By Leah Poffenberger
Everyday life is full of choices: Coffee or tea? Apples or oranges? iPhone or Android? We also have to tackle more serious questions, like who to vote for in an election. Especially for the important choices in our lives, these decisions don’t happen in a vacuum: We’re usually getting a lot of information from a lot of places, especially from people around us. But how much impact do other people’s choices have on our own?
As it turns out, social networks—and the beliefs the people in those networks already hold—can have a big influence in steering what decision people within that community make. Zachary Kilpatrick, an applied mathematician and the University of Colorado, Boulder, along with collaborators at the University of Houston set out to model how individuals make choices as they gather information over time, from both their private research and members of their social network.
“A lot of past work has focused on situations where people have access to some set of information at a precise point in time, then they discuss or share information in order to arrive at individual decisions or decisions of the group,” says Kilpatrick. “But not a lot of work has thought about how the process of mixing together private information sources, like maybe books that people read or independent research…unfolds in time along with information gleaned from knowing the decisions their friends make.”
Kilpatrick and his colleagues created a model with a group of agents, representing people in a social network, who would be fed information in favor of two different choices, choice A or choice B, until receiving enough information to make a decision. Each agent already had a preference—or bias—for one of the options, changing how much information they’d need to tip the scales in favor of A or B. Once one agent chose either A or B, that act infused the rest of the social network with one more piece of information—often causing other agents to make a decision, too.
“The act of making a decision actually provides a big boost of information to the neighbors that you have in the social network, so much so that it can drive them to immediately make a decision,” says Kilpatrick. “If your friends are tweeting out what candidate they’re voting for…and you value [your friend’s] beliefs, that may sway you to vote for that person.
|Each agent receives a sequence of pieces of evidence. Agent 1 then communicates their decision state (decided for H+, decided for H-, or undecided) to agent 2. Agent 2 then uses this information to update their belief. The graph on the left shows the flow of information as the agents update their beliefs.
Knowing how those friends might usually vote—or what their biases are—can provide extra information or influence in decision making. Someone with a bias towards a certain candidate only needs a little bit of information in line with their preference to cross over an information-threshold and cast their vote for that candidate. But if we know what someone’s preference is, and they don’t make any decision, that can be important: they must have learned something compelling that goes against their bias, causing them to move away from the decision-making threshold for their typical choice.
“Let’s say you have a person that typically votes Republican, but they take a long time to make that decision…The fact that they haven’t decided yet probably means they’ve gotten evidence in favor of the Democratic candidate,” says Kilpatrick. “If they had gotten just a little bit of evidence for the Republican candidate, they would have gone ahead and voted for them already. This means you can even get information from the fact that a person has NOT made a decision. This is true even if you don’t know what evidence they received. Just the fact that they have not made a choice by some point in time is informative.”
Likewise, if a friend goes against their usually held bias—maybe they’ve always bought iPhones and suddenly they’re buying Android phones—that tells us they must have collected a lot of data in favor of changing their position. According to Kilpatrick, a person going against their usual bias gives everyone in their network a huge burst of information. Something must be great about Androids!
Kilpatrick has also worked on models of what he calls “large cliques,” or a social network where everyone is in communication with everyone else, and how decisions are made in these groups.
“Let’s say you have many people that can all observe each other’s decisions,” says Kilpatrick. “The interesting thing that we found is that this group decision process can be kind of self-correcting: If the first person convinces only a few people to follow their choice, that reveals to others that the first decision may not have been a good one.”
Image A shows an agent in a clique making a quick, but incorrect, decision. In B, a few agents agree with the first. In C, the second wave of the agents have made a correct choice, after seeing the very small group making the choice in B. (Image credit: Kilpatrick)
While Kilpatrick’s current research has been based purely on artificial agents, for future projects, he and his collaborators are looking at ways to gather data from real people to gain further insight into decision-making.
Kilpatrick’s motivation for creating robust models of social decision-making comes from an interest in neuroscience that he picked up in graduate school. He recalls reading papers on research that showed that, when a monkey was given a two-choice decision task, the activity in certain parts of the monkey’s brain could indicate what choice the monkey preferred.
“There’s a long history in neuroscience of using these experimental methods to say what the neural mechanisms of decision making [are], and what’s so remarkable is that the things that neurons in people’s brains are doing are often what you would think an optimal statistical decoder would do,” says Kilpatrick. “We think that evolution [can often] converge on these optimal solutions.”
Kilpatrick notes that much of this neuroscience research is on one individual making a choice and not on decisions made within a social network.
“Not many people think about how the brain does social decision making. There’s a few people, but [they’re] not really thinking about serious statistical descriptions of the computations the brain might be doing while people or other animals are performing social decision making,” says Kilpatrick. “That’s where good, sturdy mathematical modeling comes in because we can look at situations where we can really understand what these models are doing and then even speculate on how the brain might implement that kind of computation.”