AI-RULE for Trustworthy AI
As AI systems become pervasive, and are applied to life-critical use-cases, there are increasing concerns about the power and trustworthiness of these systems. Human-First AI takes into account these concerns and supports the implementation of a set of actionable rules to ensure the development of trustworthy AI. The principles are set forth in the AI-RULE methodology below.
H1st models & applications should exhibit the following properties:
Item |
Trait |
Description |
---|---|---|
A |
Auditable |
Actions and artifacts are recorded for examination |
I |
Idealizable |
Aspires to ideals like fairness ethics |
R |
Reliable |
Optimizes for quality & reliability, particularly in life-critical systems |
U |
Understandable |
Overall systems and models are described for understandability |
L |
Lockable |
Optimizes for the privacy and security users |
E |
Explainable |
Particular decisions are explained for human trust & confidence |