April 21, 2017
In the immediate aftermath of a disaster an array of critical needs emerge: food, water, shelter, and the ability to locate loved ones. Another urgent need, which may be less evident to those outside of civil, government or military response entities, is for information. Yet without information, responding to urgent relief needs is difficult or ineffective, as priorities cannot be identified. Further, in order for information to be actionable intelligence it must also be verified and reliable.
After a disaster, all that was known about an area, settlement or community prior to the disaster event is now secondary data – it may no longer be true, and until it is verified it cannot be reliably acted upon. Added to the challenge of now potentially inaccurate knowledge about the situation prior to the event, are sudden gaps in information: where has damage been sustained? Where are on-going hazards such as flooding, instability or contamination as a result of the disaster?
In the short period following a critical event, but during the phase where lives may still be saved, this data void is a key challenge. Conversely, responders at both strategic and operational level are increasingly seeing a data overload occurring, particularly from social media and news outlets. The challenge here is how to turn this glut of rapidly created data from disparate sources into actionable intelligence.
At Rescue Global we put significant emphasis and effort into tackling the challenge of being able to gather, speed up and verify information following a critical event in order to create actionable intelligence that can be used to support prioritisation and action that saves lives in disasters. Our goal is to achieve a Commonly Recognised Information Picture (CRIP) that integrates multiple data sources and can be shared across diverse responding agencies. In order for this to be effective, it must be rapidly stood-up following a disaster, must be able to include non-homogenous data sources (for example, images and text), and must include the ability to attribute data sources and assign them with provenance and reliability scores.
We have been working with partners in academia and industry on both real world and pilot projects to develop this concept, using Machine Learning (ML), Artificial Intelligence (AI) and Human Agent Collectives (HACs) to manage big data, and speed up and improve decision-making in disasters.
These tools were used in a real-world situation following the 2015 earthquakes in Nepal, where Rescue Global and academics from the Orchid Project took pre and post-disaster imagery, utilised crowd-sourced data analysis and machine learning, to identify locations affected by the quakes that had not yet been assessed or received aid. This information was integrated into “heat maps” to visualise humanitarian response priorities based on this information, and shared with SAR groups to facilitate their decision-making and activities. This work was recently showcased at the World Economic Forum 2017.
A recent pilot project, funded by the Defence Growth Partnership (DGP) Innovation Challenge, to further this work was in collaboration with BMT Defence Services Ltd. and University of Oxford. The project investigated methods to properly vet and then exploit information generated from crowd-sourced data. The project used recent developments in machine learning and data provenance to build “trust” in crowd input, increasing the utility of the crowd as a resource. The pilot used a large volume of satellite imagery from a disaster area, allocated tasks to humans and machines to review the data, and then aggregated and scored this information to create “heat maps” identifying infrastructure damage.
The project also examined optimal task allocation between human subject matter experts (who will likely be limited in number), crowd-sourced data analysis input (the “crowd” is likely to consist of a large number of non expert individuals) and Machine Learning tools. Collectively, these are Human Agent Collectives. Each resource is best suited to different tasks, but bringing them together significantly increases the rapidity, volume and accuracy of information. This information, when Artificial Intelligence is applied to assign provenance and credibility, and exclude erroneous information, becomes actionable intelligence.
The potential for such technology in the realm of disaster response is undeniable and truly exciting. Ultimately, as it is further developed and fielded, it will be a tool to give disaster response leaders an informed, clearer, single-view of a disaster area, which will ultimately save lives in a crisis.