In one corner we have a computer armed with a list of things you “Liked” on Facebook. In the other, your friends, family, and coworkers, relying on an analog trove of memories and the raw strength of their own judgment. Who’s sketch of your personality is more likely to match your own?
Those who voted for the computer might be cynical, but they’re also probably right, according to a new study conducted by researchers at the University of Cambridge and Stanford University. By harvesting Facebook “Likes,” the researchers’ computer model proved more accurate at divining a person’s self-reported personality traits than their own kith and kin.
The researchers say their findings suggest that cheap, automated, and accurate personality tests could help people more accurately judge the personalities of others. Not only could that boost more targeted marketing, but it could revolutionize how people choose “whom to marry, hire, or elect as president,” says study co-author Dr. David Stillwell. It’s also a milestone in developing socially savvy, emotionally sensitive computers, say the researchers – sort of like the one Joaquin Phoenix’s character falls in love with in the movie Her.
Using a sample of more than 86,000 volunteers on Facebook who completed a 100-item personality tests and gave access to their Likes, the researchers built a model that predicted personality traits based on Likes—for instance, people who Like, say, the artist Salvador Dali or TED Talks tend to be open-minded, while those who Like the US reality TV personality Snooki are unusually outgoing. The scientists then tested the model against an additional 17,600 participants judged by one friend or family member, and more than 14,400 subjects evaluated by two.
It turned out the more Facebook Likes to evaluate, the more accurate the model. For instance, given the average number of Facebook Likes – i.e. 227 the algorithm handily bested all but spouses in more accurately predicting the subject’s personality. And for subjects with 300 or more Likes, the computer beat even significant others.
There are some important caveats to bear in mind, though.The touchstone for most of the study’s comparisons was analysis based on the “Big Five,” a set of personality measures that includes openness to experience, conscientiousness, extraversion, agreeableness, and neuroticism. Though the Big Five are commonly used for personality testing, the method is inevitably limited to the traits it measures.
Plus, the personality test friends, family, and co-workers took was only 10 items long, compared with 100 for the self-evaluation. This might limit the range of knowledge tested, noted the study.
The computer model also proved better at judging openness – something expressed mainly through a person’s interests, preferences and values. That’s not exactly surprising; while collecting and synthesizing all those tiny clues is tough for brains, computers excel at it.
In the core part of the study, the competition between man and machine came down to who knew the subject best. But subjects don’t necessarily know themselves that well. Even though the Like-based computer model was geared toward predicting a person’s own evaluation of himself based on the Big Five, the model proved eerily adept at predicting real-life behavior refracted through a user’s Facebook activities (e.g. number of status updates, political affiliations) – in some cases, the model outperformed the subjects themselves.
This article originally appeared on Qz.com
We welcome your comments at
letters@scroll.in.
Those who voted for the computer might be cynical, but they’re also probably right, according to a new study conducted by researchers at the University of Cambridge and Stanford University. By harvesting Facebook “Likes,” the researchers’ computer model proved more accurate at divining a person’s self-reported personality traits than their own kith and kin.
The researchers say their findings suggest that cheap, automated, and accurate personality tests could help people more accurately judge the personalities of others. Not only could that boost more targeted marketing, but it could revolutionize how people choose “whom to marry, hire, or elect as president,” says study co-author Dr. David Stillwell. It’s also a milestone in developing socially savvy, emotionally sensitive computers, say the researchers – sort of like the one Joaquin Phoenix’s character falls in love with in the movie Her.
Using a sample of more than 86,000 volunteers on Facebook who completed a 100-item personality tests and gave access to their Likes, the researchers built a model that predicted personality traits based on Likes—for instance, people who Like, say, the artist Salvador Dali or TED Talks tend to be open-minded, while those who Like the US reality TV personality Snooki are unusually outgoing. The scientists then tested the model against an additional 17,600 participants judged by one friend or family member, and more than 14,400 subjects evaluated by two.
It turned out the more Facebook Likes to evaluate, the more accurate the model. For instance, given the average number of Facebook Likes – i.e. 227 the algorithm handily bested all but spouses in more accurately predicting the subject’s personality. And for subjects with 300 or more Likes, the computer beat even significant others.
This graph shows the accuracy of the computer model’s personality judgements compared with those of humans, based on number of Likes evaluated.(Wu Youyou/Michal Kosinski)
There are some important caveats to bear in mind, though.The touchstone for most of the study’s comparisons was analysis based on the “Big Five,” a set of personality measures that includes openness to experience, conscientiousness, extraversion, agreeableness, and neuroticism. Though the Big Five are commonly used for personality testing, the method is inevitably limited to the traits it measures.
Plus, the personality test friends, family, and co-workers took was only 10 items long, compared with 100 for the self-evaluation. This might limit the range of knowledge tested, noted the study.
The computer model also proved better at judging openness – something expressed mainly through a person’s interests, preferences and values. That’s not exactly surprising; while collecting and synthesizing all those tiny clues is tough for brains, computers excel at it.
In the core part of the study, the competition between man and machine came down to who knew the subject best. But subjects don’t necessarily know themselves that well. Even though the Like-based computer model was geared toward predicting a person’s own evaluation of himself based on the Big Five, the model proved eerily adept at predicting real-life behavior refracted through a user’s Facebook activities (e.g. number of status updates, political affiliations) – in some cases, the model outperformed the subjects themselves.
This article originally appeared on Qz.com
