AAAS 2013: Predicting human behavior

Photo via Thinkstock.

Photo via Thinkstock.

If you’ve driven on the highway, you’ve seen it: The traffic jam appears out of nowhere and dis­ap­pears just as mys­te­ri­ously.  We blame the cars around us for their poor dri­ving skills, and slam on our own breaks. During an AAAS annual meeting ses­sion hosted by North­eastern pro­fessor Albert-​​László Barabási, North­western pro­fessor Dirk Hel­bing showed a video of cars dri­ving at con­stant speed around a cir­cular race track. Even­tu­ally, small vari­a­tions in dri­ving speed prop­a­gate into a cas­cading traffic jam that travels back­wards around the track.

You can imagine how more cars would allow this phe­nom­enon, which emerges as a result of the cars’ inter­con­nect­ed­ness, would happen a little quicker on a more con­gested track. This is how I think about our world these days. More than seven bil­lion people are trav­eling around this giant track and we’re get­ting our­selves into some nasty jams. Cli­mate change, finan­cial melt­downs, the Arab Spring, SARS: most of the large scale prob­lems we’re now facing stem from the fact that our system has grown too big for its britches.

Of course, it’s not just a problem of size. It’s also a problem of con­nect­ed­ness. Because human socioe­co­nomic and tech­nical sys­tems are deeply inter­twined through global net­works of things like mobility and money, small per­tur­ba­tions are vastly amplified.

But just as our pop­u­la­tion has grown, so too has our ability to track that pop­u­la­tion. More data has emerged in the last two years than did in all pre­ceding human his­tory. Alone that would just be an over­whelming con­cept worthy of little more than a minor panic attack, but cou­pled with insights from the phys­ical sci­ences and math­e­mat­ical mod­eling, we are now in a posi­tion to actu­ally probe all that data and use it to under­stand the under­lying mech­a­nisms that govern human behavior.

The ses­sion was titled “Pre­dictability: From Phys­ical to Data Sci­ences,” and five speakers in addi­tion to Hel­bing dis­cussed ways in which human behav­iors can be fore­casted much the same way we track the path of a hurricane.

North­eastern research assis­tant pro­fessor Chaoming Song looked specif­i­cally at the pre­dictability of human motion. Using cell phone data, Song and other member’s of Barabasi’s lab showed that human mobility is an inter­play between two phe­nomena: returning and exploring. Most of the time, we’re moving from one familiar place to another. Every now and then, we ven­ture off to some new des­ti­na­tion. Intu­itively, it’s not that sur­prising, but the fact that Song and his col­leagues are now able to model those pat­terns with extreme accu­racy is ground breaking.

Work like this informs the work that North­eastern pro­fessor Alessandro Vespgi­nani and North­western pro­fessor Dirk Brockman are doing to track the spread of epi­demic dis­eases around the globe. Vespignani’s GLEaM model com­piles a series of lay­ered human mobility data sets to pre­dict epdi­demic tra­jec­to­ries and severity. His work recently demon­strated that in order to have any kind of effect, trans­porta­tion restric­tions for con­taining dis­eases would have to shut down 99% of human move­ment in order to be remotely effec­tive. In con­trast to the way small vari­a­tions in speed amplify into major traffic jams, large inter­ven­tions can some­times have min­imal effect in a com­plex system.

Instead of using real data about human motion like Vespig­nani, Brockman looked at the move­ment of dollar bills around the globe using the web­site wheres​ge​orge​.com. Brockman used his proxy data set to demon­strate where and when dis­eases would break out next and despite the seem­ingly random pat­tern, they results were strik­ingly sim­ilar to those derived from Vespigani’s method. This urged Brockman to search for a “more fun­da­mental” approach. He decided that per­haps our global con­nec­tivity means that the dis­tances between two cities shouldn’t be described in geo­graphic terms, but rather “effec­tive” terms. If the frac­tion of people moving between two air­ports is enor­mous, those two air­ports should be con­sid­ered as very close together. But if no one travels between them, they are very far apart (even if geo­graph­i­cally they’re quite close).

He redrew the net­work of inter­na­tional air­ports using this new def­i­n­i­tion of dis­tance and now dis­eases spread just as you might expect, like a ripple prop­a­gating through water, con­cen­tric cir­cles ema­nating form a cen­tral focal point like a pebble or the ini­tial case of SARS.

Vespig­nani noted that in the 14th cen­tury it took sev­eral years for the black plague to ripple through the pop­u­la­tion, but it did so just like that: with a wave­like motion, geo­graph­i­cally prop­a­gating from one city out­ward. Brockman’s approach shows how our inter­con­nect­ed­ness changes only the shape of the net­work on which that ripple travels.

Other speakers at the ses­sion included Marta Gon­zalez of MIT and Boleslaw Szy­manski of Rens­se­laer Poly­technic Institute.