People lie every day, and it can be hard to understand whether someone is telling the truth, from harmless white lies to things like you call them sick when you're not, to more serious deceptions that can end up in front of the courthouse.
A group of scientists based at the University of Sharjah in the United Arab Emirates have discovered that AI could be the solution to this truth detection problem, but only if the machine can explain cultural and gender differences.
“Our aim was to conduct a comprehensive review of publications focused on computational predictions of deceptions, particularly using a machine learning approach,” the scientist wrote.
The team conducted a meta-analysis of 98 papers published between 2012 and 2023.
In particular, this study focused on convolutional neural networks (CNN) programs. These networks are a type of machine learning program that mimics the role of the visual cortex in the brain to recognize images.
These papers were compared to traditional approaches to AI-free lie detection, such as diagnostic questions, expert analysis, and the use of evidence.
As part of the investigation, the team also analyzed 35 short videos and two hours of footage.
“We conducted a comprehensive analysis of deception detection and provided a clear overview of field contributions and limitations,” the authors say.
Deception detection is a growing field, and many scientists want to learn more in the hopes that it will lead to a more objective understanding of human behavior.
Accurate detection of deceptions is also important for areas of society, such as legal systems, where false outcomes can be substantial.
“In such circumstances, mistaken truth and vice versa can have a major impact on the parties involved and society as a whole,” the author points out.
Compared to traditional methods of lie detection, AI and CNN programs have increased the efficiency of discovering deceptive statements and lies.
“More than half of the papers achieved accuracy of over 83%, particularly high performance, highlighting the effectiveness of the machine and the effectiveness of the deep learning models adopted,” the author writes.
“Overall, the ML-driven approach shows performance comparable to traditional methods while achieving increased efficiency.”
While this may seem promising, their study also identified potential issues that could arise using AI-based deception detection methods.
One of the main limitations of using these algorithms to detect deceptions is the inability to consider the role of gender, culture and language.
“Unless culture, language and gender are considered, the generalization of the findings can limit generalization to diverse populations,” the author writes.
The team outlined this could be due to a small scale lack of diversity within the dataset where many of these machine learning programs are being developed.
The research team at the University of Sharjah Expert System with Applications.
