Organisations globally are looking to use Artificial Intelligence (AI) in their operations, with the question being not if, but when. As governments looks to jump aboard the AI trend, we take a quick look at some key considerations.
Guidance on employee use of AI
Unless clear guidelines are in place, staff use of AI can be indiscriminate and expose governments to unforeseen risks, such as disclosure of confidential information to third parties.
Queensland Government Enterprise Architecture's (QGEA) guideline on Use of Generative AI in Queensland Government provides some key considerations, including:
- official information or personal information should not generally be used with generative AI products
- government will provide its own tailored options to access generative AI in a secure and managed environment.
A formal policy on AI use is also expected to be developed in the future. In the meantime, all government bodies should be aware of the guidelines and ensure their employees are clear on permitted and non-permitted uses.
Use of AI in decision-making
A potential use of AI in government is to assist in decision-making processes. However, care must be taken in determining when and how AI should be used. For example, given the lack of transparency in some AI algorithms and models, what if the process is flawed (e.g. bias or hallucinations) and leads to incorrect decisions? This may only be discovered much later and could lead to major problems in dealing with the consequences of those incorrect decisions.
Another important consideration is how to provide evidence for decisions involving AI. If a decision is subject to administrative review, would you be able to provide appropriate evidence from the AI-supported parts of the decision-making process? Is the model suitably transparent to support this?
Supplier contracts
As AI becomes increasingly integrated into everyday life through products such as Microsoft Copilot, governments need to consider what contractual clauses are appropriate to include to ascertain the proper use of AI. For example, you could include clauses such as:
- warranties that any use of AI in the services has been disclosed to the customer (including relevant details to be able to properly assess this use), to avoid using AI in circumstances that are not appropriate
- requirements to ensure that any data provided by the customer in using the service is not fed back into the model to form part of the training set for the next version.
EU AI Act guidance
While Australia does not yet have specific AI legislation, the European Union is leading the way with the passing of its AI Act. Some useful indications of potential future legislative approaches can be taken from the EU AI Act, which classifies AI according to risk – such as prohibiting use of AI in "unacceptable risk" situations (e.g., biometric categorisation systems), and imposing specific and detailed compliance requirements on high risk AI systems.
As a starting point, government should observe what is considered appropriate and inappropriate use of AI to help guide its own decisions on where to begin using AI and what controls should be put in place around its use.
This publication does not deal with every important topic or change in law and is not intended to be relied upon as a substitute for legal or other advice that may be relevant to the reader's specific circumstances. If you have found this publication of interest and would like to know more or wish to obtain legal advice relevant to your circumstances please contact one of the named individuals listed.