Podcast: Play in new window
BOB HIRSHON (host):
Detecting deception…I’m Bob Hirshon and this is Science Update.
Websites like TripAdvisor can be handy for finding information on unfamiliar destinations. But how can you tell if that rave review was written by someone who’s actually been there? A team of scientists at Cornell University collected a sample of actual positive hotel reviews and also paid volunteers to write a bunch of fake ones. Then they analyzed the language in those reviews with a computer model. Myle Ott, a graduate student in computer science, worked on the study.
MYLE OTT (Cornell University):
Our algorithm was actually nearly 90 percent accurate at detecting deception. The best human judge was 61.9 percent accurate. And two out of three human judges actually performed at chance.
It turns out that humans tend to believe what others say and therefore aren’t very good at figuring out when someone’s fibbing. A computer doesn’t have that bias, so it more readily tells real from fake. I’m Bob Hirshon, for AAAS, the science society.