In response to the emergence of new Generative AI programs such as ChatGPT and similar models, the Department of Health has issued an advisory outlining an interim position on their safe use in Victorian health services.
This interim position covers the use of Generative AI only. It does not apply to other types of artificial intelligence (e.g. advanced analytics).
Generative AI refers to a type of artificial intelligence that creates new data or content, such as images, text, or music, by learning patterns and structures from existing data. Popular generative AI tools include ChatGPT and Bard.
The use of generative AI such as ChatGPT presents a number of risks, including, but not limited to:
- Confidentiality and security breaches: Information entered into a generative AI system may enter the public domain, breaching regulatory requirements, patient privacy, vendor contracts, and causing reputational damage.
- Inaccurate outputs: Generative AI uses algorithms to generate content and may generate content that is factually inaccurate or outdated.
- Transparency: Generative AI is often labelled as a ‘black box’ technology. Users are unaware of how the software ‘generated’ its output. In modern patient centred care, clinicians strive to provide consumers with all the information and decision factors that are considered for their care.
- Biased outputs: Generative AI may produce biased, discriminatory or offensive content.
In response to these risks, the interim position for health services is that unregulated generative AI software such as ChatGPT and other, similar, software should not be used for any clinical purpose. This includes:
- clinical applications of generative AI, such as the generation of discharge summaries
- clinical support functions that use generative AI, such as generation of summaries from patient notes and/or EMR records
- consumer support functions, such as conversational triage ‘bots’
- integration of generative AI into clinical services, such as talk-to-text summaries and/or translations, post consultation.
Permitted uses of Generative AI:
- Content generation as a starting point for non-sensitive communication such as non-clinical email or communications: Content must be proofread and checked for accuracy by a human before being published or shared.
- Content generation as a starting point for non-sensitive documentation: Any information or recommendations generated by AI that influence decision-making should be documented in the appropriate records, along with a clear indication that the information was obtained from an AI-generated source.
- Summarisation: Generative AI may be used to summarise literature, research, or other non-sensitive information sources. The AI-generated nature of the summary must be acknowledged.
- Analysis: Generative AI may be used to analyse literature, research, or deidentified internal information sources (e.g., surveys) that do not contain personal health information or operational performance metrics. The AI-generated nature of the summary must be acknowledged.
- Educational and Training: Generative AI may be used as a supplementary tool for education and training within the health service, provided that it is utilised responsibly and within the context of established curricula and educational materials.
Important Note
In all use cases, particular care must be given to ensuring that no health information, sensitive information, classified information, unique identifiers, or personal information from Monash Health, patients, consumers, customer or third parties is used as an input (in a prompt).
Future developments
The Department of Health’s interim position acknowledges that the innovation and development environment of generative AI is dynamic.
At Monash Health we recognise the potential benefits of using AI to improve patient care, enhance clinical decision-making, and optimise operational efficiency. We are equally cognisant of the importance of using AI responsibly, safely and ethically, with consideration to data privacy concerns, and security risks.
We are currently developing our approach to these emerging technologies within the complex landscape of healthcare and look forward to sharing further details in the coming months as the regulatory environment becomes clearer.
Approved by Martin Keogh, Interim Chief Executive