AI for Queensland Businesses

Artificial Intelligence (AI) systems have become increasingly prevalent in various industries, including the public sector. Queensland Government has developed its own secure generative AI system called QChat. Additionally, there are other AI systems available such as Microsoft 365 Copilot, ChatGPT, and Bing AI. These generative AI systems have the ability to learn and evolve, allowing them to create new content based on existing information and respond to user prompts.

Generative AI Systems and Recordkeeping

It’s important to note that generative AI systems are not recordkeeping systems themselves. Therefore, any records created through these systems need to be captured in the agency’s existing recordkeeping system. As guidance on AI systems continues to evolve, it is crucial for agencies to ensure they meet their recordkeeping responsibilities within their existing framework.

Recordkeeping Requirements for Generative AI Systems

Generative AI virtual assistants, like QChat and Microsoft 365 Copilot, utilize prompts to generate responses. These systems do not independently create, action, or send information to other systems. The recordkeeping requirements for generative AI systems are the same as those for all business activities within an agency.

Agencies must assess the value of content created by generative AI systems and retain it for the appropriate time. To determine the value of this content, agencies should refer to their existing legislative and policy frameworks, Section 7 of the Public Records Act 2002, and the Records governance policy’s six policy requirements. Additionally, relevant retention and disposal schedules, including the General retention and disposal schedule, can help establish minimum retention periods for the records.

When content created by generative AI systems does not hold ongoing business or legal value for the agency, it can be classified as transitory or short-term value. These records can be retained for a shorter period before disposal.

Minimum Retention Periods for QChat Records

In the case of QChat, records are retained in the system for three months and disposed of after six months. It is important to understand that QChat serves as a repository for transitory and short-term value records only. For alternative generative AI systems like Microsoft 365 Copilot or ChatGPT, automatic retention and disposal timeframes may not be set. Agencies using these systems should investigate the available retention and disposal features within the products.

AI Interactions and Source Data

Agencies must comprehensively document all interactions related to generative AI systems, regardless of whether they function as virtual assistants or create official documents. This documentation should include prompts, responses, conversations, and all processing steps. These records must be retained in accordance with the minimum retention requirements.

When using a generative AI system response to create content, it is essential to be able to identify the source of the data. QChat allows for accessible source data. However, in some systems like Microsoft 365 Copilot and Microsoft Purview, the identification of source data may not be readily available or accessible. To ensure complete and reliable records, existing systems and processes should be used to relate the AI-generated content to the final official document.

AI Interactions with Multiple Versions of Documents

Generative AI systems rely on the accuracy of the information they source their responses from. It is important to manage redundant, obsolete, and trivial (ROT) information, such as transitory or short-term value information, to avoid confusing or incorrect responses from AI. Agencies can enable archive and deletion policies on collaborative platforms to automate the removal of transitory material. By regularly removing obsolete and irrelevant content, the focus of generative AI systems can be maintained.

Impact of Outdated Documents

Agencies may have multiple versions of a document saved within a system. When generating a response, a generative AI system may use an outdated document that was recently opened instead of the final version of the document. This can lead to inaccurate or outdated information being included in the AI-generated content. Therefore, it is crucial to ensure that the most up-to-date and accurate documents are used when utilizing generative AI systems.

In conclusion, generative AI systems have the potential to greatly benefit Queensland businesses. However, it is important for agencies to understand and meet their recordkeeping responsibilities when using these systems. By following the guidance provided and utilizing existing frameworks, agencies can effectively manage and retain records created through generative AI systems. This ensures transparency, accountability, and the preservation of valuable information for the future.