Last week on 23-24 May, VIEWS Director Håvard Hegre, Outreach and Operations Manager Angelica Lindqvist-McGowan, and PhD Candidate Maxine Leis had the pleasure of attending the EconAI workshop “Growing Together: Prediction, Prevention and Preparedness” at the Institute for Economic Analysis in Barcelona. The two-day event brought together representatives for foreign offices, UN agencies, INGOs, and researchers to discuss how AI-based forecasting systems for conflict and natural disasters can be leveraged to assist anticipatory action in the humanitarian sector.

Key take-aways from the lively discussions:

• There is growing interest, momentum, and capacity for incorporating predictive analytics into daily operations in the humanitarian sector.
• Predictive analytics is actively used to support preparedness and crisis response for natural disasters, with trigger-based funding release considered the gold standard for effective assistance.
• Integration of conflict forecasts into similar solutions is, unfortunately, a far more challenging task, by and large grounded in:
  • the perceived imprecision of the systems (and far higher expectations placed on AI systems than on traditional experts),
  • the tradeoff between aspirations for high performance and explainability of the models (and misaligned communication between “intuitive” and “analytical” staff within humanitarian organizations)
  • lack of political will and accountability for acting on identified risks.
• Nevertheless, VIEWS forecasts (along with ConflictForecast and ACLED Cast) are used internally by several humanitarian organisations to date – and there is great appetite for impact modelling that assesses the consequences of armed conflict on humanitarian outcomes, emphasising the potential of our own Societies at Risk, ANTICIPATE, and VIEWS-People In Need projects.
• In order for existing and future systems to be fully and accurately utilised, we need to be clear about the promises – and limitations – of AI-based prediction models.
• We need to continue an open dialogue on the ethical dilemmas related to conflict and conflict impacts predictions – be they qualitative or quantitative in nature – to ensure no harm is inflicted on the affected populations.
  • Open systems can help ensure accountability for acting on risk alerts , especially for conflict forecasting, but in some cases – such as displacement forecasting – the negative consequences (e.g. closing borders in anticipation of high refugee flows) may outweigh the positives.
• Open dialogues are also needed to ensure a mutual understanding of key terms used across the academic and humanitarian sectors, highlighting the common mismatch between the broad and inclusive terms used by humanitarian actors and the narrow, operationalised definitions applied by research teams.
  • “Conflict”, e.g. is most often operationalised by research teams in accordance with the armed conflict definitions offered by UCDP or ACLED, whereas humanitarian actors may incorporate both direct and indirect deaths, violent and non-violent conflicts, and broader systematic issues into their working definitions.
• Inter-sectoral collaboration and funding of continued academic research is key!

A huge thank you to the EconAI team for two incredibly insightful days and fantastic organisation!

Slides from the VIEWS presentation

Photo credit: EconAI