Not ‘The Sims,’ But Close: How Computer Models Help Public Health Researchers Plan for Disasters


(elevator music)
The toy playground epidemic. The simplest model. There are a hundred kids on a green playground, Blue kids are healthy,
red kids are sick. There’s one initial red kid. When a red kid bumps into a blue kid,
he gives him the disease with some probability, and then after you’re red
for a while you’re removed from play. Oh.
(click) That’s it. I guess in that run, one kid. But he’s gonna die, too. I just
stopped the movie at that point. (futuristic, fast-thumping music) Agent based models are
artificial societies of software people. The people interact with each other and
those interactions generate social phenomena, health phenomena that we care
about: epidemics, distributions of wealth, distributions in the burden of disease,
spatial patterns, norms of health behavior, patterns of addiction, and so
forth. The idea is that we build these agent based models, equip them with
human-like decision systems, and see if their interactions can generate the
phenomenon from the bottom up. It explains these phenomena and
sometimes suggests very interesting bottom-up solutions. It’s sort of SimCity, but much, much more serious. We’re interested in calibrating these models to real-world data.
So when you build a model of smallpox, or pandemic flu, it has
some empirical credibility which increases its impact. There are individual-based models in economics. They assume a perfectly
informed, rational, optimizing actor. This is simply not the way humans
make their decisions. And we’ve tried to develop a next
generation agent that has emotions, social networks, bounded rationality.
One piece of agent zero is a very simple toy model of fear.
Then, like humans, they have imperfect data. They analyze it
incorrectly, and they make a poor statistical inference. But they’re also
in a network of other emotionally driven and partially rational creatures.
And when you put them all together they produce genocide, dysfunctional health
behaviors, all sorts of collective dynamics that are very disturbing that
a rational collective would not do. And we’re trying to use it to study
addiction, smoking, alcoholism, all kinds of phenomena in which rational
behavior can be overridden. Agent-based modeling is the all-terrain vehicle.
You can use it to study all of these. It’s had a huge impact in the way
we do infectious disease modeling, and infectious disease policy design.
Stakeholders are really part of the modeling team. And then when
it produces a counterintuitive result, they’re not scared away from it.

Daniel Yohans

Leave a Reply

Your email address will not be published. Required fields are marked *