Paper details

Paper authors	Josee Poirier
In panel on	Towards responsible use of AI and geospatial data in preparedness and response to natural hazards and complex emergencies
Paper presenter(s) will be presenting	In-Person / Online

Abstract

The Peer Review Framework for Predictive Analytics in Humanitarian Response was developed to help translate the outputs of predictive models into timely, responsible, appropriate humanitarian action. It draws on best practices from academia and the private sector to offer advisory support through standards and processes for the development and implementation of predictive models in the humanitarian sector. It offers a terminology that facilitates communication and connection between communities, humanitarians, technologists. The Framework provides guidance on the evaluation of three aspects of a model: technical rigor, operational readiness, and ethical considerations. In particular, the proposed ethical review process aims to: help identify key issues (17 potential issues are raised in a non-comprehensive list) associated with a model and its intended use in a specific context; assess the related ethical concerns of all stakeholders; and offer recommendations for how to address these concerns. Model evaluations using this Framework result in thorough characterizations of specific applications of models, which increase transparency and accountability while reducing the risk of harm. We propose to present the Framework, share lessons learned, and leverage the Framework as a use case to collaboratively discuss gaps in best practices towards the responsible use of technology in the humanitarian sector.

Back