AI Lie Detection Tools Promises and Challenges in Identifying Truth and Falsehood
TapTechNews July 8 news, in recent years, people have encountered more and more challenges in distinguishing true and false information, especially with a large amount of fake news and exaggerated propaganda on the Internet. Studies have shown that humans do not perform well in judging true and false words.
According to TapTechNews, traditional lie detection means, such as polygraphs, have been criticized for their accuracy issues. Some people believe that artificial intelligence (AI) can help us improve the accuracy of identifying lies. One day in the future, AI-based lie detection systems may be able to help us identify false information on social media, evaluate various remarks on the Internet, and even screen the exaggerations in job seekers' resumes and interview answers. However, the problem is whether we will trust these systems and whether they are reliable.
German economist Alicia von Schenk and her team at the University of Würzburg recently developed an AI lie detection tool, and the accuracy of this tool is significantly higher than that of humans. They conducted relevant experiments to explore how people use this tool. The experimental results show that people who use this tool do improve in identifying lies, but at the same time, it will also lead to more people being suspected of lying.
Researchers mentioned in the paper published in the iScience journal that they asked participants to write weekend plans. Half of the participants were asked to lie and would get a small amount of money reward for convincing people to believe the lie. The research collected 1,536 statements from 768 people.
Subsequently, the researchers used Google's AI language model BERT to train 80% of the statements to help the algorithm distinguish true and false words. The test results show that the tool can identify the truth and falsehood of the remaining 20% of the statements with an accuracy of 67%. In contrast, the average accuracy rate of humans is only about 50%.
In order to understand how people identify lies with the help of AI tools, von Schenk's team divided another 2,040 volunteers into groups and conducted a series of tests.
In one of the tests, when people can choose to pay a small fee to use the AI lie detection tool and get a reward, only one-third of people will choose to use the it. Von Schenk speculates that this may be because people are skeptical about this technology or overestimate their ability to detect lies.
However, the one-third of people who choose to trust this tool show great dependence on it. When people actively choose to rely on this technology, they will almost always follow the prediction of artificial intelligence... They trust its judgment very much. Von Schenk said.
This dependence will affect our behavior. Usually, people tend to believe that others are telling the truth. This study also confirms this - although the participants knew that half of the statements were lies, they only marked 19% of them. However, when people choose to use the AI tool, the proportion marked as a lie has risen to 58%.
To some extent, this may be a good thing. Such tools can help us identify more lies in daily life, such as false information on social media.
But on the other hand, this will also destroy trust, and trust is the foundation of human behavior, which helps us establish good interpersonal relationships. If accurate judgment is based on the deterioration of social relationships, is such accuracy still valuable?
Secondly, the accuracy issue cannot be ignored. The researchers acknowledged that thei r goal is only to make the AI perform better than humans, but for scenarios such as the authenticity judgment of social media content or the resume review of job seekers, just better than humans may not be enough, because this will bring more misjudgments and accusations.
It is worth mentioning that there are also flaws in traditional lie detection means. The polygraph aims to measure heart rate and other physiological arousal indicators because people once thought that stress was the unique physiological reaction of liars. But this is not the case, which is why the results of polygraphs are usually not admissible in US courts.
Von Schenk pointed out that because AI lie detection tools are easy to apply on a large scale, the impact may be greater. The polygraph can only detect a few people per day, while the application range of AI lie detection tools is almost unlimited.
Given the large amount of false news and misinformation currently available, such technologies do have certain benefits, von Schenk said. However, we really need to test them strictly to ensure that their accuracy is far beyond that of humans. If the AI lie detection tool only brings a large amount of misjudgments, then perhaps it is a better choice to give up using it.