Artificial intelligence offers significant potential to enhance emergency and crisis management, in certain situations. AI can be understood as an ‘umbrella term’ that includes a diverse set of technologies, methodologies and applications, including machine learning, computer vision, and natural language processing. AI can support situational awareness, forecasting, damage assessment, and decision-making throughout the disaster risk management cycle of prevention, preparedness, response and recovery. But there is as yet no comprehensive, system-level assessment of the risks of using AI for crisis management in the EU context. Also, acute crisis phase (response) poses additional challenges to the deployment of AI tools, both ethical (life-and-death decisions need to be made with limited information and under extreme time pressure) and infrastructural (AI tools’ performance is dependent on network resilience and, in the specific European case, AI must reflect the reality of cross-border operations and multi-national crisis across Europe).
Evidence suggests that AI performs best on standardised, data-intensive tasks that are typical in frequent disasters such as floods, wildfires and droughts. AI is good at repetitive tasks that may be tiring for humans, such as continuous environmental monitoring, which is important for early warning systems. It can also be effective in damage assessment and social media processing, performing at scales and speeds that are impossible for human analysts. It is not well suited to interpreting highly heterogeneous contexts, and in new situations where it lacks appropriate training data. Moreover, morally challenging decisions and trade-offs should not be referred to an AI tool.
AI tools must uphold human dignity, transparency and responsibility, whilst meeting European standards for safety and ethics. Careful monitoring is required to ensure compliance with legal frameworks, avoid algorithmic biases and maintain meaningful human control, where people are ultimately in charge and are thus responsible for AI and any decisions made with it.
Other policy options include developing benchmarks, practical guidelines, codes of conduct and sandbox environments for AI in crisis management, which would allow the testing of AI under supervision and with ethical oversight, prior to full deployment.
An assessment of the risks associated with the use of crisis management AI tools, covering areas such as cybersecurity, AI resilience, and the EU’s strategic autonomy in this area is recommended by the Group of Chief Scientific Advisors. The assessment should be comparative in nature, evaluating the added value of AI tools for different uses against other options. The Advisors call for an inventory of AI tools, particularly those already in use.
