AI Can Help in Search and Rescue, But Collaboration with Humans is Key

Strategic Insights -

In the aftermath of natural disasters like flooding, search and rescue teams face a daunting task: quickly scouring vast areas to find survivors. Artificial intelligence (AI) can play a crucial role in this process, but a collaborative approach between humans and machines is essential. Recent studies have shown that AI can rapidly scan drone imagery, identifying areas of interest and prioritizing them for further inspection. However, current implementations of AI fall short, and the technology can be prone to producing false positives.

Key Takeaways:

  • Machine learning systems can scan drone imagery in under one second, compared to one to three minutes for a human.
  • Drones can produce up to 8,000 high-resolution images in a single 20-minute flight, overwhelming human responders.
  • Current AI systems are not up to the task, often producing too many false candidates, which can overload search teams.
  • The key to success lies in AI-human collaboration, where AI scans and prioritizes images, while human responders inspect and validate.
  • Human remains may be obscured, camouflaged, or entangled in debris, making it challenging for AI to detect.
  • Training datasets for AI are limited, and existing classifiers may miss visual indicators of victims in aerial imagery.
  • The imprecise GPS location of candidate areas requires extra time for search teams to verify.

Statistics:

  • 800 high-resolution images can be produced in a single 20-minute drone flight.
  • 10 flights can result in over 8,000 images, requiring extensive human effort to inspect.
  • AI can spend less than one second scanning an image, while a human responder can take up to three minutes.
  • The average response time for a human responder is 10 seconds per image.

Sources:

  • (The Conversation), "For search and rescue, AI is not more accurate than humans, but it is faster"
  • (The Conversation), "Recent successes in applying computer vision and machine learning to drone imagery"
  • (The Conversation), "Developing computer vision and machine learning systems for finding flood victims is difficult"