Robots Are Going to Be Doing National Security Clearance Interviews

Advancing a career in the US government might soon require an interview with a computer-generated head who wants to know about that time you took ketamine.

Psychologists at the National Center for Credibility Assessment (NCCA) are developing an interview system that uses a responsive on-screen avatar for the first stage of the national security clearance process.

Initial screening for a variety of government jobs currently requires applicants to fill out a form disclosing past drug use, criminal activity, and mental health issues, which is then reviewed during an interview—with a human.

But a recent NCCA study published in the journal Computers and Human Behavior asserts that not only would a computer-generated interviewer be less "time consuming, labor intensive, and costly to the Federal Government," people are actually more likely to admit things to the robot.

The study used US Army basic trainees as volunteer subjects for a mock national security clearance interview. The trainees were not told that the questions would be asked by a robot. After being hooked up to electrodes for cardiographic and electrodermal (heart and skin) responses the volunteers were told that the interview would be with a computer avatar, and were left alone in a chamber with their on-screen interrogator.

The program used for the study was capable of responding to vocal cues and taking multiple conversation paths depending on the subject's answers. The researchers were hoping to leverage the power of presence: the idea that people recognize another sentient being in the environment, and are more responsive as a result.

The bot is racially ambiguous and looks like a sort of cross between Barack Obama and Vladimir Putin. Clean-shaven and all business, the bot asks you to divulge your most embarrassing personal mistakes in the name of national security and trustworthiness.

And apparently these computer-generated heads had a lot of presence. Volunteers in the study were significantly more likely to disclose alcohol use and mental health issues to the avatar than to the questionnaire. Responses for drug use and criminal charges were about the same.

Using the avatar also allowed the researchers to measure pauses in conversation and take advantage of questions that would seem out of place on a paper form. At the end of each interview section the computer-generated interviewer asked the volunteers "if there was anything at all" they wished to discuss—with over 10 percent then responding with more information.

The researchers concluded, in so many words, that national security clearance interviews can totally be outsourced to a computer-generated agent. That's not an empty recommendation: The NCCA grew out of the Department of Defense Polygraph Institute and is still responsible for "lie detection" training for all branches of government. It's also tasked with developing new technologies for credibility assessment.

In other words, when "Blade Runner" is an actual job, they will likely be trained at the NCCA headquarters in South Carolina.

The NCCA predicts a bright future for its virtual agents. The study notes that computer-generated interviewers might help mitigate the gender and culture bias that affects human interviewers. It also recommends using avatars with distinctive physical characteristics and "culture-specific utterances."

The interviewer isn't quite a sentient AI; it relies on a dialogue tree similar to telephone customer service: tell the computer all the simple things, then press 0 for a human to explain the story behind your streaking arrest.

Still the idea of a computer conducting national security clearance interviews, even with human oversight, is bound be unsettling for some. But depending on the system's effectiveness and the potential cost savings, we may see national security screening being done by a screen in the very near future.