Researchers from the Tokyo Institute of Technology recently published a study in the Journal of Artificial Life and Robotics exploring how machine learning can help detect deception.
Machine learning is a subset of artificial intelligence (AI) that involves using algorithms and statistical models to enable computers to learn and improve from experience without explicit programming. In other words, it is a method of teaching computers to perform specific tasks by learning from data, patterns, and examples rather than relying on predefined rules.
Detecting deception is important in various situations, such as interviewing crime victims or suspects, as well as interviewing patients with mental health issues. Sometimes, human interviewers may struggle to ask the right questions or accurately detect deception.
To address these challenges, researchers are attempting to develop an automated deception detection system using machine learning. The goal is to establish a fair and accurate system that helps interviewees tell the truth while correctly identifying true suspects without falsely accusing innocent individuals. The team, consisting of Kento Tsuchiya, Ryo Hatand, and Hiroyuki Nishiyama, focuses on utilizing facial expressions and pulse rates to identify deception.
In their study, they collected data from four male graduate students. Instead of using artificial interview scenarios, the researchers opted for a more natural approach. They showed the participants random images and asked them to freely discuss these images while making deceptive statements.
During these interviews, the researchers recorded the participants' facial expressions using a webcam and measured their pulse rates using a smartwatch. The participants were instructed to deceive the interviewer while speaking. After each video segment, the participants themselves identified which parts of the recording contained deceptive statements.
The researchers used a machine learning technique called random forest (RF) to build the deception detection model. They utilized all the collected data, including facial expressions and pulse rates, to create a dataset for training the machine learning model.
To evaluate the model's performance, they employed a 10-fold cross-validation technique. This involved dividing the dataset into ten parts, with nine parts used for training and one part used for testing. This process was repeated ten times, and average performance metrics were calculated, including accuracy, precision, recall, and F1 score (a performance metric that balances accuracy and recall).
The research results indicate that the deception detection model performs well. The accuracy and F1 scores for each participant ranged from 75% to 80%, with the highest accuracy reaching approximately 87%. Some common clues used by the machine to detect deception include changes in pulse rate, eye movements, and specific facial areas around the eyes and mouth.
The researchers believe that their machine learning approach can be a valuable tool for detecting deceptive behavior in human interactions. However, they note, "To obtain statistically rigorous results using our method, we essentially need thousands of recorded videos and richer sensor data, as well as the help of a large number of participants who may have different cultural backgrounds and neural differentiation states, which is difficult for us." Therefore, our focus is limited to providing some case study-style analysis based on the methods proposed in the existing literature, with the help of a small number of subjects."
Due to the difficulty of obtaining a large and diverse dataset, the researchers had to work with limited resources. This limitation may affect the overall strength of their findings and restrict the scope of their analysis.
In other words, they require detailed data from a large number of individuals to develop an accurate deception detection system. Due to the challenge of obtaining a large and diverse dataset, they focused on conducting a small-scale study with only a few participants. This limited their analysis but still provided useful insights that can lay the foundation for future research.
Source: https://www.psypost.org/2023/07/decoding-lies-with-ai-new-machine-learning-model-uses-facial-expressions-and-pulse-rates-to-detect-deception-167323
Alphabet Q1 paid subscribers reached 350M, fueled by YouTube & Google One. YouTube ad revenue slightly missed at $9.88B amid Premium growth. Total Q1 revenue hit $109.9B, beating forecasts.
Anthropic releases Claude connectors for Adobe, Blender, and Ableton, enabling AI workflow automation and app control. It also sponsors the Blender Development Foundation.
Apple introduces an App Store model splitting 12-month plans into monthly installments. On iOS 26.4+, it lowers upfront costs without reducing total price, adding billing tracking.
Google launches Aadhaar digital IDs in Wallet for India, enabling secure QR/NFC verification. A phased partner rollout targets key sectors, while passport-based credentials expand globally.
YouTube tests "Ask YouTube," an AI tool blending text and video for complex queries. Initially for US Premium users, it supports interactive follow-ups, expanding Google's AI search.
Google grants US DoD classified AI access, joining OpenAI and xAI in defense deals amid Anthropic's dispute. Contractual safeguards against surveillance and lethal weapons remain unenforceable.
Mistral AI launches Workflows, a Temporal-powered orchestration platform for secure, Python-driven enterprise AI automation featuring human-in-the-loop approvals.