In our submission we share our member's feedback that AI is considered a powerful assistant for their work, increasing their productivity and freeing up time for new activities. Critically, no personal data is entered into tools and all outputs are sense checked by a human.

The role we see for regulation is to create a holistic framework of how AI can and cannot be used. This can mitigate the risk of harm to people by banning uses such as social scoring, biometric identification systems and cognitive behavioural manipulation.

Though, ultimately, it is up to an organisation or a person to choose whether they use tools based on AI and which AI tools they use.

The focus for feedback to the discussion paper was to identify any gaps in the matters under consideration. We raised the environmental impact, human rights and product ownership.

Environmental impact

The emissions of the devices used to store the vast quantities of data needed for large language models to operate emit 5x the lifetime emissions of the average car. These emissions are compounded by the energy used to execute learned algorithms plus users’ devices as they raise tens of millions of queries every month in the tools built on these models.

We call for regulation to require designers and developers of these models to publish their source of energy and the model's greenhouse gas emissions.

Human rights

Large language models need to be trained to remove undesirable content such as hate language. In a process known as reinforcement learning, humans are employed by the developer to ask questions to the model and filter out answers based on undesirable content. There has been evidence of exploitation of works and no consideration of the mental health harm caused by repeatedly dealing with undesirable content.

We call for an international standard to require designers and developers to adopt the Guiding Principles on Business and Human Rights issued by the United Nations.

Product ownership

While it is up to the individual to use AI, as the power of AI is built into tools we commonly use today such as Google, it will be difficult for a person to determine when information is generated by a credible source or when it is generated by an algorithm over a pool of data.

Where information generated by AI forms part of a service delivered by a third party, such as a financial planner, the third party can be held accountable if harm is caused. Yet, when harmful content is directly acted on by an individual there is no legal recourse available.

We propose that an international standard include accountability obligations such as the nomination of an accountable person for harm arising from toxic content generated by AI. Within Australia, consideration could then be given to the regulation of AI that incorporates similar principles to the proposed Financial Accountability Regime Bill 2023.

Training

We agree with the proposition that humans should be responsible for overseeing the use of AI in an organisation and be properly qualified and trained. Employee literacy will be key to the safe and responsible use of AI.

We also raised the need for leaders of organisations to become educated on how AI systems differ from traditional IT so they can develop new and effective policies to manage the risks of systems that can adapt without human intervention.

Conclusion

Overall, we support risked based regulation to balance protecting consumers against the harmful use of AI and supporting innovation. We propose that education is the channel to upskill people to harness the opportunities AI presents in a safe and responsible manner.