The public are receptive to AI’s use in saving time and driving efficiencies, although there is recognition that the benefits of speed can be lost if there isn’t a focus on accuracy. In this context, the public highlighted the need for monitoring the quality of outputs.

Application

• AI should not be used for automated decision-making
• As a minimum, humans should still read all consulation responses
• AI summarisation can be used for speed but AI summaries will be cross-checked by a human for accuracy and against ‘hallucinations’
• Parameters should be put in place for the types of issues that could be addressed using AI technology. Certain cases which may be more niche, complex, or require more empathy may be better suited for humans to respond to, rather than AI.  
• Typically the application of AI occurs by default where there are a large number of responses for when speed is of the essence
• AI may be used to identify patterns in the data, such as duplicate or near duplicate responses

Transparency

• We will indicate where AI has been used to assist us in the synthesis of views within the body of official output or outcome reports.
• Where AI is used to create public summaries or generate images they should be accompanied with the following text : “To improve efficiency, this content was drafted with assistance from AI. It has been fully checked by a human”
• Where AI is used in automated emails or chatbot correspondence, this will be clear to the participant
• If participants choose to use AI in the formation of consultation responses, we reserve the right to classify these as machine responses which are ineligible for further consideration.

Privacy and security

• Personally identifiable information (PII) should be redacted and not be sent to AI sub-processors
• Feedback analysed using AI should not be used to train the AI model without the consent of the participant
• Participants should be able to ‘opt-out’,  excluding their response from AI analysis
• Periodic human synthesis of results is made to ensure that any AI summaries remain accurate