Introduction
The Pacific Disaster Center (PDC) provides insight and innovative tools to support early warning, emergency preparedness and disaster risk reduction. Its DisasterAWARE platform provides alerting for hazards and tools that support decision makers to prepare for and manage disaster events. DisasterAWARE is augmented by the PDC’s AI for Humanity which not only enables a much faster processing of huge amounts of hazard monitoring data, but is also capable of recognising patterns, creating new data and providing additional insights. It utilises a variety of AI technologies – Machine Learning (ML), Natural Language Processing (NLP), and also more recently generative AI and Large Language Models (LLMs).
Background
For more than 28 years, PDC products and analytical capabilities haveaided response to some of the largest disasters and supported governments and nonprofits worldwide in the shared mission to save lives and reduce disaster risk. Through the DisasterAWARE platform, PDC provides not only real time alerting for 28 natural, biomedical and geopolitical hazards, but also additional tools that help the general public as well as disaster management decision makers to prepare, assess the potential impact, and plan their response. With AI for Humanity, the PDC is looking at where AI technology can be useful to enhance existing capabilities. Rather than having AI be a purpose in itself, the idea is that it can underpin and enhance what the PDC does.
AI development
The PDC’s initial reason for starting to use AI technologies and having AI for Humanity underpin many of its offerings was scale.
“[…] we're covering the globe. And we're harvesting all of these different types of information about early warnings […] How can we scale that in a meaningful way meaning that for instance, when we analyse the data, we could actually […] develop advanced analytics to try to guesstimate how many people might be impacted, what infrastructure might be impacted, and all of that” says Ray Shirkhodai, the PDC’s Executive Director.
To offer uniform and as complete as possible information about global disasters, deep learning and machine learning algorithms are used to analyse and forecast, delivering a level of speed and accuracy that would need an army of humans to do the work manually.

PDC also utilises AI’s ability to process natural language and employs NLP and generative AI to support the finding, analysing and curation of information on hazards that would otherwise be difficult to detect, but may be contained in difficult to read scientific documents, or have been reported on news websites. Relevant information is extracted and made available through DisasterAWARE, thereby enhancing the coverage of hazards and related information around the world.
For generative AI, Ray Shirkhodai thinks it will also be able to make a difference in a third area - communication with the general public, informing them of any hazards that may impact them and to allow them to prepare. The ability to communicate in many different languages will make DisasterAWARE services accessible to anyone with a device, whether they are fluent in English or not. This would also support the efforts of the Early Warnings for Allinitiative.
Operationalising AI
The PDC works continuously to assess the use of technology for its mission, and collaborates with other institutions that contribute expertise, research, and data to further enhance its offerings, for the benefit of the currently over 25,000 users and over three million individuals that have downloaded the DisasterAWARE app. Currently the team is testing an LLM-powered “expert system” that can support disaster management professionals by pulling together relevant information and providing recommendations, helping them to navigate the many different sources of information and layers of data that are available to them. Through a feedback function, the users will be able to provide additional information and correct any mistakes and thus improve the quality of the output provided.
Ray Shirkhodai has also for a long time pondered the idea of a “pocket expert” that would not only alert both disaster management professionals and the public to any hazards but also provide them with advice on what actions to take. In his view, these could be one of the most impactful uses of AI in the realm of disaster awareness. And with generative AI and LLMs this may now be possible – an app would be able to provide information and targeted advice to specific personas, taking into account the nature of the hazard, the actual situation at a specific location, the language spoken by the user, and so on. The PDC team is conducting a proof of concept for such a “pocket expert” with three different personas – a disaster manager, and general public fluent and basic English speaker. Also with this application, a lot of the work goes into ensuring that the narrative delivered to the audience is correct by providing a welldefined set of parameters to the LLM to avoid hallucinations and ensure the output is actually useful and relevant.

Learnings
The PDC's scope is global so being able to scale any solution while keeping the cost manageable is important. For their use of LLMs, the team is taking a phased approach, looking at the various ways to control cost, and building proof of concepts which, when showing to be useful, can easily scale when infrastructure and resources are available. In general, the most cost-efficient architecture seems to be to utilise open source LLMs run on in-house infrastructure.
While he is excited about the potential of generative AI, Ray Shirkhodai is also clear that its application needs to be handled with caution, and in line with humanitarian principles. To him, it is essential that while AI may make suggestions, it must be the human that ultimately takes the decision. This is particularly important in disaster management where the lives of many may depend on the right decision being taken.
“It's a very […] fine line. Pushing the technology but being cognisant of its potential unintended consequences”
Plans for the Future
The team is hoping to deploy the pocket expert this year, at least with the basic English persona, to test it in real live and get feedback that will help to fine tune the underlying models. If the application is demonstrating its value, the PDC will look at finding partners, potentially even in the private sector, to scale and roll out broadly.
Ray Shirkhodai can think of many other functions the pocket expert could offer. It could provide information based on the phase the disaster is in - whether it is imminent or whether there is still time to prepare. For disaster managers, it could help cascade information from larger areas to local places, adapting to what is relevant for the managers on each level. The PDC will continue to explore the possibilities of AI use for this and other, future applications.
Where to Learn More
For inquiries or to learn more about PDC’s AI for Humanity program, email media@pdc.org
Web page: AI for Humanity