Missing link: AI and rescue workers – how artificial intelligence helps helpers

Contents

The warnings about the potential flood disaster in the Ahr valley in the summer of 2021 and the subsequent crisis management are considered suboptimal , which cost the then Rhineland-Palatinate Interior Minister Roger Lewentz (SPD) his job. For example, videos from a police helicopter squadron from the evening of July 14, 2021, were considered lost meanwhile, whereupon it became apparent early on that entire houses, caravans and people had been washed away by the masses of water. Emergency services are therefore now increasingly relying on systems with artificial intelligence (AI) to detect dangerous situations with weather events such as heavy rain and flooding or fires at an earlier stage, inspect them in near real time as they occur and respond appropriately.

For authorities and organizations with security tasks (BOS), i.e. the emergency services with blue lights, everyday life is often hectic. Maintaining an overview can quickly become a challenge. More and more data and technical aids such as computer programs and, finally, AI applications are available to help rescuers with their tasks. Some solutions have already been launched and many are still in the pipeline. However, as sensitive personal information is often used as a basis, there are also potential pitfalls, for example due to the General Data Protection Regulation (GDPR). If risks are associated with the use of the technology, the new AI Regulation must also be considered.

At a recent online conference on "AI helps helpers" organized by Behörden-Spiegel, experts agreed that the greatest potential of artificial intelligence in the BOS sector lies in situational awareness. The technology is predestined to "get an overview of the scene as quickly as possible", explained Sirko Straube, Deputy Head of the Robotics Innovation Center at the German Research Center for Artificial Intelligence (DFKI). This involves, for example, evaluating images and data from drone cameras or body sensors (wearables) together in the control center and then drawing the right conclusions and networking with other emergency services.

Early warning systems with AI

Katharina Weitz, Project Manager at the Fraunhofer Heinrich Hertz Institute (HHI) in Berlin, cites the EU-funded Tema (Trusted Extremely Precise Mapping and Prediction for Emergency Management) project, which aims to improve crisis management in the event of natural disasters, as an example of early warning systems with AI. The HHI is primarily responsible for human-interpretable explanations, AI-generated predictions and recommendations. As part of Tema, heterogeneous data sources such as drones, sensors and satellites are to be evaluated, including topographical information. In addition, there is AI-based object recognition, which should be able to differentiate between pedestrians, bicycles or motorized vehicles and identify phenomena such as mass panics, sources of fire and fires.

Weitz explains that users need to know why the model has classified objects in one way or another. The team relies on Layer-wise Relevance Propagation (LRP) here. This involves interpreting individual predictions by assigning values that quantify the significance of the individual input features for the prediction. Using the implemented neural network, "we send the image backwards again to generate a visual explanation", the computer scientist explains. As a result, individual relevant pixels for the assessment are highlighted in color. These are then signs of smoke and fire, for example. The AI also provides "learned concepts with examples". This is ideal for early warning systems.

There are already models for emergency services "that are based on explainable AI", says Simon Franke, team leader for research projects at the Information and Technology Center (ITC) of the Rhineland-Palatinate branch of the German Red Cross (DRK). The center uses such a system for emergency call support. "It can be rule-based," says the practitioner. For example, the signs mentioned by the caller that could indicate a heart attack can be brought together.

"Everyone is a bit scared of the AI Act"

In general, Franke also "dreams of" using AI to obtain a uniform picture of the situation at an early stage. The ITC is initially concentrating on "integrated control centers" to achieve data exchange, including with partners in industry such as BASF. Due to the large number of systems in use, the basis for using AI is often still lacking. The order of the day is therefore to break up isolated solutions with standardized tenders and create interfaces. The Federal Ministry for Economic Affairs and Climate Protection (BMWK) is now demanding a uniform data standard for funding - in line with this.

However, Franke does not believe that there is a need for specific AI products for disaster control. Existing technologies already help with live translation, image recognition and the evaluation of traffic cameras, for example regarding the transportation of hazardous goods. There are also already AI-supported solutions to reduce peak deployment times in the event of major emergencies. However, there are often legal concerns before a deployment, as "many things are complicated by data protection". "Everyone is a bit scared of the AI Act," reports the platform manager of the Spell project for innovative technology in rescue control centers. The new regulation is currently creating "a lot of uncertainty".

Franke also points to technological hurdles when using AI: "We can't just attach ourselves to any cloud provider." It is necessary to operate infrastructures for the computer clouds yourself. This leads to "entirely different hardware requirements" than for private users. It is also necessary to convince employees, authorities and ministries. Many employees ask themselves: "Will I be abolished or controlled by AI?" The advantages of the technology should therefore be clear. It must also be easy to use and must not disrupt the workflow of dispatchers in particular. On the other hand, the networker is not particularly worried about mobile network failures such as those in the Ahr valley - radio relay and satellite communication are available.

AI projects for control centers sprout from the ground

The AI in Rescue Chains (Aircis) project, which is funded by the German Federal Ministry for Digital and Transport, has also set itself the task of making the work of rescue services easier with the help of AI. The starting point here is that there are currently no tools to simulate or plan the rescue chain under the influence of extreme weather events. The aim is therefore to use AI to forecast the volume of operations based on real data from the Cottbus control center and to develop a simulation to map the entire rescue chain using a digital twin. In the event of heavy rainfall or during periods of heat, for example, it would then be easier to calculate different routes.

In addition to Tema, the HHI alone is involved in another EU funding initiative, MedEWsa (pronounced Medusa), which aims to use AI to facilitate the prevention of natural hazards ranging from fires and heatwaves to volcanic eruptions. A German variant of this is the Daki (AI for Disaster Early Warning Systems) project, which runs until the end of 2024 and, according to Weitz, will create a dashboard and interfaces to inform industry, the public and politicians in good time. Daki is part of the BMWK's AI innovation competition and is being subsidized with around 12 million euros. One aspect of such initiatives is to use satellite information from the European Space Agency (ESA) to predict the risk of forest fires and flooding.