Can an AI lie detector tell when you’re fibbing?

Artificial intelligence is everywhere—but here’s a use you may not have considered: lie detection. It sounds like science fiction, but such an AI system is possible. The question is: How accurate can it be? Rada Mihalcea, a professor of computer science and engineering at U-M, has worked on deception detection for about a decade.

Mihalcea’s used 121 video clips from actual court cases and the corresponding transcripts of what they said. About half represented deceptive statements, and half truthful. They then used this data to build machine learning classifiers that discovered verbal and physical cues indicating that a person was perhaps not being truthful—an AI “lie detector” in other words.

But Mihalcea’s work is “not perfect,” she concedes. The system her team developed has between a 60 to 75 percent accuracy rate. “As a researcher, we are excited we were able to get to 75 percent.” But looked at another way, that’s an error rate of one in four. “I don’t think it’s ready to be used in practice.”