Two researchers from China want to use artificial intelligence to identify criminals – even before they commit a crime
In the 19th century, the physician Cesare Lombroso asked himself the question whether it is possible to identify a criminal before he commits his crime?
With his work “L’Uomo delinquente” he founded the studies on the classification of offenders. For him, criminals belonged to the group of the “insane and primitive”. Lombroso was convinced that criminals were inclined to look at their inclination externally. His theory of perpetrator types was already critically discussed in the 19th century.
His studies experienced a renaissance during the National Socialist era. With the race doctrine the theory was pseudo-scientifically exhausted to the point of the typical “pest of the people” or “master man”.
The idea of identifying criminals by their faces or physical characteristics was later rejected as ethically unacceptable.
The human being – a mixture of predisposition and education?
Until recently, the “investment-environment formula” was used, which implies that a person’s behaviour is influenced by his or her innate genes as well as his or her social environment and childhood – i.e. socialisation.
Which factor has a greater influence on later behaviour – heredity or upbringing – remained unclear, although exciting twin studies have been carried out on this. Identical twins that grew up separately are invaluable for such research. They have the same genes, but different socialization. With twin studies, scientists wanted to find out how genetically identical people develop under different social conditions.
For this reason, twins who grew up in different places – for example, adoptive parents in different environments – were particularly popular for research. The result? Identical twins often have similar behaviour, preferences and habits, although they grew up separately. For example, they often fell ill with the same disease at the same time – regardless of their lifestyle, for example smokers and non-smokers. These studies gave rise to the assumption that there is definitely a considerable genetic predisposition.
However, these did not provide a reliable answer as to whether crime is innate or learned behaviour.
With artificial intelligence the seemingly insoluble question is taken up again
With the help of artificial intelligence, two researchers have again addressed the question of the type of perpetrator. Xiaolin Wu and Xi Zhang from Jiao Tong University in Shanghai have taught an artificial intelligence to recognize criminals by images.
Police and Internet provided the photos
For this purpose they fed the artificial intelligence with about 1800 photos of Chinese men. 700 photos of them were provided by the police – photographed on them – criminals with criminal records. The researchers have another 1100 photos from the Internet – pictures from the professional network LinkedIn. 90 percent of the existing pictures were initially used to train artificial intelligence. The remaining photos were used to test the neural network.
Criminals look different – says the artificial intelligence
Their investigation revealed that three different facial features indicate criminal potential: A special curvature of the upper lip, a certain distance between the eyes and two special lines from the tip of the nose towards the mouth.
Research result is often criticised
There is talk of multiple bias of the data: On the one hand, the data basis is criticized, which is by no means representative, since neither other ethnic groups, nor age groups nor gender are represented.
Even the comparative material is said to be limping: In the LinkedIn photos, the persons photographed are beautifully dressed – with hair and make-up – often perfectly lit by a professional photographer and photographed with a tie or collar. The pictures of the criminals, on the other hand, are police photos, taken under completely different conditions – that is the nature of things.
Large data sets and new possibilities for analysis
Nevertheless, it is expected that such studies will be continued – for example, for face checks at airports and borders. The appeal of using artificial intelligence to predict behaviour and to recognise and prevent crimes in advance is too great.
Artificial intelligence and smart data analysis are simply too attractive for criminological research. After all, police files and statistics provide huge amounts of data that can be tested and evaluated.
Presumably there will be some more attempts in criminological research in the coming years to predict crimes using artificial intelligence.
At the same time, the study by the two Chinese researchers makes it clear how carefully the new technical possibilities must be used in order to avoid stigmatising individuals.