Polygraph+ would add machine-learning scoring systems to a process that still relies heavily on measurements such as blood pressure, pulse, breathing and perspiration. The program also calls for “standoff sensing,” which would collect physiological signals without requiring sensors to be physically attached to the person being evaluated.
Earlier Pentagon work provides some indication of how that could function. In 2023, the Defense Innovation Unit selected Presage Technologies and Altec Research to develop prototypes for deception detection. Presage says its technology can estimate heart and breathing rates using ordinary cameras, while an Altec prototype shown by the DIU tracked signals including head movement, facial skin temperature and pore activity.
The federal government already conducts tens of thousands of polygraph examinations each year as part of personnel screening. The DCSA handles background investigations and other vetting work across the government, while the Department of Defense has a workforce of about 2.8 million people.
The proposed investment arrives as the Pentagon has increased its use of polygraphs under Defense Secretary Pete Hegseth. The New York Times reported in September that about 50 Joint Staff officers underwent testing following news reports about depleted U.S. weapons stockpiles during the war with Iran.
At the same time, researchers who study deception detection continue to question whether adding AI can overcome longstanding problems with the underlying method. “It’s a misguided effort to reduce the complex to something that is tangible,” said Kyri Kotsoglou, a professor at Northumbria Law School who studies polygraphs in the justice system.
Scientific scrutiny of conventional polygraphs stretches back decades. In 1983, Congress’s Office of Technology Assessment found little evidence supporting their use for employee screening. A 2003 U.S. National Research Council review described the evidence for polygraph effectiveness as “weak at best.”
The American Polygraph Association puts accuracy between 80% and 94%. But the National Research Council warned that even a system operating within that range could generate large numbers of false positives when applied to a workforce as large as the Defense Department’s.
Other problems include differences in how examiners interpret results and the possibility that trained subjects can manipulate physiological responses. One known technique involves deliberately increasing the body’s reaction during baseline questions, making comparisons with more consequential questions less useful. “If you know how it works, you can beat it,” said Sophie van der Zee, an associate professor at Erasmus University who studies deception.
Van der Zee argues that polygraphs can still have a deterrent effect because some subjects confess before testing begins. That effect, however, depends partly on people believing the test can reliably detect deception. “But that only works if people think a polygraph works,” she said.
Researchers have spent decades experimenting with other approaches, including thermal imaging, pupil monitoring, brain scans, eye tracking, voice analysis and body-motion measurements. Systems such as Silent Talker and the U.S. border-focused AVATAR project attempted to combine several signals, but those efforts did not develop into broadly reliable lie-detection systems outside laboratory settings.
AI could theoretically process more signals at once and identify patterns that human examiners might overlook. Van der Zee describes deception detection as an attempt to capture three areas: physiological stress, cognitive load and the deliberate effort involved in concealing information. Traditional polygraphs primarily measure the first.
A multimodal AI system could potentially analyze signals associated with all three. But researchers say that creates another fundamental problem: training an algorithm requires reliable examples of what counts as truth and deception. “Even if you have all the records in the world from polygraph tests, you don't know whether those polygraph tests are right or not,” said Marion Oswald, a professor of law.
Kotsoglou described combining AI with polygraph testing as “the worst of both worlds,” arguing that algorithmic analysis does not resolve uncertainty in the measurements being used as its foundation.
Van der Zee makes a similar point about decades of attempted technological fixes. “There is still no Pinocchio’s nose,” she said.
If Congress approves the Pentagon’s request, Polygraph+ would move that long-running search into a new phase centered on AI scoring and remote sensing. The core challenge will remain the same: determining whether more sophisticated analysis of physiological signals can reliably distinguish deception from the many other reasons a person’s body may react during questioning.
This analysis is based on reporting from MIT Technology Review.
Image courtesy of Popular Science.
This article was generated with AI assistance and reviewed for accuracy and quality.
