Quit spoofing! Algorithm can spot lies in emails and dating sites
Published 05/07/2016 | 14:04
Researchers have created a computer program that can detect lies, be it an email, dating profile or visa application.
The algorithm created at City University London can tell if a person is lying just by analysing their word use, structure and context, according to the researchers.
To create the algorithm, researchers compared text in tens of thousands of emails that contained lies and truthful contents.
The comparison revealed that people who are lying are less likely to use personal pronouns - such as "I", "me", "mine" - and tend to use more adjectives, such as "brilliant" and "sublime". Reasons for this language use could be that liars try to dissociate themselves from the content of a message, while clouding its meaning in unnecessary description.
Other clues that someone is lying include linking sentences to each other so that thoughts appear to be connected, and mirroring the sentence structure of the person they're communicating with.
The algorithm is better at detecting lies than the average human. People manage to spot a lie 54 per cent of the time, according to the researchers, whereas the computer lie detector detects it 70 per cent of the time.
"Humans are startlingly bad at consciously detecting deception," said Tom van Laer, one of the researchers who specialises in marketing at the Cass Business School, City University London.
It was designed to help organisations protect themselves against scam or phishing emails, which can lead to identity theft and financial loss.
"We want to put it to work to fight fraud wherever it occurs in computerised content," said van Laer. "And as the technology evolves, we want to wholly automate its warning."
But before it can be put into practice van Laer said they want to increase the algorithm's accuracy.
When it comes to messages from a contact that you know, the best ways to spot a lie include looking out for deviations from the normal pattern of writing, non-committal phrases, such as "pretty sure", "maybe" and "probably", changes in tense in the middle of a story, and qualifying statements, such as "to be honest" and "I hate to tell you".
In a separate study, researchers at the University of Michigan taught computers to detect lies using courtroom videos. The machine learning algorithm was able to identify people telling the truth 75 per cent of the time.