Crises such as pandemics, armed conflicts, or earthquakes force decision-makers to act quickly under high stakes and deep uncertainty. This makes them a natural test case for AI-based decision support. But using AI here raises a prior question: does it actually improve decisions, or does it introduce new risks? 

Crisis situations involve difficult trade-offs, and it is often unclear which trade-offs an AI system is making, whether it has the right information to make them, and whether decision-makers know what to expect from its advice. 

In this workshop we will discuss how AI can be used responsibly in crisis situations. We will start off with a report of our experiences modeling policies for the COVID and Ebola crises. Based on those experiences, we will discuss the following four aspects:

• Data and models: Do we have the right information to create models or train AI systems for crisis settings?

• Guarantees and validation: What guarantees can be given on the output of the AI system?

• Preparedness and adaptation: Can we prepare systems before a crisis happens, and adapt them to the specific situation as it unfolds?

• Project organization: How do we organize the process of building an AI system during a crisis, under time pressure and with limited resources?

Participants are invited to register their interest in one or more of these four topics. When submitting an expression of interest, please indicate which topic(s) you wish to contribute to and how, for example through a case study, a methodological perspective, or a position on open questions. Contributions will be grouped by topic to structure the workshop's discussion sessions. 

Due to the highly interactive program, the workshop will be in-person only. 

The workshop is co-located with the AIES conference and is free to attend for accepted participants. 

• Paul Davidsson, Malmö University

• Frank Dignum, Umeå University, contact person: dignum@cs.umu.se

• Bertilla Fabris, Umeå University