Alex Pentland, a pioneer in the field of behavioral analysis, described computer models that could reliably foresee the outcomes of romantic dates, predict traffic jams, and optimize the structure of corporate meetings for productivity in the keynote speech at the Dartmouth computer science research symposium on Saturday.
Much of the information generated by human interactions is background noise with no predictive value, according to Pentland, a professor at MIT. Hidden beneath the abundance of noise however, are measurable patterns that can be harnessed to predict the outcome of an interaction. Pentland’s goal in his research is to sift through the noise to find these “honest signals” to better understand human behavior.
For example, Pentland’s team analyzed conversational patterns in five-minute speed-dating sessions to predict whether a couple would share contact information at the end of the date. Each participant carried a PDA that measured patterns during the conversation, such as activity level, voice stress, and length of speech.
The team found that short back-and-forth exchanges often indicated romantic interest, while longer exchanges did not. Using such “honest signals”, they were able to account for over one-third of the variation in outcome. Additionally, they found that the woman’s engagement in the conversation had greater correlation to the dating outcome than the man’s, indicating that she shared a greater portion of the decision.
Pentland stressed that “honest signals” were universal. They did not depend on the words or even on the language used in the interaction.
“A great deal of this is folk knowledge, but it never gets taken seriously until we can quantify it,” he said.
He also noted that these signals were indeed “honest,” since even people who are aware of them have trouble faking them. People who attempt to fake the signals quickly lose track of the conversation.
“In doing this research, I’ve come to pay attention to it,” he said. “But it’s hard to pay attention and talk at the same time.”
Pentland added that similar systems could be used on larger scales to predict traffic patterns or social demographics. Many cars and cell phones now have GPS systems built in, allowing scientists to track and analyze human movement. He found that taxicab routes in San Francisco are far from random, and that their paths reveal different blocks within the population. This data may pave the way for friend-finder software that tracks people with similar behavioral patterns. Businesses may soon use it to optimize their location to their target demographic.
Pentland also warned, however, that “You should be scared of some of the things going on with this data.”
A GPS chip in a phone could locate someone at any time, he said, explaining that he is a proponent of legislation to protect such personal data. Some of the solutions he suggested included watermarking or encryption of the data.
While there are dangers associated with this new technology, Pentland expressed optimism that these models will soon change the way people look at society.
“So you think you can model humans?” one audience member asked.
“We are modeling humans,” Pentland answered. “Just not comprehensively.”
Alex Pentland’s research on human behavior modeling is the subject of his new book, Honest Signals: How They Shape Our World (Bradford Books).