Using AI tools in research

AI tools may be trained on sources that have racial, gender and ableist biases. Researchers who incorporate AI-generated content into their research without critical assessment risk perpetuating these underlying biases. Additionally, AI tools may be trained on data created by or about Indigenous people without their permission. AI tools can generate outputs that imitate Indigenous art styles that constitute infringement of Indigenous Cultural and Intellectual Property and that do not maintain appropriate cultural sensitivity. Researchers should consult relevant frameworks such as the AIATSIS Code of Ethics and seek guidance when working with Indigenous data or cultural expressions.

While AI tools can be used to generate ideas for research topics, identify relevant sources for research and improve the grammar and readability of existing text, researchers should be wary of using GenAI to autonomously generate significant portions of any academic work. AI tools are trained on vast swathes of data and the sources AI tools are trained on are not necessarily acknowledged within the tools themselves. Researchers should therefore critically evaluate all AI outputs with the same level of discernment they would any other source. Additionally, as a researcher, you should be able to differentiate between your own contribution and that of AI and acknowledge where AI has contributed to your work.

To ensure copyright compliance and maintain data privacy, researchers should avoid uploading sensitive and copyright-protected materials to generative AI platforms that use uploaded content for model training.

Uploading copyrighted content without proper clearance may breach copyright law and violate terms of use. Copyright clearance may need to be conducted for each item before use with AI platforms that use uploaded content for model training, including text, documents, datasets, and multimedia. This may involve confirming that the licensing permits the work to be reproduced digitally or obtaining permission from the rights holder. Materials from library databases, course content, and other third-party sources are generally unsuitable for use with AI platforms that claim rights over user-submitted content.

If you're uncertain about the copyright implications of using AI tools in your research, we strongly encourage you to get in touch with the Library. Copyright in the context of AI remains a complex and evolving area, and guidance is best provided on a case-by-case basis.

AI systems can also pose privacy risks. The complex nature of many AI systems mean that it is not necessarily transparent how data uploaded into the systems is used and even developers of the AI systems may not be able to articulate how all aspects of the system work. There are also risks of data breaches, with attackers attempting to gain unauthorised access to the datasets AI models are trained on. Before using any tool, we encourage you to take a moment to look over its privacy policy.

In comparison to many GenAI tools freely available online, the University-endorsed Microsoft Copilot with Enterprise Data Protection (Copilot Chat) has a higher level of data security. Furthermore, Copilot Chat ensures that user data is not used for model training. For further details on Copilot Chat, please consult Copilot Chat or review this blog post.