Last year, the Dutch National Police launched a system that will predict crimes using accurate and specific data from criminality. The Dutch National Police have already experimented with this Crime Anticipation System (CAS) in Amsterdam and The Hague. (NOS, 2017) With this system, the police use previous data and real-time data to make decisions about high-risk profiles. Like any other machine learning system, the CAS system classifies people as ‘low-risk’ and ‘high-risk’, and also defines ‘hotspots’ anywhere in the city. An example of the use of AI to predict crimes is by using an algorithm that uses ‘crowd analysis’ to identify suspicious patterns of individuals to classify them into a ‘low-risk’ or a ‘high-risk’ group: ‘If someone buys a kitchen knife that’s OK, but if the person also buys a sack and a hammer, that person is becoming suspicious’ (Martin, 2017).
Benefits
‘Predictive analytics that incorporate social factors and local demographics can play a significant role in enhancing the intelligence-led law enforcement that will help police anticipate crime, tackle chronic recidivism and manage risk more effectively’ (Accenture, n.d.). By using the machine learning capabilities, the police can make more informed decisions, which lead to better use of police resources.
Problems
We can use machine learning and algorithms to try to figure out who are going to be the bad types; which are the people to focus on. But those algorithms and machine learning capabilities have to be based on data we define to be sufficient as training set. We, humans, define what are high-risk characteristics. The problem with these machine-learning systems is that the National Police might be provided with striking gender and racial biases. People might be defined as ‘high-risk’, because they are part of a class higher ranking in criminal statistics, and can be precautionary arrested while having done nothing wrong.
Another problem is that criminals might learn how these machine-learning systems work, and create situations where attention is drawn to a certain place, away from where these criminal actually planned a criminal activity.
If we can figure out where the trouble is coming from, and therefore not intrude on the lives of people that are not going to be causing trouble, that is terrific. Yet again, the trick is to be honest about what is available to us, to have a frank conversation about it, to developed rules we can all see upfront, and apply those rules to make policing the best it can be.
References
Accenture, (n.d.) Policing gets smarter. Retrieved September 15, 2017, from https://www.accenture.com/us-en/insight-outlook-policing-gets-smarter-public-safety-government
NOS, (2017, May 15). Politie gaat misdaad voorspellen met nieuw systeem. Retrieved September 15, 2017, from https://nos.nl/artikel/2173288-politie-gaat-misdaad-voorspellen-met-nieuw-systeem.html
Martin, S. (2017, July 24). AI POLICE STATE: China to use technology to predict crimes BEFORE they happen. Retrieved September 15, 2017, from http://www.express.co.uk/news/science/832390/AI-minority-report-POLICE-China-predict-crimes-BEFORE-they-happen
Given the current state of affairs, when many people live in fear from the next terroristic attack, the responsible authorities need to find new ways to prevent such incidents from happening. Because though the last few year, we have all seen that the current approaches and technologies are not develop enough to ensure out safety. Thus, when I saw the title “Predictive policing”, I was like “okay, this is a step in the right direction”. The use of machine learning to identify suspicious behaviors will definitely help police anticipate and even prevent crime but at what cost. As you said, we are the ones that define what is considered a high-risk individual and we, as humans, are led by numerous biases. It is not a secret the due to the recent events, people have become extremely prejudiced when it comes to Muslims and I have personally observed many situations when the authorities question and attest Muslims without a valid reason but only because of their origin. Now, imagine what the effect of using predictive policing will be in a country as the Netherlands, where the Islam is the second most popular religion. Thus, I completely agree with you have before implementing this ML-based system, the police should thing of how to overcome the problems with the biases we have all displayed recently.
Thank you for sharing this insight! I would 100% vote for predictive crime analysis, for the application of ML-based crime prediction has already proved empirically efficient in the US( G.O.mohler,2015), and the considering abrupt terrorist attacks worldwide, which claimed lives of the innocent, the precautions backed up by algorithms is of great significance.
As for existing problems, the author fears that person-based algorithm might aggravate current gender and racial biases, which go against civil rights. However, I think before we cry for equality, we can prove by unbiased statistics, adding races and gender, for example, as factors to impartially conduct analysis. If the result proves no significant correlation between factors and criminal behaviors, then done. If certain correlation do exist, we won’t rush to someone’s house and hold him/her in custody, but instead keep a closer eye on the specific person. Moreover, there are still other types of algorithms such as location-based and other ones, such as proximity of liquor store. But those data contains business code, making it a bit difficult for fully implementation. Another source of data worthy of mentioning is social media. Looking back to school violence and murders recently, many murderer have even explicitly revealed their violence inclination on social media(Jeremy, B.2017). Similarly we have valuable information from cell-phone calling records. Despite the fact that the collection of those private data might invoke dissatisfaction of the mass, but if the police promise justifiably the lawful use of data, people can understand and accept that, since it is for the well-being of the whole society.
Overall, the precision of prediction model do needs further improvement, and the initial costs( including Trial and error) tremendous, but it is definitely worth of investment. And as the counterreconnaissance ability of criminals advances, only if we consistently moves forward with intelligent prediction model can we prevails over criminals.
Sources:
https://www.cfr.org/blog/predictive-policing-not-predictive-you-think
http://amstat.tandfonline.com/doi/abs/10.1080/01621459.2015.1077710?journalCode=uasa20#.WcVr6q17FE4
https://www.insidehighered.com/news/2017/05/23/arrest-u-maryland-student-stabbing-death-bowie-state-student-shakes-both-campuses