Robocalls No Match for AI-powered Virtual Assistant
With the help of artificial intelligence (AI), malicious robot calls may soon become a thing of the past thanks to a smartphone application developed by Georgia Tech researchers.
Created by Georgia Tech computing alum Sharbani Pandit (CS Ph.D. 21), RoboHalt uses machine learning and natural language processing to intercept and screen incoming calls for spam. The smartphone compatible virtual assistant successfully blocked 95% of mass robocalls, 82% of evasive robocalls, and 75% of targeted robocalls in Pandit’s research.
Traditional methods of blocking robocalls use blocklists, which are ineffective against caller ID spoofing. Robohalt solves this problem by intercepting the call and randomly asking the caller to hold or continue the conversation. The program asks a set of questions like “Who are you trying to reach?” and “Can you tell me more about it?” to unknown callers and determines if the call should be put through.
“A phone call from a number not saved in a contact list would get picked up by the virtual assistant which would then screen the caller,” Pandit said. “If the caller is a robot and cannot answer the questions the assistant asks, the call is terminated. If the caller passes, the call along with a transcript of the screening is forwarded to the user.”
While they are a nuisance to millions, robocalls also pose significant security threats. Like many scams, robocalls rely on their targets being convinced that the call is real. Examples include impersonating the IRS, law enforcement, or announcing that the caller has won a contest they never entered. Robohalt helps mitigate this by acting as a CAPTCHA test to filter out most of the spam.
Pandit graduated in Fall 2021 and now works on the Amazon Customer Trust and Partner Support group as an applied scientist. She assists Amazon in detecting fraud through research and maintaining the company’s current systems.
The research paper, Combating Robocalls with Phone Virtual Assistant Mediated Interaction, will be presented at the 32nd USENIX Security Symposium in August. Contributors to this project include Krishanu Sarker of Georgia State University, Roberto Perdisci of the University of Georgia, and Georgia Tech, as well as Mustaque Ahamad of Georgia Tech and Diyi Yang of Stanford University.