People tend to make snap judgments on each other in a single look and now an algorithm claims to have the same ability to determine trustworthiness for obtaining a loan in just two minutes.
Tokyo-based DeepScore unveiled its facial and voice recognition app last week at the Consumer Electronics Show that is touted as a ‘next-generation scoring engine’ for loan lenders, insurance companies and other financial institutions.
While a customer answers 10 question, the AI analyzes their face and voice to calculate a ‘True Score’ that can be help companies with the decision to deny or approve.
DeepScore says its AI can determine lies with 70 percent accuracy and a 30 percent false negative rate, and will alert companies that fees need to be increased if dishonesty is detected.
However, scientists raise concerns about bias saying the app is likely to discriminate against people with tics or anxiety, resulting in these individuals not receiving necessary funds or coverage, Motherboard reports.
Dr. Rumman Chowdhury, founder of Parity AI, an algorithmic bias auditing platform, told Motherboard: ‘You can say that, in aggregate, there are some general trends in human beings when they are lying.’
Scroll down for videos
While a customer answers 10 question, the AI monitors micro-movements and stress shown on the person’s face to calculate a ‘True Score’ of trustworthiness that can be help companies with the decision to deny or approve
‘It doesn’t disaggregate to an individual human being. I might touch my nose because it’s a nervous tic that I’ve developed. It doesn’t mean that I’m lying.’
Scientist are using AI to read people’s faces in order to uncover hidden truths, such as political affiliation or sexual preference – so it may not come as a surprise that one system is trained to determine trustworthiness.
The DeepScore AI, according to the company, watches a customer’s face and listens to their voice.
‘When you tell a lie, you feel stress, your eye/mouth moves and your voice will skew.’ DeepScore CEO Shirabe Ogino shared in a statement.
DeepScore says its AI can detect lies with 70 percent accuracy and a 30 percent false negative rate. It uses a person’s facial expressions and voice cues to calculate a ‘True Score’
Ogino explained to Motherboard that DeepScore’s technology is based on more than 200 research papers about the correlation of micro-movements and stress with a person’s trustworthiness.
On the other hand, researchers say facial expressions are not indicators of mental state and vary from person to person and even throughout different cultures.
In addition to that, facial recognition has also been found bias towards women and people with darker skin color.
Ogino seems to be aware of the issues that plague AI and told Motherboard that DeepScore has been calibrated to address physical differences.
He also noted that DeepScore is not the final say for decision-making among lenders and insurers, but just a single part leading to the final verdict, saying ‘no algorithm is 100 percent.’
However, scientists raise concerns about bias saying the app is likely to discriminate against people with tics or anxiety, resulting in these individuals not receiving necessary funds or coverage
DeepScore shared its pitch deck with Motherboard, with one slide laying out how the AI is used by insurers. If the app believes the customer replied dishonestly to any of the questions, it will alert companies to ‘fee increases or additional examinations’
DeepScore shared its pitch deck with Motherboard, with one slide laying out how the AI is used by insurers.
‘To detect ‘if [an applicant is] diagnosed with cancer but hiding it since it’s at an early stage.’
If the app believes the customer replied dishonestly to any of the questions, it will alert companies to ‘fee increases or additional examinations.’
Along with bias, DeepScore has also raised the alarm among privacy advocates over the firm’s lack of privacy policies on its website – there are no protocols or regulations shown.
