Social well-being and Planet
Robustness & Stability
Accountability
Transparency
Privacy & Safety
Fairness & Equity
Models with uncertainty estimates to improve robustness
Set standards and assign responsibility for AI-generated content
Employ diverse and representative training datasets
Develop standards for transparency and disclosure
Carbon aware computing
Set data privacy policies
Document data sources, model architecture, and training procedures
Interpretability techniques such as feature importance etc.
Techniques like Reinforcement Learning from Human Feedback
Adversarial training to protect against attacks
Regularly audit AI models for fairness and equity
Sustainability reporting
Data augmentation, data compression, and selective data sampling
Data augmentation, reweighting, or
Watermarking, metadata embedding for attribution
Guidelines to generate synthetic data
Error Analysis and Validation
algorithmic adjustments
Minimize PII data collection and ensure secure handling
Strategic roadmap to scale
CXOs
Ethics Committee
Privacy by Design
AI Engineers
The Dynamic Duo: Synergy of human behavior and AI systems After laying out the fundamental principles, attention should be directed toward the actions of humans and AI systems. When considering behaviors, areas for consideration include the following:
1. CONTESTABILITY
Contestability is the ability for individuals and stakeholders to challenge or contest the decisions, processes, and outcomes generated by AI systems. For instance, if you applied for a loan and your application got rejected, it is important to have the right to contest that decision. The RAI framework should promote contestability as a desirable behavior. Users should be free to raise concerns or contest decisions related to ethical challenges or sensitive topics. Empowering users to contest AI outcomes is crucial for maintaining transparency and fairness.
2. HUMAN-CONCENTRICITY
When it comes to enterprise adoption of AI systems, the goal should not be to implement them for the sake of doing so. Instead, careful consideration and strategic planning are essential to ensure that implementing AI benefits your business. In today's tech-savvy world, users have become overly reliant on technological artifacts, making decisions primarily influenced by technological developments. Henceforth, it becomes crucial to prioritize human-centric AI systems, catering to their needs as the mainstay, rather than expecting individuals to put AI at the center.
© 2023 Fractal Analytics Inc. All rights reserved
06
Made with FlippingBook - PDF hosting