China facial image criminality inference

Released: November 2016

Can you improve this page?
Share your insights with us

According to Xiaolin Wu and Xi Zhang, researchers at Shanghai Jiao Tong University, software can automatically detect with 85.5% certainty that a human is a criminal by analysing their facial features.

The research study results (pdf) indicate that people with smaller mouths, curvier upper lips and closer-set eyes are more likely to be criminals. The research looked at 1,856 faces of Chinese men aged 18-55 with no facial hair, facial scars or other markings, of which 730 belonged to criminals.

Wu and Zhang stated in their study that they they do not intend to or are not 'qualified to discuss or debate on societal stereotypes', they stand accused of conducting physiognomy, phrenology, and pseudoscience.

They also said their system cultivates 'no biases whatsoever due to past experience, race, religion, political doctrine, gender, age, etc,' before admitting it should be tested using a dataset of different races, genders and facial expressions before it could be implemented on a broader scale. 

Given the software's potential for inaccuracy and bias, there are concerns that the system could wrongly identify criminals if used in the real-world. 

In 2020, a similar study by researchers at Harrisburg University had to be withdrawn after an academic and industry backlash.

Developer: Xiaolin Wu; Xi Zhang; Shanghai Jiao Tong University
Country: China
Sector: Research/academia
Purpose: Recognise/predict criminality
Technology: Facial analysis; Computer vision; Deep learning; Neural network; Machine learning
Issue: Accuracy/reliability; Bias/discrimination - race, gender, age, income; Ethics Transparency: 

Page info
Type: System
Published: October 2021
Last updated: January 2023