Can people and AI cooperate in disaster response across borders?
On Tuesday, October 8, 2019, at 9 a.m. (JST) and 7 a.m. (IWST) (i.e., 0 a.m. (UTC)), the world's first “International Cyber Disaster Response Drill” will be held simultaneously in Ehime Prefecture, Banda Aceh City of Indonesia, and at the University of Syiah Kuala.
The drill is designed in the CREST CyborgCrowd Project, which is a collaborative research project involving University of Tsukuba, the University of Toyama, and Kyoto University, based on the experience of the West Japan Heavy Rain Disaster (2018 West Japan Flood). The aim is to examine the possibility of prompt understanding of the damage situation (flooded area) through the joint work of people across borders and AI, making the best use of the latest knowledge and technology of Japan and the world.
Currently, large-scale natural disasters occur frequently worldwide, causing damage to tens of thousands of people. In those situations, it is necessary for both the national and local governments to investigate and summarize the situation of the damages, but grasping the whole situation is not always an easy task.
In the areas damaged by a disaster, while rescue operations are ongoing, there are also many activities to support the evacuation procedures and the survivors in many ways, involving the participations of many volunteers. However, it is possible to support the affected areas from a distance. This kind of support can be provided by everyone little by little, or it can involve people who can provide specialized work such as AI development.
Both Ehime Prefecture and Banda Aceh City have experienced wide-scale disasters.
In the past, when a wide area was hit by a disaster, it was difficult to identify the damages, and to fully grasp the entire situation.
Based on their hard experience, Ehime Prefecture and Banda Aceh City will collaborate internationally to implement a new disaster response drill for the realization of prompt damage assessment.
During the training, starting from participants in Japan and Indonesia, many others will be recruited worldwide. Using the aerial photographs of the 2018 West Japan Flood, their aim will be the one of uncovering the inundated area as quickly as possible by bringing together the collective power of people. In addition, we will use AI that was developed through open recruitment and pursue its possibilities.
Specifically, participants from all over the world will judge the flood situation from satellite images and aerial photographs. At the same time, AI learns the judgment results and estimates the total damages. The results of human assessments and AI estimation are put together and send back to the affected area (Ehime Prefecture) to speed up decision making for disaster response. In other words, this represents a new form of support that consolidates the power of remote participation of people and the power of the latest AI technology to collect data on the disaster situation and return them to the disaster areas.
In the future, we aim to break through the “72-hour wall” where the survival rate is extremely lowered by utilizing these technologies.
Thanks to ICT (Information and Communication Technology) and AI, it is becoming possible to respond to natural disasters where citizens from all over the world collaborate remotely across borders. In this disaster drill, we will examine the application of this new type of natural disaster response that was impossible before.
a) Send images of the affected area (aerial images of the 2018 West Japan Heavy Rain Disaster) to the server.
b) Based on the acquired image, the server generates a number of tasks (small tasks) for grasping the situation and distributes them to the world.
a) For the tasks distributed on the web, people all over the world perform “inundation judgment”.
b) At the same time, AI will detect the image and perform “inundation determination”.
a) The work results of human and AI are integrated and expressed on a single map. In addition, the progress of work and the degree of AI involvement can be visualized using graphs.
b) These maps are updated in real time, and are always distributed and projected in the affected areas.
a) In disaster areas, disaster management staff will examine the extent of damage, what kind of response is required, how much personnel and equipment are required, etc., based on the “inundation judgment result”.
b) Since this is a training, we will review the possibilities of our approach, and summarize the expectations, requests, and issues for the CyborgCrowd Project.
c) Mention the possibility of international collaboration and its usefulness at the global level.
※ Participation: University of Tsukuba, University of Toyama, Kyoto University, Niigata University, CNRS（France）
Cooperation: Yahoo Japan (Yahoo! Crowdsourcing), Geographical Survey Institute, University of the Philippines Open University.
In addition, we received advice from Mr. Itsuki Noda of AIS.
The demonstration experiment will be conducted mainly with the support of JST-CREST “Development of Intelligent Information Processing System that Realizes Creative Collaboration in Harmony with Humans” (Research Supervisor: Norihiro Hagita).
In the task, you will see a picture of a part of the disaster area and judge the situation.
(1)“The whole area is non-flooded”
(2) “The whole area is flooded”
(3)“Partially Flooded / Partially covered by clouds / Not sure”
(4) “The whole area is covered by clouds”.
Only select (1),(2),(4) when all of the areas apply to it.
For example, only select (4)“The whole area is covered by clouds” if the shown image is completely covered by clouds.
Select (3)“Partially Flooded / Partially covered by clouds / Not sure” if some parts of the image are flooded or covered by clouds, or if you are not sure.
For example,the image above may appear to be flooded, but it is a mixture of flooded and non-flooded areas.
In this case select (3)“Partially Flooded / Partially covered by clouds / Not sure” instead of (2)“The whole area is flooded”.
Only select (2)“The whole area is flooded” when the shown image looks like the above.
The image is a mixture of clouds and non-flooded areas.
In this case, select (3)“Partially Flooded / Partially covered by clouds / Not sure”.
Only select (4)“The whole area is covered by clouds”when the shown image looks like this.
＞＞ Please let us know what you think about this project.
Goto Feedback Form
| Name | Affiliation | Research field |
|---|---|---|
| *Kosuke Uo | University of Tsukuba | Crowdsourcing and AI |
| *Masaki Kobayashi | Doctoral Programs in guraduate school of Tsukuba | Crowdsourcing and AI |
| Munenari Inoguchi | University of Toyama | Natural Disaster& IT |
| Itaru Kitahara | University of Tsukuba | Computer Vision |
| Kouyou Kobayashi | Master's Programs in guraduate school of Tsukuba | Computer Vision |
| Hidehiko Shishido | University of Tsukuba | Computer Vision |
| Keishi Tajima | Kyoto University | Data Engineering |
| Keiko Tamura | Niigata University | Natural Disaster Responses |
| Hisatoshi Toriya | Doctoral Programs in guraduate school of Tsukuba | Computer Vision |
| Flavia Fulco | Tohoku Univeristy | Natural Disaster Reconstruction and Culture |
| Masaki Matsubara | University of Tsukuba | Cognitive Science, Human-Machine Intereactions |
| Atsuyuki Morishima | University of Tsukuba | Crowdsourcing and Human Computation Systems |