Host: David Riedman, creator of the K-12 School Shooting Database
Guest: Dr. Rodrigo Nieto Gómez, professor at the US Naval Postgraduate School and expert in criminal innovation, AI, and national security.
Articles referenced:
How does OpenAI's ChatGPT-4o assess school shooting threats compared to human police officers?
Did OpenAI create the best weapon detection software available with ChatGPT-4o?
During this episode we discussed:
Rodrigo’s unique background being born in Mexico and educated in Paris, yet he is an expert in US national security and defense policy.
Criminal innovation moves faster than the systems designed to stop crime.
ChatGPT-4o turned everything upside down with an omni-model that can analyze text, audio, and images. This is 10 years ahead of where experts thought AI would be today.
Standalone machine learning classification models (flagging guns on CCTV, hearing gunshots) are irrelevant now in the same way that paying for a voice transcription service is no longer needed.
ChatGPT is different from any existing single-purpose software model because it has been trained with 13 trillion pieces of data. The first generation of AI school security companies are now like Kodak when affordable digital cameras hit the market.
Unlike image classification models, ChatGPT can identify a toy gun being used in a squirt gun fight because it knows what water, toy, playing, smiling, grass, and kids all are. ChatGPT understands the entire context of the situation instead of searching for specific objects that might match training data.
Without any special training ChatGPT analyzed school shooting threats and gave scores comparable to the aggregate scores of 240 human experts. ChatGPT performed significantly better than most of the individual police officers!
Without being prompted, ChatGPT created a threat assessment model based on multiple factors including time, location, and specificity of the threat. ChatGPT also suggested emergency actions to take.
When humans hallucinate (give the wrong answer), we can’t really tell what happened or easily fix it. If ChatGPT gets something wrong, we can see where it went wrong and create a better prompt for the next response.
My biggest takeaway from talking to Rodrigo is that things that we thought were impossible for computers to understand are now possible, and this changes everything for the future of school security.
David Riedman is the creator of the K-12 School Shooting Database. Listen to my recent interviews on Freakonomics Radio and New England Journal of Medicine.
Share this post