Skip to main content
Search form
  • 00:00

    [MUSIC PLAYING][Studying Complex Systems of Social BehaviorUsing Agent Based Modeling]

  • 00:09

    JOSHUA BECKER: My name is Joshua Becker.[Joshua Becker, Postdoctoral Student,Northwestern University] I'm a postdoctoral researchfellow here at the Kellogg School of Managementand with the Northwestern Institute on Complex Systemshere at Northwestern University.And I study collective intelligence,which means I use a combination of formal computational modelsand experimental design to understand

  • 00:30

    JOSHUA BECKER [continued]: how things like the network structure of communicationimpacts the accuracy and optimally of group decisions.[What recommendations would you have for someonelooking to do similar research]I went to a very small college, St. John's Collegein Annapolis, Maryland, that emphasized

  • 00:50

    JOSHUA BECKER [continued]: historical science and the great books, is what we called it.So we studied a lot of the history and philosophyof math and science.And one of the things we studied, of course,was Isaac Newton's Principia.And it blew my mind that Isaac Newtonwas able to understand the physical world--

  • 01:11

    JOSHUA BECKER [continued]: this is the foundation of contemporary physics--simply by sitting at his desk and working outwhat some people call thought experimentsusing formal mathematical modeling.And I wanted to know if we could bringthat rigor and a priori analysis to our understandingof the social world.

  • 01:33

    JOSHUA BECKER [continued]: So when I got to grad school at the University of Pennsylvania,by a bit of luck as well as some intention,I ended up in a lab that did just that.So we use formal computational models to build theoriesabout the social world and about howsocial interactions structure collective behavior and group

  • 01:55

    JOSHUA BECKER [continued]: decisions.[What recommendations would you have for someonelooking to do similar research]One thing I think that's important to noteis that not all mathematical models are statistical models.So when I talk about formal computational modelsfor social behavior, what I'm talking about

  • 02:16

    JOSHUA BECKER [continued]: is, what some people call, generative models.So we're not looking at data and running statistics on itto try to understand what already happened.But we're actually sitting down and thinking about,how do people behave?How does one person respond to their environmentand to stimuli?And then we think about, OK, what

  • 02:37

    JOSHUA BECKER [continued]: if you had 100 or 1,000 people.And we take that model of individual behaviorand scale it up to help us understand complex systems.So some people call it agent-based modeling.Some people call it generative modeling.It's a very broad set of methods that

  • 02:58

    JOSHUA BECKER [continued]: tries to understand the social world by growingcomplex social behaviors from the ground up.So in that sense, it's fundamentallydifferent from statistical modelingbecause it's more like these thought experimentsthat Isaac Newton used to develop contemporary physics,in that you're just sitting down and thinking rigorously

  • 03:20

    JOSHUA BECKER [continued]: about how people may behave without any actual dataabout the world.And what's amazing is that it works.So within my lab, we've published several papers.And of course, they published papersbefore I even got there that started with these models.We then run these experiments.

  • 03:40

    JOSHUA BECKER [continued]: And it's always incredible to seethat these experimental groups performas you'd expect them to.Now, they don't always perform exactlyas you'd expect them to.And that's where the statistics do come in.We do actually get data about people and their interactions.So then we look at that data, and it surprises us.

  • 04:01

    JOSHUA BECKER [continued]: If you're not surprised in science,you're not doing it right.So at that point, we take the points that surprise us.And we go back to the model.And this is why having formal models of social behavioris so important because, when you develop this formal model,you're forced to think through, whatare my assumptions about people's behavior?

  • 04:22

    JOSHUA BECKER [continued]: What are my assumptions about how peopleinteract with each other?And when you get that data, you can look exactlyat what your assumptions are and seeexactly where you went wrong.And then you adjust your model.You adjust your model to see if you can better fit the data.And then you run more experiments.And it's that circle between theory and data

  • 04:44

    JOSHUA BECKER [continued]: that really makes social science move forward.[What recommendations would you have for someonelooking to do similar research]I can't really tell you what I mean by complex systemsbecause people throw that term around very broadly.It's not it's not a precise scientific or methodological

  • 05:05

    JOSHUA BECKER [continued]: term.One of the things that's common to what a lot of peopledescribe when they talk about complex systemsis unintended consequences or unintended outcomes.So a complex system is one where youmake some change or some intervention,maybe you tweak a parameter or start some messaging campaign

  • 05:25

    JOSHUA BECKER [continued]: to try to change people's behavior.And the system goes off in a completely unexpecteddirection.Colloquially, people are familiar with the ideaof the butterfly effect.That's also known as sensitivity to initial conditions.It means that very slight changesin any setting or parameter within the system,

  • 05:47

    JOSHUA BECKER [continued]: no matter how small the change, can lead to enormous changesin the qualitative behavior of the system.Within our experimental work, we createsuch simple social networks.We have 40 people.And they're connected in this very geographically simplesystem that I'm not really sure what I do is actually

  • 06:09

    JOSHUA BECKER [continued]: complex systems at all.But it's sort of a terminology, a field signal, if you will,in terms of the kind of things that people study.And they have me at these conferences,so there must be some relevance, I suppose.[What recommendations would you have for someonelooking to do similar research]

  • 06:30

    JOSHUA BECKER [continued]: I would recommend using formal generative modelsanytime you want to study social behavior thatinvolves more than one person.Really, people have used formal models, generative models--I resist the term agent-based model--to study segregation.That's one of the classic models,

  • 06:52

    JOSHUA BECKER [continued]: is Schelling's segregation model.[Schelling model of segregation] People haveused it to study the emergence of conventions.People have used it to study group decision-making.There are a few general cases, not so muchthe substantive area in terms of the type of social behaviorbut the type of scientific question you want to ask.

  • 07:14

    JOSHUA BECKER [continued]: That's really what distinguishes inappropriate or ineffectiveuse of formal modeling.So the best types of questions to ask with formal modelingare, how is this possible?When you have some social phenomena,some social behavior, and you can't explain how that emerged,

  • 07:35

    JOSHUA BECKER [continued]: if you can come up with a generative model that generatesthat social behavior, you can't be certainthat you have the right model.But you now have a candidate.You have a proposed solution, whatwe call sufficient conditions.You can point to your model and say,I don't know if this is right.

  • 07:56

    JOSHUA BECKER [continued]: But it's sufficient to generate the behavior that interests us.Another very useful place to use agent-based modelsis when you want to develop testable predictions.So especially if you're planning on doingany sort of experimental design or intervention,formal models can be used to ask the type of question,

  • 08:19

    JOSHUA BECKER [continued]: what if we changed this one parameter?So for group decision-making, youcan ask, what if we change the network structure from onein which people communicate very efficiently to onein which people communicate very inefficiently.And it can lead to surprising outcomes.There's a very famous paper by Lazer and Friedman

  • 08:43

    JOSHUA BECKER [continued]: that asked, how does communication efficiency impactgroup problem solving?And the intuitive response would be that, well, we'resolving problems.We're exchanging information.We want people to be able to share that information.So efficiency is good, right.As it turns out, the more efficientthe groups are, the less optimal their problem solving was

  • 09:06

    JOSHUA BECKER [continued]: in the long run, because it leads to thingslike herding behavior.Everybody groups around one idea instead of exploringdifferent potential solutions.[What recommendations would you have for someonelooking to do similar research]My dissertation had some very unexpected results.And we were only able to understand them

  • 09:28

    JOSHUA BECKER [continued]: because we had a formal model thatforced us to clearly lay out exactlywhat our assumptions were.So we were doing experiment on the accuracy of group beliefformation.So for example, if you have a jar of jelly beansand you ask people how many jelly beans are in the jar,the wisdom of crowds theory tells us

  • 09:48

    JOSHUA BECKER [continued]: that the average belief in a large group of peoplewill actually be incredibly accurate.In fact, it's mathematically guaranteedto be more accurate than the average individual randomlyselected from the group.Now, the wisdom about the wisdom of crowdsis that, in order to have these accurate group beliefs,you have to keep people statistically and socially

  • 10:10

    JOSHUA BECKER [continued]: independent, which means they can't talk to each otherbecause, when people start talking to each other,you get herding behavior.You get group think.Their errors become correlated.And they no longer cancel out.So you lose the wisdom of crowds when people talk to each other.That was the common wisdom on the wisdom of crowds.

  • 10:31

    JOSHUA BECKER [continued]: But we used a formal model to actually studyexactly what we think would really happen whenpeople influence each other.And our model said social influence is notbad for the wisdom of crowds.If you let people talk to each other,they converge directly towards the group mean.And because the mean is accurate,

  • 10:51

    JOSHUA BECKER [continued]: individuals will get better, but the group will stay the same.Social influence shouldn't have any impact at allon collective intelligence.So we designed an experiment to test this theory.And we started collecting data.We started running trials.And we're getting this data in.And we actually start to see, not only our groups not gettingworse, they're getting better.

  • 11:14

    JOSHUA BECKER [continued]: So at first, I thought this was a statistical fluke.And then we collect more data.We collect more trials.And the results get stronger.I can't deny it anymore.Groups get more intelligent when people talk to each other.We're not getting herding.We're actually getting an improvementin collective intelligence.So because we had this formal model, we could go back

  • 11:34

    JOSHUA BECKER [continued]: and see exactly what we had thoughtwas true that turned out not to be true in the waypeople formed their belief.We were able to adjust our model,account for people's empirical behavior in the experiments,and come up with a model that produced sufficient conditionsto explain our observations.So now we have a model that says groups

  • 11:57

    JOSHUA BECKER [continued]: become more accurate when people can talk to each other.So it wasn't possible to have this findingwithout the experiment.It wasn't possible to understand this finding without the model.So you really need both of those piecestogether to have a rigorous scientific examination

  • 12:18

    JOSHUA BECKER [continued]: of human behavior.[What recommendations would you have for someonelooking to do similar research]A lot of people don't get this message early on.And I think it's incredibly importantto know that, if you push on an idea, it will go somewhere.When you first start a research project,

  • 12:38

    JOSHUA BECKER [continued]: you have an inkling, a thought, maybe something that just kindof interests you.So you start sketching it out.And it looks OK.Maybe you have a bit of a model.You have a little bit of data.It's nothing too exciting.You have to push on that idea.You have to take that abstract, that framing,that first paragraph, where you lay out

  • 12:59

    JOSHUA BECKER [continued]: the puzzle or the paradox.And you write it 100 times.And you write it, and you rewrite it.And you turn that question over in your head.And you find the interesting angle.And that's how all of the really great papers started.They started as an idea.And you're interested in that for a reason.

  • 13:21

    JOSHUA BECKER [continued]: You have that idea because there's somethinginteresting about that.There's a grain of truth or science somewhere in there.And you have to find that, and develop it, and bring it out.And don't stop working on the abstractonce you have the data.Go back to the original question.Go back to the framing and turn it over again and again.

  • 13:43

    JOSHUA BECKER [continued]: And eventually, you'll find the really exciting piece about it.And it seems so obvious in hindsight.But it's many hours of labor and thinkingthat get you to that really winning idea.Don't be methodologically oriented.When I started grad school, I loved formal models.

  • 14:03

    JOSHUA BECKER [continued]: And I said I want to do a physics of the social world.And I went around asking, what questions can I askwith these agent-based models?And you're methodologically oriented,you end up with really shallow questions.Start with a good question and then

  • 14:25

    JOSHUA BECKER [continued]: find the method that is best suited to answer that question.


Joshua Becker, PhD, postdoctoral research fellow at Northwestern University, discusses his research on complex systems of social behavior using agent-based modeling, including his interest in complex systems and group decision making, types of methods and models used in his research, defining what a complex system is, aspects of social behavior and phenomena best studied using agent-based modeling, unexpected or surprising results, and advice to new researchers starting out in this field.

Looks like you do not have access to this content.

Studying Complex Systems of Social Behavior Using Agent-Based Modeling

Joshua Becker, PhD, postdoctoral research fellow at Northwestern University, discusses his research on complex systems of social behavior using agent-based modeling, including his interest in complex systems and group decision making, types of methods and models used in his research, defining what a complex system is, aspects of social behavior and phenomena best studied using agent-based modeling, unexpected or surprising results, and advice to new researchers starting out in this field.

Copy and paste the following HTML into your website