Artificial Intelligence Tip Sheet

This guide outlines key responsibilities for public servants using AI in any policy, program, or service. It is grounded in federal directives, standards, and legislation, with a focus on identifying and removing bias and barriers.

Core Principles for Responsible AI

Every public servant involved with an AI project must understand and apply these foundational principles:

• Human Oversight: A human must always have the final say in decisions that impact a person's rights or well-being.
• Fairness by Design: Equity is not an add-on or a nice to have. Legal considerations require users to proactively identify, challenge, and mitigate bias at every stage.
• Data Integrity: The quality and fairness of AI systems depend entirely on the quality and fairness of the data it's trained on.
• Accountability & Redress: Clear mechanisms must exist for people to challenge AI-driven decisions and seek recourse.

1. Validating & Managing Data

AI systems learn from data. If this data reflects historical or societal bias, the AI will learn, perpetuate, and even amplify that bias.

• Audit Your Data Sources: Before any development, rigorously analyze your datasets for historical biases. For example, if historical data shows a certain demographic was disproportionately denied a service, using that data without correction will teach the AI to continue that discriminatory pattern.
• Ensure Data Representativeness: Ensure data reflects the diversity of people in Canada. If there are gaps (e.g., underrepresentation of Northern communities or persons with disabilities), develop a strategy to address them before proceeding.
• Practice Data Minimization: Only collect and use the data that is absolutely necessary for the system’s purpose. Every extra data point increases the risk of introducing bias and privacy violations.
• Establish Clear Data Governance: Appoint clear ownership and accountability for the data's quality, lifecycle, and ethical use.

Key Resources:

• Guideline on Service and Digital (See Appendix C: Standard on Enterprise Data)
• Statistics Canada's Quality Assurance Framework
• PIPEDA fair information principles - Office of the Privacy Commissioner of Canada

2. Developing & Updating Policies

When creating or changing policies that involve AI, assess its impact on people from the very beginning.

Conduct an Algorithmic Impact Assessment (AIA) to determine the system's risk level. This must include an explicit assessment of the proposed data sources for potential bias.

Incorporate foundational legislation like the Accessible Canada Act, the United Nations Declaration for the Rights of Indigenous People and the Employment Equity Act  in the policy analysis.

Challenge policy assumptions in areas like risk scoring, eligibility determination, and the underlying assumptions of the policy itself that could lead to discriminatory outcomes.

Key Resources:

• Algorithmic Impact Assessment (AIA) Tool
• Accessible Canada Act
• United Nations Declaration on the Rights of Indigenous Peoples Act
• Employment Equity Act
• Guide to Peer Review of Automated Decision Systems - Canada.ca

3. Designing Programs & Services

Embed fairness directly into the architecture of any AI-powered program.

Ensure meaningful human oversight and provide plain-language notices to users explaining how the AI works and how to challenge a decision.

Include systemically marginalized groups in all phases, from initial design to final testing and implementation.

Audit all AI tools for equity, especially internal systems, to ensure they do not perpetuate bias and barriers.

Embed accessibility and bias mitigation throughout design, testing, and implementation.

Key Resources:

• Directive on Automated Decision-Making
• Policy on Service and Digital
• Accessible and Equitable Artificial Intelligence Systems - Accessibility Standards Canada
• Guide to Peer Review of Automated Decision Systems

4. Procuring Technology & AI Systems

Require conformance with accessibility standards in all procurement contracts. This includes both the hardware/software standards and the specific standards for AI.

Require potential vendors to disclose the sources of their training data, their data-cleaning methods, and the steps they took to mitigate bias in their models.

Require conformance with accessibility standards in all procurement contracts, including ICT and AI-specific standards.

Mandate an external, independent peer review for any high-impact AI system before a contract is finalized and before deployment.

Key Resources:

• CAN/ASC - EN 301 549:2024 Accessibility requirements for ICT products and services (EN 301 549:2021, IDT) - Accessibility Standards Canada
• ASC-6.2 Accessible and Equitable Artificial Intelligence Systems - Accessibility Standards Canada
• Directive on Automated Decision-Making (See section 6.3.3.6 on GBA Plus)

5. Managing & Supervising Teams

As a leader, help build your team's capacity to work with AI ethically and inclusively.

Obtain training on bias, equity, and accessible design principles.

Actively engage employees from systemically marginalized groups to gather feedback on AI tools and processes.

6. Working in Human Resources (HR)

Exercise caution to prevent AI from creating discriminatory barriers in recruitment, promotion, or talent management.

Do not use AI in hiring or promotion unless:

AI training data was audited and corrected for biases related to gender, race, disability, and other protected grounds.

The system has been independently audited for equity impacts in a Canadian context.

Interfaces are fully bilingual and accessible.

Ensure AI-enabled learning or assessment platforms are barrier-free and have been co-designed with meaningful consultation from systemically discriminated groups.

Conduct an Algorithmic Impact Assessment for any system that automates decisions affecting employees' rights or careers.

Key Resources:

• Artificial intelligence in the hiring process - Canada.ca
• Enhance fairness and reduce bias in the content of assessment tools - Canada.ca
• Employment Equity Act
• Accessible and Equitable Artificial Intelligence Systems - Accessibility Standards Canada
• Algorithmic Impact Assessment (AIA) Tool

6. Supporting Indigenous Rights & Self-Determination

Ensure AI systems respect the rights and data sovereignty of Indigenous Peoples.

Align AI systems with the principles of the the United Nations Declaration on the Rights of Indigenous Peoples Act and the OCAP® Principles (Ownership, Control, Access, and Possession).

Include Indigenous Peoples, leaders, and networks in the design, procurement, and governance of any AI system that may affect them.

Avoid automated systems that could reinforce or create new systemic inequities for Indigenous Peoples.

Key Resources:

• The First Nations Principles of OCAP®
• United Nations Declaration on the Rights of Indigenous People Act

7. Monitoring, Evaluating & Auditing

Continuously assess the real-world impact of AI systems to ensure they remain fair and effective over time.

Assess AI-related impacts using GBA Plus assessments, program evaluations, and privacy and accessibility audits.

Continuously monitor system outputs for unexpected or inequitable results. If an AI system starts flagging a specific demographic at a higher rate, it requires immediate investigation.

Report transparently on AI risks, mitigation efforts, and any updates made to the system.

Establish clear feedback and redress mechanisms so users can challenge an automated decision

Key Resources:

• Directive on Automated Decision-Making- Canada.ca 6.3,3.6
• Gender-based Analysis Plus (GBA Plus) - Canada.ca.

8. Engaging the Public & Partners

Foster public trust through clear communication and meaningful collaboration.

Provide plain language explanations of what AI tools do and how they impact people.

Ensure all outreach is culturally relevant, linguistically accessible, and inclusive of marginalized communities.

Co-design AI systems with systemically marginalized groups and by recognizing that persons with disabilities must be involved in creating policies and services that affect them.

Key Resources:

• ASC-6.2 Accessible and Equitable Artificial Intelligence Systems - Accessibility Standards Canada

9. Designing Training & Communications

Create educational materials that are accessible and inclusive.

Use inclusive and accessible formats like screen-reader-compatible documents, captioned videos, and translated materials.

Co-create content with  systemically marginalized groups to ensure it is relevant and respectful.

Go beyond "performative" training. Invest in meaningful education that helps public servants understand and confront systemic racism, ableism, and colonialism.

Key Resources:

• ASC-6.2 Accessible and Equitable Artificial Intelligence Systems - Accessibility Standards Canada

10. For All Public Servants: Your Personal Responsibility

Regardless of your role, you have a part to play in ensuring the responsible use of AI.

Seek out training on data literacy, AI ethics, and unconscious bias. Understand the basics so you can ask critical questions.

If you are asked to work on an AI project, ask "What are the risks of bias in this data?" and "Who might be negatively impacted by this system?"

Actively consult with systemically marginalized groups and colleagues from different levels, locations and classifications. Their insights are invaluable for spotting potential issues.

Key Resources:

• AI Strategy for the Public Service
• Guiding principles for the use of AI in government
• Guide on the use of generative artificial intelligence