Burciaga emphasizes that documentation should capture the traceability, clarity, consistency, and accuracy of data, algorithms, and human factors within AI. He also suggests that data and algorithms should have set expiration dates to ensure they are still operating as expected. Lastly, he recommends an Analysis of Alternatives & Actions (AoA&A) approach, which evaluates the performance, operational effectiveness, suitability, and estimated costs of possible alternative systems. He argues that establishing transparency in AI requires discipline in documentation, understanding the life cycle of data, and systematic testing of real-time AoA&A.
Key takeaways:
- Aaron Burciaga has developed a six-point responsible AI framework that includes accountability, impartiality, resilience, transparency, security, and governance.
- Transparency in AI, also known as "explainability," is about ensuring users understand how data, output, and decisions are used and made. It is defined by three key features: documentation, expiry, and analysis of alternatives & actions.
- Documentation involves recording the traceability, clarity, consistency, and accuracy of data, algorithms, and human factors within AI. Expiry refers to the need for data and algorithms to have set expiration dates to ensure they are still operating as expected.
- Analysis of Alternatives & Actions (AoA&A) allows for a full evaluation of different courses of action, considering upstream, downstream, and sidestream effects with explainability and feedback. This approach involves continuously experimenting with different models and monitoring results.