Webinar Explores the Use of AI in Emergency Medical Services

The webinar aimed to foster an open discussion on the role AI could play in enhancing the EMS/EMT sector, focusing on both its challenges and opportunities in improving emergency response and citizen safety.

Agenda and Key Speakers:
The session covered several key topics, including insights from the KnowEMS project, a comprehensive report by the European Emergency Number Association (EENA) on the use of AI in Public Safety Answering Points (PSAP), and contributions from both researchers and practitioners in the field.

Notable speakers included:

• Mr. Peter Lonergan, EENA, who presented the organization's report on AI in PSAPs.
• Prof. Claudia Goldman-Shenar and Dr. Yuval Bitan, leading researchers in AI and emergency services.
• Mr. Thomas Seltsam and Mr. Chaim Rafalowski, practitioners with hands-on experience in emergency response.
• An interactive open discussion session followed, encouraging feedback from all participants.

Main Discussion Points:

• AI's Role in Enhancing Safety and Emergency Response:
  Participants agreed that AI has significant potential to improve both the safety of citizens and the effectiveness of responses to emergency situations. However, the unique and chaotic nature of emergencies requires AI tools to be specifically tailored for such environments.
• Challenges of Applying AI to Emergency Situations:
  AI systems designed for more structured environments often struggle in chaotic, noisy, and dynamic emergency scenarios. For instance, voice recognition tools must contend with loud, disorganized environments where language is not always clear. Similarly, AI tools designed to count casualties in mass casualty situations face difficulties distinguishing between victims and bystanders, especially when individuals are moving.
• Data Challenges:
  Machine learning algorithms require large and high-quality datasets, but such data is currently scarce. There are gaps in both clinical data (patient records) and operational data (such as recordings from emergency scenes and resources' locations). This lack of data hinders the development of AI tools that can be applied in real-time emergency situations.
• Sensitive Data and Privacy Concerns:
  Emergency services manage highly sensitive information, and the use of AI in this context is considered high-risk under the EU's AI Act. Participants stressed that AI tools need to be designed with data protection in mind, particularly since sensitive information cannot be uploaded to the cloud where AI systems typically operate.
• Synthetic Environments for AI Training:
  Creating synthetic data to train AI systems that accurately reflects real-world emergencies is both complex and costly. Without realistic simulations, there's a risk that AI systems may fail to adapt effectively to actual emergency situations.
• Collaboration Between Researchers and Emergency Experts:
  Effective integration of AI tools into emergency services requires strong collaboration between researchers, industry leaders, and emergency professionals. All parties must dedicate sufficient resources for ongoing development, and extended trial periods should be incorporated to allow for iterative feedback and improvement.
• Adjusting Expectations and Validation of AI Tools:
  Developers must align expectations with users regarding the readiness of AI tools. Some technologies are still in the early stages of development and are far from market-ready. Validation of these tools in operational settings is crucial. Additionally, trade-offs between speed and accuracy, such as in translation, need to be considered.
• AI as Part of the System:
  AI tools should be seen as components of the overall emergency response system. Certain outcomes, like noise cancellation, can sometimes be better achieved by other means, such as the phone’s built-in capabilities, rather than relying solely on AI.
• Studying Human-AI Interaction:
  Understanding how humans interact with AI tools, especially in stressful and high-pressure situations, is essential for improving their effectiveness in emergency contexts.
• Integrating Patient Data for Critical Care:
  The integration of vital signs from patient monitoring devices with data from exams and treatments is key to advancing AI-driven machine learning in critical care. However, challenges remain in attaching devices to casualties in a way that supports the use of machine learning algorithms.
• Formalizing Critical Care Assessments:
  Certain aspects of critical care, such as trauma assessments, need to be formalized into standardized scoring systems to facilitate their integration into AI algorithms.

Overall, the webinar highlighted both the promise and challenges of AI in emergency medical services. To achieve meaningful progress, collaboration between various stakeholders, thorough testing, and a focus on data protection will be crucial in shaping the future of AI in emergency response systems.