Columbia University National Center for Disaster Preparedness (NCDP)
Climate Week in NYC: AI 4 Resiliency Panel at the Disaster Readiness Summit, hosted by Columbia University’s National Center for Disaster Preparedness (NCDP), located at Columbia University’s Teachers College, and led by the Global Enabling Sustainability Initiative (GeSI).
The Disaster Readiness Summit explored how communities, leaders, and practitioners can strengthen climate resilience through anticipatory action, artificial intelligence, and disaster readiness.
Jumptuit Founder and CEO, Donald Leka at Columbia University, National Center for Disaster Preparedness (NCDP)
AI 4 Resiliency Panel at the Disaster Readiness Summit: Transcript of Donald Leka's Remarks
“So, I think we have to be cautious when we use the term AI. It's an umbrella term... and when most people refer to AI, they're really talking about Large Language Models.
And Large Language Models are really historically focused, and so there's certain limitations. They do not really bring you to the present moment in time, and they don't engage in the type of quantitative reasoning that would enable us to really be able to predict events as they occur.
Of course, there's a lot of hype now with the IPOs that are coming out as well, so it's important to keep that in mind. What I would say is look back at the genome research that was done, look at what people were saying about that 30 years ago. And what they were saying was that by somewhere between…the year 2020 and 2030...most major diseases would be eliminated. Has that happened? No.
Okay, and one of the reasons it didn't succeed, and the critical piece to understand, is that it did not take into account the exposome, which is all the physical factors that affect health and affect people. And so that now is calculated to be over 80% of what determines whether or not you get diseases or not. Of course…there is the climate factor. As temperatures go up, that actually facilitates diseases…to fester and grow.
So what was missing was the exposome, which is all those external types of variables and factors.
And when you look at LLMs and you look at the benchmarks for them, there is not a single benchmark — and you can go do the research, I had teams at Yale and Georgetown, do extensive research — and they could not find a single benchmark test that was measuring success in the physical world. So it's all theoretical, it's test-taking, it's automation of digital tasks, but all of it functions in the digital world.
100% of human beings live in the physical world. No human beings live in the digital world.
So that's a critical thing to understand, and the vast majority of the global economy, about 85% of it, resides in the physical world. So it's critical that this transition to understanding the physical world, and understanding those external variables, and the diversification of AI models occurs in order for us to achieve our objectives."
September 22, 2026
