Response to Wollondilly’s draft AI Ethical Use Policy

Hi Eric,

Thank you for inviting feedback on Wollondilly's draft AI Ethical Use Policy. Please accept the following submission, reference CM 14789~21, from Alex Carpenter in a personal capacity.

I support the policy's practical approach to enabling useful AI adoption. The protections for confidential information, human accountability and review of consequential decisions provide a sound starting point. I suggest six refinements to help staff apply these commitments consistently.

These comments address the exhibited policy. Where an arrangement is already covered in the accompanying AI Governance and Ethical Framework, a clear cross-reference and a short instruction for the user may be sufficient.

1. Make training practical and accessible

The responsibilities on page 12 include completing required training. Please specify a short induction before first use, followed by practical refreshers when tools or approved conditions change. Staff should practise checking invented facts and references, recognising sensitive information, identifying bias, and deciding when to escalate or avoid AI. Include contractors, volunteers and Councillors in guidance suited to their responsibilities. Managers need additional preparation to assess proposed uses. Provide time and support for learning, and check understanding through realistic tasks rather than attendance alone.

1. Give staff a clear approval path

The amber rules on pages 7–8 require manager approval, while higher-risk uses follow a separate route. A one-page decision guide could identify the approval role, required information, expected response time and escalation contact. An approval should record the purpose, tool, permitted data and review conditions. Repeated use within those conditions could rely on the existing approval, with reassessment when the purpose, data or tool changes. This would make the safeguards easier to follow without requiring a fresh application for every routine task.

1. Classify risk by the use and its consequences

Pages 5–6 list chatbots, content generation and emotion or sentiment analysis as examples of Limited Risk. The policy should make explicit that these examples do not determine the category by themselves. A tool used to summarise public information presents different consequences from the same tool used to assess staff or influence a resident's entitlements. Add a short rule that the purpose, affected people, information used and consequences determine the classification, with prohibited and High Risk provisions taking precedence. Include one worked example showing how a familiar tool can move into a higher category.

1. Make human review usable in practice

The accountability principle on page 5 already provides for human review or escalation. Please make the route visible to residents when AI materially informs a decision affecting them. Identify how to request review, who can reconsider the decision, and how errors will be corrected. The reviewer needs sufficient information, time and authority to challenge the output. Retain an accessible telephone or in-person route for people who cannot use a digital service, and ensure that raising a concern does not depend on knowing how the AI works.

1. Include staff experience and service quality in evaluation

The consultation mechanism and annual reporting on pages 10–11 offer a useful basis for learning. Involve affected staff and their representatives before material changes to their work. Assess workload, skill development, service accuracy, accessibility, complaints and corrected errors alongside time and cost savings. Where tasks change, include training and redeployment planning. The goal should be better services and worthwhile work, with people able to exercise judgement and develop their capabilities.

1. Carry these principles into procurement and ongoing review

For enterprise systems governed through the Framework, ask suppliers for proportionate evidence about data handling, security, limitations, accessibility and support for human review. Consider continuity of service, the ability to change providers and environmental resource use where material. Start with a defined service problem, test whether the proposed tool helps, and retain the ability to stop or change an unsuccessful use. Review after a material system change or serious incident as well as on the annual cycle.

These refinements would connect the policy's principles to everyday decisions while preserving room for useful experimentation. Thank you for considering them.

AI tools assisted with drafting this submission.

Warm regards,

Alex Carpenter