US Government Chatbot Gets Everyone to Spill Their Secrets

Friday, 25 July 2014 - 11:38AM
Artificial Intelligence
Friday, 25 July 2014 - 11:38AM
US Government Chatbot Gets Everyone to Spill Their Secrets

Members of the U.S. Army are more likely to tell their deepest, darkest secrets to a robot than to a faceless questionnaire. Dean Pollina and Allison Baretta of the US National Center for Credibility Assessment published research which stated that robots are more effective at screening individuals for security clearance than a written questionnaire.

 

When a person is technically eligible for security clearance to access certain classified documents, there is still an abundance of red tape to cut through in order to actually grant said clearance. Pollina and Baretta hypothesize that the bureaucratic nightmare could be streamlined significantly if the detailed screening process was conducted by robots. In order to test this theory, they interviewed 120 soldiers with a robot capable of voice recognition asking them questions via an on-screen avatar.

 

Their findings were surprising; not only were the robots adequate, they actually elicited more confessions of wrongdoing from the participants than the current method. When interviewed by a robot, subjects were far more likely to admit previous mental health problems, drug use, and criminal activity. From the journal article in Computers in Human Behavior: "The results indicated that the number of relevant admissions during the CG interview exceeded the number of admissions made using a self-report questionnaire."

 

Pollina and Baretta are extremely optimistic about the findings, claiming that their study proves that these interviews will be more efficient if they are conducted by chatbots. Some believe this optimism is premature, such as Ehsan Hoque, who works on similar technology at the University of Rochester: "As a scientist, I love the overall promise of their findings, which are consistent with what we know about animated characters and the future we predict for them." But he feared that the variables weren't completely controlled: "It would have been better if the authors had another condition where participants would simply respond to questions from a blank screen and be recorded. That would have helped us to see the exact effect of using a computer-generated agent versus just being recorded."

Science
Technology
Artificial Intelligence

Load Comments