TLO-35306 2018-01 Global Information Systems Management Group assignment
Dóra Bereczki Julien Haslé Daniel Ochs Jeanne Tremblay
ABSTRACT
Improving Predictive Policy with Artificial Intelligence Tampere University
Group assignment, 14 pages March 2019
This report examines the current usage of Artificial Intelligence to predict up-coming crime areas, so called Predictive Policing. In the first part we introduce the topic and review ethnical and ethical problems that are related to Predictive Policing. Based on the problems and the current situation we suggest a more proactive approach and how-to extent the model trying to avoid the problems discussed before. The last part covers how this approach can be applied in the real world and how other private institutions and com-panies are able to implement a similar system.
Keywords: Predictive Policing, Artificial Intelligence
1. INTRODUCTION
The AI currently being developed around the world has many capabilities. Some AI’s are used to improve the development of technology; however, many are still using for enter-tainment and do not provide much-added value when using AI. AI is very useful when it comes to making choices that humans are not able to make instinctively. The AI is well suited for planning complex tasks when it comes to finding the optimal solution in a large number of variants [White 2017].
Assign people to workstations, machines to tasks, develop a schedule, schedule manufac-turing processes, manage traffic, cable a building, assign tracks to trains, platforms to boats... All these problems, which are faced by manufacturers, transport companies, en-gineering companies, and training organizations, can be summed up in a single question:
find the optimal solution based on given resources. Those are problems that, if not too complex, are usually solved by hand. However, computer science and more specifically AI provide a multitude of solutions to solve these problems when they become complex [Ismail 2017].
In the context of public service interventions, but more specifically those of the police, the use of AI can be a real source of improvement. Indeed, the cyber-police strongly in-creases its capacities in the field of AI. An AI platform is being tested in the US police services and claims to be able to predict where crimes will be committed. The tool comes from a company called PredPol, which is short for "Predictive Policing". This company claims that the software can predict crime in an algorithmic way, based on the theory of broken windows applied to predictive fonts. Other companies such as Palantir, CrimeScan and ShotSpotter Missions, argue that the version integrating AI, at the predic-tive police level, goes beyond the traditional hot spot analysis, which involves reacting according to everything that has happened before, in a given place, whereas here it is more a question of anticipating what is likely to happen in the future [PredPol]
[ShotSpotter]. The establishment of such an AI could represent a real step forward in combating the rate of crime and criminality.
This paper is made as an assignment for the course TLO-35306 Global Information Sys-tems Management at the Tampere University of Technology. In the first part we will look at the use of PredPol and what they try to avoid. In the second part we analyse problems that come with use of predictive policy tools in general so we can then later look on how to avoid them. The focus in this report is on the ethnical and ethical prob-lems because in our eyes those are the biggest issues with predictive policy. After
ana-lysing the problems that can occur, we examine how PredPol’s technology can be ex-panded and optimized without raising those ethnical and ethical questions discussed be-fore. In the last chapter we suggest steps to implement possible solutions and how other fields and sectors are able to use these solutions as well.
2. HOW PREDPOL WORKS
First, it is important to mention how PredPol is working so far. They rely on a machine-learning algorithm. They use only three types of data, which are crime type, the crime location and the time the crime has happened. They find the information in historical events of the city where they wish to predict crime. Then, the machine-learning algorithm is able to find the most probable location of the crime. The location can be visualized in a Google Maps interface, where red boxes point to the location. An example is presented below. [Fig.1]
Figure 12. The Google Maps interface of PredPol
The boxes measure 150 meters by 150 meters. With this information, it is easier to see where the most important places are to send officers to patrol. Usually, the police make sure that their officers spend about 10% of their shift in the PredPol boxes, as it is still important to patrol everywhere else [PredPol].
In addition to this map, PredPol also has another program that helps with resource allo-cation. In fact, they created the PredPol Heat Map, which shows how much time the box is patrolled in total during the day. It can be used to see if the areas have too little officers, which could make it dangerous, or too much, which makes them lose their time. An ex-ample is shown below. [Fig.2]
Figure 13. Total Patrol Dosages
To PredPol, privacy and information security are very important. They use cloud-based software that is secured in many ways from hackers. Still, the data is easy to read and accessible quickly for the police department. It is also possible to share it with the local government. On their website they are saying that no personally identifiable information is ever collected or used and that they believe that protecting the privacy and civil rights of the residents of their communities is as important as protecting them from crime.
Therefore, they try to avoid raising questions about questions like ethical or ethnical ones are.
So far, PredPol is mostly used in the United States of America. In the state of Kent, the police department noticed that street violence has been reduced by 6% within the four months of testing this new technology [PredPol].
We think that some other data could be interesting to use in order to push this idea even further. The main goal would be to prevent crime in the locations PredPol suggests. With the data already used, it would be interesting to see if there are other links with what causes crime. By example, looking around in the highest crime zones could give clues on what kind of areas are dangerous. Are they close to a liquor store? Or not well enlight-ened? Finding out some links between the zones could be a way to prevent crime even more effectively than only using the three data points PredPol is familiar with. Then, the goal would be to resolve those issues in every place that has them. Hopefully, those places would become less dangerous.