Go and tell your boss that we have been charged by society with a sacred quest. If they provide us with remuneration, they can join us in our training quest for the Holy Grail.
Well, I'll ask him, but I don't think he'll be very keen. He's already got one, you, see?
"Monty Python and the Holy Grail” (Adapted)
The quest to reliably detect deception has long captivated psychologists, law enforcement, and technologists alike. While most lies in everyday life are trivial, in legal and regulatory settings the consequences of mistaken judgments—misclassifying truth as deception or vice versa—can be profound. The stakes are high: wrongful convictions, investigative misdirection, and erosion of public trust.
In recent years, this pursuit has taken a technological turn. A 2025 review found over 540,000 academic papers on “deception detection,” a testament to both its urgency and elusiveness. The use of machine learning (ML), particularly neural networks, is gaining momentum, with some models now claiming 80% accuracy. But this technological progress is not without deep ethical and practical concerns.
AI systems trained to detect deception promise consistency and speed. Some outperform traditional methods under experimental conditions. However, even an 80% confidence rate is a 20% error rate which can be devastating in justice settings. That’s 1 in 5 people misclassified—a margin far too high when liberty is at stake.
Furthermore, algorithmic opacity—often described as the “black box” problem—raises questions of accountability. If a system labels someone deceptive, yet cannot explain why, how can that decision be challenged in court? Who is responsible if that label leads to legal consequences? The interviewer? The agency? The developer? Existing legal frameworks are not equipped to answer these questions.
Compounding the risk is automation bias—our tendency to defer to machine judgments, especially under pressure. In investigative settings, this could erode professional judgment and reduce critical thinking skills, much like how rigid adherence to the Reid Technique has shaped decades of interview practices in the U.S.
In contrast, science-based interviewing models, such as the PEACE framework, offer an ethical, empirically grounded approach. They prioritize:
- Information gathering over confessions
- Open-ended questioning and memory science
- Strategic (but ethical) use of evidence
- Rapport and respect over coercion
Instead of asking “Is this person lying?”—a binary trap prone to error—the science-based interviewer asks: “What do I need to know to fully understand this account?” This reframing removes the pressure to “catch a lie” and centers the conversation on obtaining complete, accurate, and reliable information.
Deception, if it exists, reveals itself through inconsistencies with verifiable facts—not through guesswork based on body language or emotion. Interviewers trained in this model move away from pseudo-scientific behavioral analysis and toward cognitive psychology, narrative recall, and structured inquiry.
ML does have a place in this ecosystem—as a support or training tool, not as judge or interrogator. If carefully designed, it can:
Highlight patterns for interviewer reflection
Assist in training by surfacing cues linked to cognitive load
Analyze large volumes of data post-interview
But AI must never replace human judgment, especially in high-stakes contexts. As we’ve seen with COMPAS—a predictive policing tool whose algorithmic decisions have been widely challenged—overreliance on opaque systems can embed bias and erode accountability.
“The goal of science-based interviewing is not to find shortcuts to the truth, but to create the conditions under which the truth is most likely to be revealed.”
The allure of the “Holy Grail” of deception detection—fast, reliable, and objective—is strong. But the real prize may be something more achievable and ethical: skilled, science-informed interviewers using tested methods to uncover the truth without coercion.
If you’re in this field, ask yourself: Are you seeking a confession—or understanding? Are you building pressure—or rapport? Are you chasing a machine’s certainty—or cultivating your own professional insight?
References;
- Elbatanouny, H. et al. (2025). A comprehensive analysis of deception detection techniques leveraging machine learning. Expert Systems with Applications, 283, 127601. DOI
- Vrij, A. (2008). Detecting Lies and Deceit: Pitfalls and Opportunities.
- Meissner, C. et al. (2012). Interview and interrogation methods and their effects on true and false confessions. Behavioral Sciences & the Law, 30(6), 689–705.
- von Schenk, A. et al. (2024). Lie detection algorithms disrupt the social dynamics of accusation behavior. iScience, 27(7). DOI
- BU (2023). Do Algorithms Reduce Bias in Criminal Justice?
