
The Flux by Epistemix
Welcome to The Flux - where we talk data, decisions, and stories of people asking the what-if questions to create an intentional impact on the future.
The Flux by Epistemix
From Neural Networks to Synthetic Populations: Hamdi Kavak’s Journey
What if we could predict the impact of a pandemic, a policy change, or even a war before it happens? Hamdi Kavak joins The Flux to discuss how agent-based modeling enables researchers to explore "what if" scenarios and simulate real-world behaviors at both small and massive scales. We dive into the applications of ABM in government-funded projects, the difficulties of modeling millions of agents, and how better communication of simulation insights could improve public trust in data-driven decision-making. If you're curious about the future of modeling and simulation, you won’t want to miss this conversation.
Welcome to The Flux, where we hear stories from people who have asked what if questions to better understand the world and talk about how data can help tell stories that impact decisions and create an intentional impact on the future. This is your host, John Cordier, CEO at Epistemix. In a world where the flux capacitor from back to the future does not yet exist, people have to make difficult decisions without always knowing how the future will play out.
Our guests are people who've taken risks, made decisions when uncertainty was high, and who have assisted decision makers by using data and models. We hope you can turn lessons from our podcast into foresight, so you or your organization can make better decisions and create an intentional impact for others.
Hey there, welcome to another episode of The Flux. Today, we’re on with Hamdi Kavak. Thank you for jumping on the podcast today.
Hamdi: Of course. Happy to be here. Thanks for the invite.
John: Yeah. So, we’re here in Santa Fe at the Complex Social Systems Conference. A lot of people here are talking about agent-based modeling, system dynamics modeling, and other ways to understand complex systems in our society today.
Hamdi, what got you into the field in the first place?
Hamdi: So, my background is in computer engineering—that was my undergrad degree, and I did a lot of programming. I worked as a software engineer in the past.
When I moved to the U.S. for my education, I wanted to venture into a new field, so I picked a school that was unique in that aspect. I did my master’s and PhD at Old Dominion University in the Department of Modeling and Simulation. Imagine a whole program dedicated to developing models of different types.
Unfortunately, I heard that they got rid of the department and made it a PhD-only program, but at least it still exists.
In that program, during the first year, you take a variety of courses covering different modeling techniques—you learn about system dynamics, discrete event simulation, mathematical models, and agent-based modeling was one of them.
As a computer engineer, I was really into what we used to call neural networks—now we just call it AI. I loved the concept of developing something that mimics neurons, and back in college, I even wrote my own code to develop a neural network model. The idea of mimicking real-world systems really fascinated me.
Agent-based modeling had that same appeal—it allows you to build something that mirrors real-world behaviors, sometimes down to an individual level. That was what first got me into the field, and from that point on, I’ve tried to contribute as much as possible—all the way to where we are now, here at the CSS Conference.
John: Cool. When you look at using agent-based modeling to replicate reality, some people might say, "Well, you need a dataset to start with." Others might focus on building synthetic populations or reconstructing parts of a system with available data.
What kinds of realities are you and your teams working on? Are you building large-scale synthetic populations? Smaller populations? A mix?
Hamdi: Yeah, actually, the scale of the model is usually dictated by the project.
For example, we worked on a DARPA project called Ground Truth, and before that, there was another DARPA program called SocialSim, which ran between 2016 and 2019. After that, they cut a lot of their social science-related agent-based modeling projects. I don’t know if that was because of our performance or something else going on, but either way, it was a shift.
For the DARPA project, we were asked to develop models with up to 10,000 agents. When you’re working with that scale, you can add a lot of detail to individual agent behavior, which makes things easier.
Now, we’re working on a different project funded by IARPA. Initially, we started with 10,000 agents, but now we’re scaling up to 30 million.
John: Wow.
Hamdi: Yeah. In that case, we have to start with a synthetic population, and every decision we make has to be very deliberate. Right now, we're at about 2 million agents, modeling mobility for an entire metropolitan area.
At that scale, adding complexity becomes challenging because every additional layer increases runtime. You could add one new feature and suddenly the whole simulation grinds to a halt.
John: Right.
Hamdi: So, one of the biggest challenges is balancing model detail versus scale. It’s always a trade-off in modeling.
John: Absolutely. When you compare agent-based modeling to other areas of computer science and analytics—like machine learning, neural networks, or large language models—what do you think has held agent-based modeling back? And what’s changing now that might allow it to flourish?
Hamdi: That’s a great question.
I think one big factor is that AI and machine learning got a major boost from companies—Google, Amazon, Facebook, and others invested heavily in these technologies.
Back in 2008, Google developed Hadoop and kicked off the big data movement. Then you had things like ImageNet that advanced the field further. These big companies pushed AI and machine learning into mainstream use. Now, AI is an everyday tool—I don’t know anyone who doesn’t use ChatGPT at this point.
Agent-based modeling hasn’t had that same corporate push.
Years ago, we actually tried to start a company to run agent-based models in the cloud so that people wouldn’t need to install any software. But it required a lot of investment, and we couldn’t get it off the ground.
For agent-based modeling to take off, we need companies to invest in it at scale.
John: Yeah, making agent-based modeling more accessible is key. Right now, it’s mostly used by researchers and specialists. If more people could easily interact with it—especially outside the ABM community—that would be a game changer.
Hamdi: Exactly. And another thing that could help is integrating AI and machine learning into agent-based models. That was actually the focus of my dissertation—combining machine learning with ABMs. If we can bridge those two worlds, I think we’ll see more adoption.
John: Yeah, that makes sense. If someone is new to agent-based modeling, how do you explain its value to them? Sometimes we compare it to SimCity—it’s like building a simulation of the real world. Or we reference The Tipping Point by Malcolm Gladwell to show how agent-based modeling can help us understand emergent effects in a population.
How do you introduce ABM to people who aren’t technical?
Hamdi: I think the best way is to highlight the "what if" aspect. Agent-based modeling is great for exploring scenarios that don’t exist in the real world yet.
If someone is brand new, I recommend starting with NetLogo. It’s beginner-friendly and has a huge library of example models. Even if you don’t know how to code, you can play with existing models and see how they work.
But if someone wants to do serious agent-based modeling—like for research or industry applications—they’ll eventually outgrow NetLogo. At that point, I’d suggest tools like Repast or Mesa, especially Repast for Python, since Python is so widely used now.
John: Great suggestions. So, we named this podcast The Flux after the flux capacitor from Back to the Future. If you could rewind history and apply today’s simulation capabilities to a past event, what would be your "What the Flux" moment?
Hamdi: Hmm. Well, on an idealistic level, I’d love to go back to World War I or II and show people the consequences of their actions—how many millions of lives would be lost if they went to war.
But if I think realistically, I’d say the early days of COVID. In January 2020, I saw the estimated spread rate from that cruise ship case study, and I knew it was going to be bad. In March, I showed my students a simple simulation predicting widespread infection, and I told them, “This might be our last in-person class.” No one believed me at the time.
Even with data and simulations, people struggle to accept future risks. That’s a science communication problem.
John: Yeah, getting people to trust and act on simulations is a challenge.
Hamdi: Exactly. We need better ways to communicate scientific insights to the public. More podcasts like this, more educational outreach, and more people who can bridge the gap between research and real-world decision-making.
John: Absolutely. Well, thanks for stepping away from the conference to join us today.
Hamdi: Thank you for having me.
John: Have a great rest of your day.
Hamdi: You too.