If you haven't read George Orwell's 1984 or seen Minority Report, there's something in this article that ebbs very close to the dystopian concept of "Precrime" in this new research published by researchers from Shanghai. That's the notion that people can be convicted of a crime by just having the intention to commit it, before they actually carry it out. Some criminologists have claimed in the past that they can recognise people more likely to commit crimes by just looking at their facial features.
The debate has been rife between criminologists and psychologists over the years, who have enlisted statisticians time and time again to analyse facial patterns and see if there is any merit to the seemingly absurd notion.
In 2016, we can try a new method of analysis. Researchers from the Jiao Tong University They took ID photos of nearly 2000 Chinese men between 18 and 55, half of which were criminals, and used 90 percent of them to train a neural network to recognise the difference between the criminals and non-criminals. They then tested the neural net on the remaining 10 percent of the images.
The results are unsettling. The neural network could correctly identify the remaining 200 images as criminals or noncriminals with an accuracy of 89.5 percent. This would require much further testing of course, and not just confined to one country or one sex, but the worry of course is how humans might use these machines in the future.
It's not hard to imagine how this could be applied to data sets like passport or driver license photos of an entire country, and flag some of the population as more likely to commit crimes.
This of course raises important ethical questions. How will we use the newfound data that machine-empowered statistics can give us? Will it look like George Orwell's vision and be full of fear, or will we use the technology to in fact change our society for good? This adds yet another ethical quandry on to the ever-growing pile of the early 21st century.
The results are unsettling. Xiaolin and Xi found that the neural network could correctly identify criminals and noncriminals with an accuracy of 89.5 percent. “These highly consistent results are evidences for the validity of automated face-induced inference on criminality, despite the historical controversy surrounding the topic,” they say.