University academics are using facial recognition technology to predict when a person is telling a lie.

Silent Talker is a lie detector device built by experts at Manchester Metropolitan University which monitors a person’s face when they are asked a series of questions then classifies their behaviour.

Jim O’Shea, a senior lecturer in computing, maths and digital technology, came up with the device which looks at “fine grained gestures” – those that the human eye does not necessarily detect.

During extensive research, the system has learned by example to identify behaviours that suggest lying, O’Shea says, adding that it has been 87 per cent accurate based on sample groups.

He tells BusinessCloud it would be impossible to list the exact gestures people make when lying.

“We tried converting the learning to a rule-based system but that would need thousands of rules to explain what’s going on,” he says.

“It doesn’t look at smiles but at the things that happen in between the gestures – a change of gaze or an eye moving from half open to fully open."

Silent Talker

The patented system began as a research topic in 2000 and a limited company was formed in March 2016 after O’Shea had identified the right investors.

He is about to start a Horizon 2020-funded project to use the system at border crossing points in Europe.

READ MORE: How airports are using facial recognition technology

“The potential applications commercially are enormous,” he adds.

“There are big opportunities in security, in war zones where there is risk of green on blue attacks, at airports or in schools to protect from random shooters.

“I’ve also been asked if we can do a system for traders in the City to predict when somebody is going rogue.

“People might have an occasional interview with the system and we’re train it to detect precursors to this activity happening.

“The idea is to intervene to stop these things happening in the first place.” 

BELOW: Flick through the October edition of BusinessCloud's interactive digital magazine

BusinessCloud digital edition