bugs, and managing datasets.
- AI systems require continual monitoring, adjustment, and upkeep for accuracy and fairness.
- Databases and physical infrastructure supporting AI must be regularly maintained to prevent harmful outputs.
Labor and Ethics of AI Maintenance
- The essential labor of AI maintenance spans low-paid data workers, engineers, policy analysts, and more.
- Maintenance labor is often invisible and undervalued compared to high-profile technical work.
- Ethical AI must recognize and improve the working conditions and agency of maintainers.
Alternative Frameworks for AI Maintenance
- Indigenous and feminist frameworks promote viewing AI and data as living, relational systems.
- Māori data sovereignty principles emphasize kaitiakitanga (guardianship) and ongoing care for data and AI models.
- Maintenance includes not only supporting systems but also reflecting, modifying, or decommissioning them when necessary.
Operationalizing Ethical AI Through Maintenance
- Maintenance can make AI ethics actionable by enabling transparency, accountability, and equitable outcomes.
- Documentation and small, continual changes support AI oversight and auditability.
- Maintenance practices can challenge the status quo and offer space for system improvement and alternative worldviews.
Key Terms & Definitions
- Maintenance — Ongoing work to ensure systems continue to function safely, effectively, and fairly.
- Kaitiakitanga — Māori concept of guardianship over systems and data, emphasizing care and respect.
- Data Sovereignty — The principle that data should be governed by the people or communities it concerns.
- Sociotechnical Assemblage — A system combining technical components (like code and data) with social elements (like human labor and institutions).
Action Items / Next Steps
- Reflect on how maintenance is (or isn't) incorporated into current AI systems you study or use.
- Research more about Māori data sovereignty and other alternative frameworks for system maintenance.
- Consider the roles and recognition of all workers involved in the AI lifecycle, especially maintainers.