Leveraging AI and Big Data to Bolster Disaster Resilience
Every year, about 160 million people are affected by natural disasters. Although countless governmental, non-governmental and private organizations around the world work tirelessly to minimize the devastation resulting from such extreme events, an average of 90,000 die every year. In 2019 alone, natural disasters cost mankind 140 billion USD.
While the disaster management sector has made great strides in recent years, there remains considerable prospects for further enhancement of disaster planners’ and responders’ primary mandates- early identification of potential threats, disaster risk reduction, damage mitigation and aid delivery.
Microsoft defines Artificial Intelligence (AI) as “Computers and software performing tasks we typically associate with people, such as recognizing speech or images, predicting events based on past information, or making decisions. AI tools use data to learn a task, and they continue to improve at functions such as…answering questions by quickly finding relevant information in databases or documents, detecting patterns in data, making decisions about simple queries, and predicting someone’s behavior based on past conduct.”[i]
AI indeed holds massive potential to add value to the disaster management community’s capacity to identify threats, reduce risks, mitigate damage and streamline responses. “Disaster resilience is heavily dependent on information collection, analysis and dissemination, and those are areas in which AI tools made great advancements over the past two or so decades,”[ii] says Partnership for Public Service Vice President Katie Malague.
While the potential of AI to enhance disaster resilience is indisputably substantial, this opportunity is not risk-free. The current Disaster technology landscape is still in a nascent stage and many emerging solutions include imperfect algorithms. The hope, however, is that these incipient solutions predicate cogent alternatives which, in time, will supersede their more primitive predecessors.
To understand both the opportunities and threats that AI poses to disaster resilience around the world, it’s important to consider the potential advances alongside the credible risks and limitations of AI solutions in order to gain a pragmatic understanding of where we are today, where we want to be in the future, and what we need to drive in order to get there.
II. The Current Limitations of AI in the Disaster Tech Landscape
Current AI solutions are limited in scope. They’re typically limited to specific disasters or to a specific handful of response agencies. According to the World Economic Forum “many private-sector initiatives involve one or a few government or NGO partners, and focus on specific use cases, often in relative isolation from the larger disaster-relief community and without integration into established disaster relief protocols.”[iii] Future AI solutions should be integrable across entities, sectors, and IT systems.
Moreover, Disaster Planning and Response involves the merging of information from multiple sources (such as data from previous events, social media outlets, 911 and 311 calls, and evolving geological, epidemiological and atmospheric information). This is typically a complex and tedious process for humans to complete, as even the most seasoned professional has limited cognitive bandwidth. AI, however, holds the possibility to seamlessly merge data from all those different sources to help ensure disaster planners and responders don’t overlook important pieces of data that could bolster their service provision.
III. AI’s Potential Contribution to the Disaster Tech Landscape
AI enjoys great potential to enhance disaster management agencies’ capabilities to predict disaster pathways and impacts such as displacement and relocation patterns. By leveraging data from past events and merging it with geological, climatological and telecommunications and social media data, AI processors will be able to predict shifts in disaster pathways and expected impacts with unprecedented accuracy. The improvement of disaster planners’ and responders’ abilities to understand human behaviors during disasters is critical to enhancing future resiliency.
AI also holds promising potential to refine the way humans assess damages and population vulnerabilities. In the hours and days following a major disaster, emergency managers must swiftly understand the extent of damage to structures, roadways and human populations in order to implement an appropriate response. How can responders deliver aid to the areas that need it most, and how can they identify those areas? What is the best evacuation route for survivors? Which roads can be cleared and reopened and which are too damaged for immediate use? As with any major crisis- disaster managers are left to make major decisions based on incomplete information. Thanks to its unmatched ability to collate massive amounts of data from multiple sources, AI is positioned to quickly fill this information gap.
AI also holds potential to streamline relief aid planning and implementation. From assessing the extent of the need for essential aid such as food, water and other basic supplies (and where to allocate them) to more sophisticated services such as providing medical and mental health care, setting up temporary shelters and repairing critical infrastructure- AI solutions can drive lean relief aid planning as well as more effective and efficient delivery of goods and services.
IV. The Social Media Element:
“The convergence of social networks and mobile has thrown the old response playbook out the window,” Michael Beckerman, president and CEO of the Internet Association, told the House Subcommittee on Emergency Preparedness, Response back in 2013[iv].
The advent of social media has paved a two-way road for disaster managers and the public to provide and receive disaster-specific information both before and after impact. Emergency managers are able to share relevant information (such as weather reports, road closures, evacuation orders, and shelter locations) across social media platforms and can collect information from these platforms to guide appropriate resource allocation.
According to the United States’ Department of health and Human Services “In addition to building community relationships and setting expectations pre-disaster, planners can use social media to identify and monitor potential threats to public health, and communicate with residents about threats (e.g., infectious disease), pending incidents (e.g., severe weather), and the location and availability of services (e.g., shelters and points of distribution). Tools such as crowdsourcing (collecting information from a large group of people via the Internet) and data mining bolster these efforts.”[v]
Social media has already proven to be a valuable crisis communication tool for both disaster managers and laypeople alike. In addition to integrating data from multiple databases, optimal AI solutions will also integrate data from social media platforms.
V. The Risks of AI Proliferation
As with essentially every type of data analysis, AI solutions risk misrepresenting threats and damages. AI doesn’t always perfectly achieve what it claims to in theory and without established processes to verify the validity of data claims and algorithm methodologies- the risk of misrepresentation will persist.
An algorithm is only as accurate as the data it uses to create its predictions. If the data input into the algorithm is biased, the algorithm’s predictions will be biased as well. There is a clear and critical need to develop consistent methods by which data input processes and algorithm methodologies can be assessed and validated in the context of AI development.
Moreover, the lack of robust regulation in the realm of AI poses additional risks, namely the exploitation of information for political agendas and economic motivations, the outcome of which involves a major threat to the Disaster Management community- misinformation.
According to Forbes “Recent developments in artificial intelligence point to an age where it’s not just humanity that will be upgraded, it’s misinformation. Now we know AI contributes to the forgery of documents, pictures, audio recordings, videos, and online identities which can and will occur with unprecedented ease. We are unleashing an open-source toolkit of cybersecurity weapons that will complicate our online interactions.”[vi]
The lack of established laws, codes of ethics, and government and corporate accountability and transparency in the AI domain mean there is tremendous data at risk of being manipulated, accessed and applied in unintended manners (both accidentally by well-intentioned individuals and deliberately by malicious actors).
VI. The Resolution: integrable solutions that build Shared Situational Awareness
Situational awareness (SA) is the perception of environmental elements and events with concern to time and space, the comprehension of their meaning and the projection of their future status. Shared situational awareness means two or more people have commonly understood a given situation as well as the causative factors behind it and the implications of their effects. Developing shared SA requires lean information-sharing channels, effective coordination mechanisms and utilization of a common language within and across organizations.
These universal requirements of SA- information-sharing channels, coordination mechanisms and use of a common language- are crucial in the context of developing shared SA in a largely unregulated domain. Disaster Managers who can leverage these components to develop shared SA amongst their decision-makers will be better positioned to prepare for, detect, respond to and contain undesirable data analyses and usages, often before they escalate and produce untenable outcomes.
Developing shared SA across sectors (in terms of entities- government, private sector, NGOs and in terms of hazards- natural disasters, manmade disasters, infectious diseases, cyber and physical threats, etc.) is critical to realizing the full potential of AI and Big Data in Disaster Resilience. How can disaster managers drive this prospect?
a. Develop integrable datasets- from satellite, geo-spatial, telecom, social media and financial sources- that can be merged to produce interpretable, actionable insights.
b. Establish a multi-player, data-driven platform to support responses of all response agencies (government, private, NGO) by providing a landscape for visualizing shared SA.
c. Cultivate enhanced coordination between current initiatives, technological solutions and relevant entities across all levels (local, state/regional, national, international), through domain-specific associations/partnerships.
d. Establish standards and processes to validate AI functionality and data output validity – such as an algorithm review process to ensure AI processors meet specific standards (of accuracy) before they’re widely used.
The expression “the whole is greater than the sum of its parts” underscores an important principle in advancing AI capabilities in the disaster tech domain- no single entity or technological development will realize AI’s massive potential to bolster disaster planning and response; only through concerted cross-sectional efforts and a utilitarian approach to both cooperation and development can the disaster management community marshal an age of AI fruition.
Microsoft and the Partnership for Public Service acknowledge a financial incentive for embracing AI: “Increasing the use of AI in disaster response also demands that government officials shift to a more proactive approach, which could also prove to be financially beneficial.[vii]” Indeed, research run by the Federal Emergency Response Agency has shown that every $1 spent on mitigation saves an average of $4 in recovery costs[viii].
By embracing new technologies, increasing collaboration and data integration, and developing processes to enable consistent, validated algorithm output, entities in the Disaster Management domain will cultivate promising enrichments to the disaster tech landscape. Disaster Managers should lean-in to technological progress across sectors to be able to advance from the ubiquitous approach of reacting to the impacts of hazards to the proactive approach of foreseeing and managing future hazards (as well as opportunities) well before they materialize.