Voices is a new mini-series from Humanitarian AI Today. In daily five-minute flashpods we pass the mic to humanitarian experts and technology pioneers, to hear about new projects, events, and perspectives on topics of importance to the humanitarian community.
In this flashpod, Jessie Pechmann, Humanitarian GIS and Data Protection Lead with Humanitarian OpenStreetMap, speaks with Humanitarian AI Today producer Brent Phillips about satellite imaging, GIS, and the uses of AI in assessing building damage. They touch on how different AI models and methods can produce wildly different results for the same area, highlighting the need for transparency and better validation practices, including humans in the loop providing local knowledge and oversight. They also discuss the importance of "data commons," the open, shared data resources that humanitarian organizations rely on, and the challenges of supporting them amid a shift away from traditional government funding, which risks data becoming "siloed" as funding moves toward philanthropic or paid-for services.
Substack notes: https://humanitarianaitoday.substack.com/p/jessie-pechmann-from-humanitarian
All content for Humanitarian AI Today is the property of Humanitarian AI Today and is served directly from their servers
with no modification, redirects, or rehosting. The podcast is not affiliated with or endorsed by Podjoint in any way.
Voices is a new mini-series from Humanitarian AI Today. In daily five-minute flashpods we pass the mic to humanitarian experts and technology pioneers, to hear about new projects, events, and perspectives on topics of importance to the humanitarian community.
In this flashpod, Jessie Pechmann, Humanitarian GIS and Data Protection Lead with Humanitarian OpenStreetMap, speaks with Humanitarian AI Today producer Brent Phillips about satellite imaging, GIS, and the uses of AI in assessing building damage. They touch on how different AI models and methods can produce wildly different results for the same area, highlighting the need for transparency and better validation practices, including humans in the loop providing local knowledge and oversight. They also discuss the importance of "data commons," the open, shared data resources that humanitarian organizations rely on, and the challenges of supporting them amid a shift away from traditional government funding, which risks data becoming "siloed" as funding moves toward philanthropic or paid-for services.
Substack notes: https://humanitarianaitoday.substack.com/p/jessie-pechmann-from-humanitarian
Sergio Coronado on Blind Spots in AI Safety and International Humanitarian Law
Humanitarian AI Today
45 minutes
2 months ago
Sergio Coronado on Blind Spots in AI Safety and International Humanitarian Law
This 100th Humanitarian AI Today episode focuses on blind spots in AI safety and aligning AI with International Humanitarian Law. Guest host, Andre Heller, Director of Signpost at the International Rescue Committee (IRC), speaks with Sergio Coronado, Chief Information Officer with NATO’s Support and Procurement Agency (NSPA), about important research that he is heading at the Luxembourg Tech School studying “blind spots” in AI safety at the intersection of artificial intelligence and International Humanitarian Law (IHL).
Dr. Coronado speaks in detail about his team's groundbreaking research, which tested leading AI models against codified rules of humanitarian law. The conversation delves into the chilling discovery that while models refuse obviously harmful requests about 90% of the time, they can still for example be prompted to generate malicious code for targeting civilian infrastructure like hospitals, contrary to IHL.
This dialogue moves beyond identifying the problem to explore tangible solutions, highlighting how simple interventions can dramatically improve AI's adherence to legal principles. It serves as a powerful call to action for the humanitarian and technology communities to bridge this dangerous gap and champion the development of AI that is not just powerful, but principled and fundamentally law-adherent.
Interview Notes: https://medium.com/humanitarian-ai-today/sergio-coronado-on-blind-spots-in-ai-safety-and-international-humanitarian-law-40b64590a119
Humanitarian AI Today
Voices is a new mini-series from Humanitarian AI Today. In daily five-minute flashpods we pass the mic to humanitarian experts and technology pioneers, to hear about new projects, events, and perspectives on topics of importance to the humanitarian community.
In this flashpod, Jessie Pechmann, Humanitarian GIS and Data Protection Lead with Humanitarian OpenStreetMap, speaks with Humanitarian AI Today producer Brent Phillips about satellite imaging, GIS, and the uses of AI in assessing building damage. They touch on how different AI models and methods can produce wildly different results for the same area, highlighting the need for transparency and better validation practices, including humans in the loop providing local knowledge and oversight. They also discuss the importance of "data commons," the open, shared data resources that humanitarian organizations rely on, and the challenges of supporting them amid a shift away from traditional government funding, which risks data becoming "siloed" as funding moves toward philanthropic or paid-for services.
Substack notes: https://humanitarianaitoday.substack.com/p/jessie-pechmann-from-humanitarian
