Background
Artificial Intelligence (AI) simulates human intelligence processes and can enhance functionality in standalone applications or alongside existing ones. While AI offers numerous opportunities, it also introduces new risks that require careful management. This Statement outlines our approach to AI use within ACIAR, in accordance with the DTA's policy for responsible AI in government. This policy applies to all ACIAR staff, contractors and service providers, noting AI is not used for decision-making purposes at ACIAR.
Policy
AI technologies will be adopted only after a thorough risk assessment and monitored for compliance. Official information must not be entered into public AI services unless already public. Email accounts cannot be used with public AI services without approval. Commercial AI tools require executive approval, and any AI-produced content must be reviewed before publication. Automated decision-making tools need significant risk assessment and approval. AI tools must not access restricted information, perform illegal activities, or assess staff performance. Misuse or risks identified must be reported immediately. Administrative decisions aided by AI will be clearly indicated.
AI applications
Currently, ACIAR primarily employs Microsoft's Copilot as its AI tool. Copilot facilitates staff in various activities by leveraging AI capabilities in the following ways:
- Automation and Efficiency - Copilot automates repetitive tasks such as meeting transcriptions, minute-taking, email searching and data sorting, thereby enabling staff to concentrate on more complex and value-added activities.
- Office Productivity - Copilot enhances textual interpretation and generation by understanding and responding in natural language. This includes structuring documents, summarising written material and assisting with drafting content.
- Collaboration - Copilot aids in real-time collaboration by providing suggestions and insights during document creation, ensuring consistency and improving the overall quality of the work produced.
By utilising Copilot, ACIAR aims to enhance operational efficiency, improve productivity and support collaborative efforts, while maintaining a high standard of quality and accuracy.
Monitoring and governance
Bias can arise from data selection or software models, leading to skewed results. AI might produce false information and/or result with privacy risks. It may also provide outdated or inaccurate data and could diminish individual creativity, resulting in homogenised content.
Monitoring and governance of AI systems are crucial to mitigate the risks associated with bias, false information and privacy concerns. Establishing robust frameworks that include continuous monitoring, regular audits and compliance with established policies can help ensure the ethical and responsible use of AI. Transparency in data sources, algorithmic accountability and stakeholder involvement are key components of effective governance. By implementing these measures, we can foster trust and integrity in AI applications, ensuring they deliver accurate, reliable and fair outcomes.
ACIAR acknowledges the rapid evolution of AI technology and commits to continuous monitoring to ensure that policies, tools and their usage remain current. Internal governance mechanisms will be employed to oversee these developments.
Compliance
We will use the DTA’s AI assurance framework to review the impact of AI on ACIAR’s strategic footprint. ACIAR will also adhere to any relevant legislation and regulations governing AI across Commonwealth agencies, such as the DTA's policy for responsible AI in government.
