For the national security community, trust is vital, yet increasingly tricky to discern. In an era of battlefields without borders and ambiguous adversaries, whom can we trust?
Scott Kuzdeba and Troy Lau, research engineers at BAE Systems, had a chance to help find a technology solution to this problem through their participation in the Intelligence Advanced Research Projects Activity’s (IARPA) Trustworthiness Challenge known as INSTINCT.
INSTINCT, which stands for Investigating Novel Statistical Techniques to Identify Neurophysiological Correlates of Trustworthiness, is IARPA’s first open-source challenge and part of its larger research program called TRUST (Tools for Recognizing Useful Signals of Trustworthiness). As part of the TRUST program, neuroscience researchers conduct experiments to learn how one’s neural, psychological, physiological, and behavioral signals can reflect and predict a partner’s trustworthiness. Using existing data generated by TRUST experiments, INSTINCT competitors were asked to develop algorithms that improve predictions of trustworthiness.
The competition lasted two and a half months after IARPA’s February announcement seeking the public’s participation. And out of hundreds of competitors, Kuzdeba and Lau won first place for their entry “JEDI MIND,” short for Joint Estimation of Deception Intent via Multisource Integration of Neuropsychological Discriminators. Not only did they win first place in the overall competition, but the two remained in the top spot throughout each phase of the competition. As first place winners, Kuzdeba and Lau were awarded $25,000, which will go toward supporting future innovative efforts similar to INSTINCT.
“We achieved some pretty impressive performance results, beyond what we thought was possible when we started,” said Lau. “It is exciting to think about what might come next, in terms of continuing to explore the idea that the self’s own signals can improve decision-making about others.”
The algorithm they used is part of BAE Systems’ machine-learning expertise that applied to customer solutions, such as detecting patterns in data from multiple sensors.
Both Lau’s and Kuzdeba’s education and experience gave them an edge in this competition.
“Understanding human brain function was an essential aspect for choosing algorithms to employ in this challenge,” said Kuzdeba, who is pursuing a Ph.D. in computational neuroscience from Boston University. “With this realization, our background in neuroscience positioned us well for the interaction between human behavior and machine learning.”
Kuzdeba’s research into brain machine interfaces and speech neuroscience helped provide insight into the methods and features for the algorithms. He said he also relied on prior experience with feature selection and fusing data from multiple sources for some Department of Defense programs.
Lau, who received a Ph.D. in physics from the University of Michigan and also completed postdoctoral research in human neuroscience at the U.S. Army Research Lab, used his expertise in human behavior, large-scale data mining, and machine learning to build the algorithms for the challenge. Additionally, Lau served as an investigator on DARPA’s RCI Program that built an EEG brain controller for a robotic arm prosthesis, and on a DoD program that seeks to predict stock market activity from trade, news, and external market data.
For Lau and Kuzdeba, participating in the crowd-sourced challenge and aiding in expanding capabilities for the intelligence community was a rewarding journey, but the most exciting factor was working to exploit the mechanics of the human brain, which they say, still remains the most complex and powerful computer we possess.