Ethical AI Decision Guide¶

Ready to Use

Purpose¶

This guide provides a practical framework for making ethical decisions when designing, developing, and deploying AI systems in government. It translates Australia's AI Ethics Principles into actionable decision-making tools.

1. Australia's AI Ethics Principles¶

1.1 The Eight Principles¶

1.2 Principle Interactions¶

Some decisions involve trade-offs between principles:

2. Ethical Decision Framework¶

2.1 The ETHICS Decision Process¶

flowchart LR
    E["<strong>E</strong><br/>Examine<br/>the situation"] --> T["<strong>T</strong><br/>Think through<br/>stakeholders"]
    T --> H["<strong>H</strong><br/>Highlight<br/>risks & harms"]
    H --> I["<strong>I</strong><br/>Identify<br/>options"]
    I --> C["<strong>C</strong><br/>Choose &<br/>justify"]
    C --> S["<strong>S</strong><br/>Safeguard &<br/>monitor"]

    style E fill:#e3f2fd,stroke:#1976d2,stroke-width:2px
    style T fill:#e8f5e9,stroke:#388e3c,stroke-width:2px
    style H fill:#ffcc80,stroke:#ef6c00,stroke-width:2px
    style I fill:#fff3e0,stroke:#f57c00,stroke-width:2px
    style C fill:#f3e5f5,stroke:#7b1fa2,stroke-width:2px
    style S fill:#e0f2f1,stroke:#00796b,stroke-width:2px

2.2 Step 1: Examine the Situation¶

Questions to answer:

Risk Tier Assessment:

2.3 Step 2: Think Through Stakeholders¶

Stakeholder Analysis:

Key Questions: - Have we consulted affected communities? - Are vulnerable groups disproportionately affected? - Have we included diverse perspectives in design? - Are there cultural considerations we need to address?

2.4 Step 3: Highlight Risks and Harms¶

Harm Categories:

Harm Assessment Questions:

2.5 Step 4: Identify Options¶

Option Generation:

Consider alternatives including: 1. Proceed as planned - Accept current design 2. Modify the approach - Add safeguards or constraints 3. Alternative technology - Non-AI or different AI approach 4. Enhanced oversight - Add human review 5. Phased approach - Start limited, expand gradually 6. Delay or defer - Wait for better solutions 7. Do not proceed - Reject the use case

Options Evaluation Matrix:

2.6 Step 5: Choose and Justify¶

Decision Documentation:

Justification Test: - Would I be comfortable if this decision was made public? - Would I be comfortable if I were the person affected? - Can I explain this decision to a non-expert? - Does this decision align with our stated values?

2.7 Step 6: Safeguard and Monitor¶

Safeguards Checklist:

Ongoing Monitoring: - Define ethics metrics to track - Schedule regular ethics reviews - Establish feedback mechanisms - Document issues and responses

3. Common Ethical Scenarios¶

3.1 Scenario: Using AI to Prioritize Service Requests¶

Situation: An agency wants to use AI to triage and prioritize citizen service requests.

Ethical Analysis:

Decision Framework: - LOW risk if used to assist humans who make final decisions - MEDIUM risk if it significantly affects service timing - HIGH risk if it effectively determines service access

3.2 Scenario: Predictive Risk Scoring¶

Situation: Using AI to identify high-risk cases for intervention (fraud, child safety, health).

Ethical Analysis:

Red Lines: - No fully automated decisions for significant outcomes - No use of protected attributes as direct inputs - Must be able to explain why a case was flagged

3.3 Scenario: Chatbot for Public Services¶

Situation: Deploying an AI chatbot to handle citizen inquiries.

Ethical Analysis:

Guidelines: - Always identify as AI - Provide easy escalation to human - Don't handle high-stakes decisions - Don't collect more information than needed

3.4 Scenario: Using External AI Services (e.g., Large Language Models)¶

Situation: Using third-party AI services for government functions.

Ethical Analysis:

Usage Guidelines: - Never input classified or sensitive information without appropriate approvals - Never use outputs for decisions without verification - Document limitations and failure modes - Ensure appropriate data agreements with vendors

4. Ethics Review Process¶

4.1 When Ethics Review is Required¶

4.2 Ethics Assessment Template¶

Section 1: AI Description | Field | Response | |-------|----------| | AI system name | | | Purpose | | | Type of AI | | | Decision type | | | Affected groups | |

Section 2: Principle Assessment

Section 3: Risk Assessment | Risk | Likelihood | Severity | Mitigation | Residual Risk | |------|------------|----------|------------|---------------| | | | | | |

Section 4: Recommendation | Recommendation | Conditions | Review Date | |----------------|------------|-------------| | Approve / Approve with conditions / Reject | | |

4.3 Ethics Governance Structure¶

flowchart TB
    EXEC["<strong>EXECUTIVE SPONSOR</strong><br/>Accountable for AI ethics"] --> EB
    EXEC --> EL
    EXEC --> PT

    subgraph EB["<strong>ETHICS BOARD</strong>"]
        EB1[Tier 3-4 approvals]
    end

    subgraph EL["<strong>ETHICS LEAD</strong>"]
        EL1[Guidance & advice]
    end

    subgraph PT["<strong>PROJECT TEAMS</strong>"]
        PT1[Ethics-by-design]
    end

    style EXEC fill:#e3f2fd,stroke:#1976d2,stroke-width:2px
    style EB fill:#f3e5f5,stroke:#7b1fa2,stroke-width:2px
    style EL fill:#fff3e0,stroke:#f57c00,stroke-width:2px
    style PT fill:#e8f5e9,stroke:#388e3c,stroke-width:2px

5. Ethical Red Lines¶

5.1 Prohibited Uses¶

The following AI uses are generally prohibited in Australian government:

5.2 High-Risk Uses Requiring Special Approval¶

5.3 Ethical Boundaries Checklist¶

Before proceeding, confirm: - [ ] This is not a prohibited use - [ ] Appropriate approval obtained for high-risk use - [ ] Human oversight proportionate to risk - [ ] Appeal/contestability mechanism in place - [ ] Affected individuals informed of AI use - [ ] Data use is lawful and proportionate - [ ] No discrimination against protected groups

6. Handling Ethical Dilemmas¶

6.1 When Principles Conflict¶

Resolution Hierarchy:

1. Fundamental rights take priority (no discrimination, privacy)
2. Safety and wellbeing next (prevent harm)
3. Procedural fairness follows (transparency, contestability)
4. Operational considerations last (efficiency, cost)

6.2 Escalation Process¶

If you encounter an ethical dilemma you cannot resolve:

1. Document the dilemma clearly
2. Consult Ethics Lead or designated ethics advisor
3. If unresolved, escalate to Ethics Board
4. If still unresolved, escalate to Executive Sponsor
5. Document final decision and rationale

6.3 Whistleblowing and Raising Concerns¶

If you believe an AI system raises serious ethical concerns:

1. Raise with your manager in first instance
2. If unresolved, contact Ethics Lead
3. If still unresolved, use formal complaint channels
4. Public Interest Disclosure provisions apply

Protection: Staff who raise ethical concerns in good faith are protected under PID legislation.

7. Embedding Ethics in AI Lifecycle¶

7.1 Ethics at Each Stage¶

7.2 Ethics in Agile/Sprint Processes¶

8. Tools and Checklists¶

8.1 Quick Ethics Checklist¶

Before any AI decision or milestone, confirm:

Fairness: - [ ] Tested for bias across demographic groups - [ ] No use of protected attributes as direct inputs - [ ] Diverse training data

Transparency: - [ ] AI use disclosed to affected parties - [ ] Decision logic documented - [ ] Limitations documented

Accountability: - [ ] Clear ownership assigned - [ ] Escalation path defined - [ ] Monitoring in place

Contestability: - [ ] Appeal process available - [ ] Human review option exists - [ ] Review timeframes appropriate

Privacy: - [ ] Data collection minimized - [ ] Consent obtained where required - [ ] Security controls implemented

8.2 Ethics Conversation Starters¶

Use these questions in team discussions:

1. "What's the worst that could happen with this AI?"
2. "Who might be harmed by this, and how?"
3. "If this decision was in the news, how would we feel?"
4. "Would we be comfortable if this happened to us?"
5. "Have we heard from the people affected?"
6. "What are we assuming that might not be true?"
7. "Can we explain this to a non-technical person?"
8. "What would we do if this goes wrong?"

8.3 Ethics Decision Tree¶

flowchart TB
    START([Is AI being used?]) --> Q1{Does it affect<br/>individuals?}

    Q1 -->|No| STD[Standard process]
    Q1 -->|Yes| Q2{Are effects<br/>significant?}

    Q2 -->|No| CHK[Ethics checklist]
    Q2 -->|Yes| Q3{Does it affect<br/>rights/access?}

    Q3 -->|No| ENH[Enhanced review]
    Q3 -->|Yes| FULL[Full ethics<br/>board review]

    style START fill:#e3f2fd,stroke:#1976d2,stroke-width:2px
    style STD fill:#c8e6c9,stroke:#388e3c,stroke-width:2px
    style CHK fill:#fff9c4,stroke:#f9a825,stroke-width:2px
    style ENH fill:#ffcc80,stroke:#ef6c00,stroke-width:2px
    style FULL fill:#ef9a9a,stroke:#c62828,stroke-width:2px

9. Resources and Support¶

9.1 Internal Resources¶

9.2 External Resources¶

9.3 Training and Development¶

10. Glossary¶