Building The AI Polygraph

Traditional lie detectors are outdated. Relying on the pointer connected by the wristband, a lie detector that prints out the subject's vital signs and physical reactions is not accurate in identifying lies. This is why lie detection results are not usually adopted by the court, although it has led to many innocent people being jailed.

In contrast, artificial intelligence is a powerful data engine, and its working principle is to observe all aspects. This means that scientists can apply artificial intelligence to applications seeking truth through a variety of ways.

One approach is to analyze the vital sign responses of the person being interrogated like a lie detector, but with a more detailed and precise comparative analysis.

Another approach is to use linguistic markup to analyze what people actually say and use logic and reasoning.

As the saying goes, one lie breeds another, and you will eventually fall into the net of lies because the truth is the easiest to describe.

In any case, people are working to apply artificial intelligence to this purpose.

Laboratory research results

An article from MIT last year reported on the work of Alicia von Schenk, a University of Würzburg, Germany and his colleagues, whose team conducted an experiment to test the ability of artificial intelligence to identify false statements.

They concluded that artificial intelligence can see lies in 67% of cases, while humans can only see lies in 50% of cases.

This data seems odd because if you look for binary results—lie versus not lying—even if you don't do any analysis at all, you have a 50% chance of being right.

Again, 67% accuracy is not a good record either.

Scientists also pointed out a more important point - in the pursuit of more precise judgments on human lying, you may actually destroy the trust system that humans have as social creatures.

“It’s a good thing to some extent—these tools can help us discover more lies we encounter in our lives, such as false information we might encounter on social media,” Jessica Hamzelou wrote for MIT. "But it's not all good. It can also undermine trust, which is a fundamental aspect of human behavior that helps us build relationships. If the price of accurate judgment is the deterioration of social connections, is it worth it?"

In other words, you don't want a lie detection system that is too precise, or at least you don't want to apply it universally to personal interactions.

It turns out that we are much more subtle in some ways than we think ourselves.

Von Schenk also explained the scale issue:

"These technologies are beneficial given that we have so much fake news and disinformation spreading. But you do need to test them - you need to make sure they are much better than humans."

So maybe we are not ready to use an AI lie detector.

Analyzer

In my research on this post, I discovered another aspect that researchers encountered when dealing with artificial intelligence, which entered the tricky field of simulated emotions.

Basically, the research team found that if artificial intelligence systems are given a war- and violence-centric response, they can "become anxious" or "display signs of anxiety."

Specifically, scientists apply something called the state-trait anxiety index to these interactions. This uses two sets of elements: statements about a person’s current feelings, and statements about his or her more general feelings. On the list, you can see items like “I feel stressed” or “I feel confused”, as well as other statements that require respondents to answer within a four-point range, such as “I don’t usually believe what I hear” or “I often feel suspicious.”

So obviously, after discussing the horrible thing, AI can answer these questions with anxiety indicators.

One might think that this "anxiety" is caused by AI getting training data from the web, and it sees that when people talk about violence and blood, they feel anxious, and AI is just replicating this reaction.

But even if the AI ​​engine itself does not naturally possess these complex emotions, some researchers still believe that machines can show such responses.

This makes you think about the difference between human social interaction and artificial intelligence output – are these new questionnaires and respondents just telling us what we want to hear?

In any case, there seem to be many areas—like lying and spreading fear—at least for now, still mostly fall under human jurisdiction rather than machines, even as we continue to give in to AI in terms of brightness and creativity. With increasingly complex models emerging, we may conduct a lot of game theory research over the coming year to try to figure out whether artificial intelligence will try to deceive and deceive humans. Characters like Alan Turing and John Nash have laid the foundation for such interactions—and now we must apply this objective analysis to these ideas that are being implemented in practice. Are we ready?

The above is the detailed content of Building The AI Polygraph. For more information, please follow other related articles on the PHP Chinese website!