Principles IFAI
for Responsible Artificial Intelligence
A comprehensive guiding framework that places people and human values at the heart of the development and use of artificial intelligence systems, integrating fairness, accountability, transparency, explainability, privacy, safety, and responsible governance.
A flexible framework for development and use Responsible
Before discussing models and algorithms, responsible use begins with a set of commitments that define who the technology is built for and how it should operate.
AI systems should be designed to serve individuals and communities without unjustified discrimination, while taking into account differences in groups, backgrounds, and needs, and ensuring equitable access to the benefits and opportunities enabled by the technology.
Human dignity, rights, and well-being should remain the highest reference when designing intelligent systems. Technology should support human decision-making, not undermine individuals’ ability to understand, choose, and challenge decisions when their interests are affected.
AI systems should be continuously tested and monitored to ensure they operate within expected boundaries, minimize the risk of harm or unintended use, and provide clear mechanisms for responding to errors and incidents.
Ethical best practices should be integrated into every stage of the development and operation of an artificial intelligence system.
Define the Purpose
Identify the actual need for the system, its intended beneficiaries, and the outcomes it is expected to achieve.
Assess the Risks
Analyze the potential impact on rights, privacy, safety, and affected groups.
Design and Testing
Build safeguards and test performance, bias, robustness, and explainability.
Responsible Deployment
Provide appropriate disclosure to users, define responsibilities, and establish support and appeal mechanisms.
Monitoring and Improvement
Monitor performance, incidents, and changes, and continuously reassess risks.
Fairness
Fairness begins with the data used to build the system and extends to algorithms, decisions, and outcomes. AI systems should be regularly tested to identify unjustified biases and reduce their impact.
Accountability
Responsibility for automated decisions should never be unclear. The roles and responsibilities of design, development, and operations teams must be clearly defined, with appropriate controls in place to reduce risks and clear channels for appeal and escalation when needed.
Explainability
It should be possible to understand the key reasons behind important decisions made by AI systems, with clear and accessible explanations provided to users and those affected by the decision, without relying solely on technical terminology.
Transparency
The system’s lifecycle should be traceable, and users should be informed when they are interacting with a decision or service supported by artificial intelligence, while being provided with an appropriate level of information without compromising privacy or legitimate rights.
AI ethics cannot be treated as a static document. Technologies, risks, and use cases continue to evolve, and ethical guidelines and policies must evolve alongside them.