Welcome back, AI prodigies!
In today’s Sunday Special:
📜The Prelude
🫀History of the Polygraph
👁️Lies in the Eyes
🔑Key Takeaway
Read Time: 7 minutes
🎓Key Terms
Machine Learning (ML): Leverages data to recognize patterns and make predictions without being explicitly programmed to do so.
Neural Network (NN): A network of interconnected nodes that process large amounts of complex data to identify hidden patterns.
🩺 PULSE CHECK
If machines could read minds, should they decide who’s guilty?
📜THE PRELUDE
Imagine you’re a detective sitting across from a man suspected of murder. As fluorescent bulbs flicker overhead, the windowless interrogation room hums with an electric murmur. The air smells faintly of disinfectant, barely masking the sour smell of the damp carpet underfoot. A yellow legal pad, paired with a pen sharp with purpose, rests on the cold, cracked, and chipped metal table between you.
🚓 You begin: “What did you do after work yesterday?”
🚗 He responds: “I left work around nine o’clock and drove straight home.”
🚓 You nod slowly: “And once you got home?”
🚗 He shrugs slightly: “I passed out on the couch watching a movie.”
🚓 You quickly clarify: “What movie?”
🚗 He confidently chuckles: “Shutter Island. I love Leonardo DiCaprio.”
As you listen, his story feels too succinct, and his voice too measured. He sits with a calm, charming, and almost commanding composure. His eyes remain fixed on you, as if they’re trying to stop themselves from wandering. If he’s lying, he’s great at it. Nearly every word, every pause, and every gesture feels controlled and confident.
Suddenly, the screen beside you blinks: “Deception Score = 22.” You stare at him, stunned. Nothing about his facial expressions, body language, or speech patterns gave him away. And yet, the AI doesn’t believe him. Today, a new wave of AI-powered lie detection methods promises to catch lies before they’re even uttered. So, how does it all work?
🫀HISTORY OF THE POLYGRAPH
⦿ 1️⃣ The Blood, Breath, and Sweat?
In 1915, American psychologist William Moulton Marston proposed that lying provokes stress, which causes variations in BP: the pressure of blood against vessel walls as the heart beats. In 1939, American inventor Leonard Keeler added two critical components to detect lies:
These additions produced the modern polygraph. Despite ongoing ethical concerns about examiner bias and coerced confessions, the modern polygraph became firmly linked to law enforcement by the mid-20th century, viewed as a technological upgrade over human judgment.
⦿ 2️⃣ The Control Question?
In 1960, Chicago lawyer John E. Reid developed a structured protocol called the CQT, where a murder suspect might be asked:
💬 Control Question: “Have you ever lied to get out of trouble?”
💭 Related Question: “Did you kill the person at the park last Friday?”
If the murder suspect conveys stronger physiological reactions to the direct criminal accusation, the results are interpreted as deceptive.
⦿ 3️⃣ Is It Admissible?
The courts view the modern polygraph as inadmissible because it doesn’t provide substantive proof, which refers to evidence that answers: “Did this happen?” The modern polygraph primarily measures physiological responses like sweating and breathing, which are also the physical changes your body makes when faced with fear, anxiety, and nervousness.
👁️LIES IN THE EYES
⦿ 4️⃣ Eyes Are Windows to the Soul?
Today, the most popular AI-powered lie detection method is EyeDetect. Since 2009, 12 peer-reviewed studies have been published evaluating EyeDetect’s performance, consistently reporting accuracy rates between 86% and 88%. While this significantly outperforms the accuracy of the modern polygraph, it falls short of the legal standard of conclusive evidence.
⦿ 5️⃣ How It All Works.
🔴 Data Collection:
To record involuntary eye movements, EyeDetect leverages a high-resolution eye tracker while a participant answers a series of yes-or-no statements. The high-resolution eye tracker captures anything from a rapid series of micro-blinks to jerky eye movements that shift gaze between fixation points.
Behavioral vision scientists isolate the raw eye movements from the high-resolution eye tracker and convert them into a gaze stream, which captures blinks, fixations, saccades, and pupil size. This process enables them to extract Features, such as how long a participant’s pupils dilate before lying.
🟡 Pattern Detection:
An ML model is trained on thousands of involuntary eye movement recordings, each labeled with several different Features like “anxious” or “relaxed.” Over time, the ML model learns to associate specific eye movements with certain emotions and behaviors. As a result, when EyeDetect feeds new eye movements into the ML model, it’s able to instantly classify them accurately.
Using ML models to classify involuntary eye movements works, but a single participant can produce thousands of overlapping eye movements. The sheer volume of eye behavior can quickly exceed what behavioral vision scientists can manually interpret. More importantly, the behavioral vision scientists only extract Features they already expect to see. This means EyeDetect’s ML models are only trained to classify the expected.
🟢 Pattern Recognition:
DL models are designed for situations where the patterns are too subtle, too numerous, or too complex for humans to label. Instead of telling a DL model what Features to look for, behavioral vision scientists let it independently discover them by stacking multiple layers of learning on top of one another to form the NN.
In a NN, each layer transforms the data slightly, emphasizing certain patterns and suppressing others. By the time the involuntary eye movements reach the final layers, the DL model detects patterns of patterns that convey more specific actions like emotional arousal triggered by deceptive intent that’s masked by deliberate nervousness.
🔑KEY TAKEAWAY
The truth always seems so elusive. Now, AI-powered lie detection methods promise to finally capture it. They already outperform the modern polygraph, but remain far from perfect. Perhaps that’s for the better, because a mind-reading machine would wield too much societal and cultural power.
📒FINAL NOTE
FEEDBACK
How would you rate today’s email?
❤️TAIP Review of The Week
“I like that the narrative is accompanied by hard data.”
REFER & EARN
🎉Your Friends Learn, You Earn!
{{rp_personalized_text}}
Share your unique referral link: {{rp_refer_url}}
