When AI Goes Wrong

If an AI crisis, such as an AI-enabled attack on a national power grid takes place tomorrow, governments will largely depend on everything they know about crisis communications which is based on natural disasters, pandemics and cyber attacks. These communication protocols have not been tested for AI incidents. Established principles exist across the following domains AI safety, AI governance, crisis communications, behavioural science and trust and misinformation research but this work is largely fragmented. Evidence from other high consequence emergencies suggest that untested communication approaches leave the public less informed and more vulnerable to misinformation. This project will test what can be transferred from other fields to serious AI incidents in the Global South context.

My research will identify useful practices and principles across multiple disciplines and then I will synthesize what works well for AI crisis communications. I will develop an initial adapted approach which covers communication principles, uncertainty management and trust building techniques.

The next step is designing realistic AI incident scenarios. I will then run a pilot - testing with participants recruited in Grenada including students, community groups and residents from different socio-economic backgrounds, levels of AI literacy and government trust levels. The pilot will include scenario-based workshops and surveys with 80-100 participants. I will compare different message structures, visuals and sequencing to help pinpoint which communications approaches improve understanding, trust and behavioural intentions.

I have selected Grenada as it is an underrepresented Small Island Developing State (SIDS) where AI preparedness and governance is still developing. SIDS inclusion will help broaden the research base beyond the US and Europe, which is needed as AI risk is global. Grenada has also experienced repeated high consequence emergencies e.g. Hurricane Ivan in 2004 and Hurricane Emily in 2005 requiring public communications in rapidly changing environments. I will partner with Professor Wendy Grenade, Associate Director and Professor of Political Science at St. George’s University in Grenada, who will help with participant recruitment.

A key part of this work is to build and refine communication approaches that work in settings like Grenada, where varying levels of institutional capacity, resources and levels of AI literacy are present.

The outputs from this work will include:

• A report which includes the pilot findings, a new conceptual framework and an assessment of what is applicable to broader settings, what requires adapting and what appears to be counterproductive.

• An operational playbook covering response sequencing, tested message templates and decision guidance for varied institutional settings.

I will be the lead researcher for this project. My skills in this space are rooted in practice. I led national crisis communications at the Department for Education during COVID-19 school closures and reopenings. Additionally, I was a Resilience Emergencies Division Duty Officer at the Ministry of Housing, Communities and Local Government, acting as the out of hours link between central government and local responders during severe weather and public health emergencies. Both roles highlighted how misinformation and distrust can easily take over and why getting it right matters.