Responsible intelligence: ethical AI governance for climate prediction in the Australian context

Abstract As artificial intelligence (AI) becomes increasingly integrated into climate prediction systems, questions of ethical governance and accountability have emerged as critical but underexplored challenges. While international frameworks provide general AI governance principles, their application to environmental science contexts remains limited, creating potential gaps in oversight of high-stakes climate prediction systems. Given Australia’s absence of mandatory AI governance for climate science, this study investigates the ethical and governance challenges associated with AI-driven climate prediction, examining how stakeholders navigate these challenges without formal frameworks, and proposes a tailored governance framework for responsible AI deployment in environmental contexts. A qualitative research design was employed, combining 24 semi-structured interviews with stakeholders across government agencies, academic institutions, and non-governmental organizations, supplemented by three focus group discussions and analysis of 47 policy documents. Data were analysed using thematic analysis to identify patterns in ethical concerns, governance approaches, and stakeholder priorities. Three key findings emerged: (1) AI interpretability challenges manifest differently across sectors, with government prioritizing policy communication, academics focusing on technical validation, and NGOs emphasizing public understanding; (2) XAI implementation remains fragmented, with significant gaps in bias mitigation particularly among government and academic institutions; and (3) ethical frameworks vary substantially across sectors, creating concerning blind spots in stakeholder impact consideration and regulatory oversight. Current AI governance approaches in Australian climate prediction are inadequate for managing the risks and responsibilities associated with high-stakes environmental decision-making. Government institutions demonstrate concerning regulatory complacency despite high confidence in AI systems, while bias mitigation receives attention primarily from resource-constrained NGOs rather than technical institutions. The study proposes a four-pillar governance framework emphasizing institutional coordination, technical standards, participatory governance, and adaptive management. This framework addresses identified gaps while accommodating Australia’s federal structure and Indigenous knowledge systems. The findings contribute to emerging literature on algorithmic governance in scientific contexts and provide practical guidance for developing responsible AI practices in environmental applications.