AI & Policing
By: Jazmyn Stokes, Dempsey Wade, Julia Medici & Tyler Marchiano
We implemented an Artificial Neural Net to effectively identify areas of high crime. Specifically, our AI determines what are high crime areas, and how many crimes are going to happen in desired district of Chicago in hopes to decrease response time and increase arrests made. We applied a neural net that imports keras in order to run. Here is how we got there!
Our Motivation
We live in a time where tension between society and law enforcement is high, there has been countless protest of police and government officials. Our hope in creating our AI is to make police more effective by decreasing response time and placing more units in areas that have a tendency to experience higher crime rates. This will hopefully decrease police focusing in on minorities, in turn decreasing the backlash on police. There is also an idea by increasing the efficiency of police there will be an increase in arrest and decrease in crimes, making communities feel more safe.
Additionally, we would like to use the information outputted by our AI, to allow us to better understand “black box” predictive algorithms. From what we find we can make implications on the moral and ethical consideration while using a predictive algorithm in this way.
Our Dataset
We looked for datasets that were very informative, we were fortunate to have a topic where there needs to be an influx of data taken for government records. Our data set for this project is the Chicago police department’s records from 2001 to 2017 and contains information of each crime reported. Some of the relevant information in these records are case number, date, type of crime, community area, if an arrest is made, and if the crime involves domestic matters. In a past project by Dempsey Wade, our dataset being used has already been clean from having irrelevant and repetitive data. Dempsey also broke down the information by year so that our neural net would run faster and be able to train.
The Challenge with our dataset
After evaluating the dataset we were faced with two major points that could influence the outcome of our AI:
- The bias of the policing: Living in a society where past policing has, and will inevitably be skewed is something we have to take into consideration when working with our data.
- The amount of crimes that go unreported: Having crimes that go unreported also skew our data from being unrepresentative of the data we have.
Implementation
We implemented a neural net by importing Python Keras. We have had some experience with Keras and decided that for a project like this it would be better to use Keras instead of TesorFlow because of the lack of knowledge with it, we also were not able to make our own neural net due to a lack of time.
Since we were doing an extension from Dempsey Wade’s last project, we were able to eliminate the beginning steps of finding and cleaning the data.We began by creating a dataframe used to categorize the necessary information. This dataframe was named df1, and contained the district, year, and type of crime. This information is then passed to our Predict(x, df) function where x is a district number and df is the dataframe, df1. The function then creates a new dataframe (named df2) which contains the summation of crimes for that particular district. Afterwards this dataframe is passed into our Keras neural network, which trains the agent based on the summation of crimes for each year. The Keras neural network uses a Sequential model with relu and sigmoid activation functions to to carry out quick operations. The number of optimal epochs and batch_size varies with the district, but these are the best universally for a quick run. The agent then outputs the total predicted number of crimes for 2020 for that specific district.
Results & Analysis
We outputted the predicted amount of crimes per crime type for each district. After analyzing the data we had some observations.
We were able to train the Keras neural network to predict the quantity of every crimes for each district in 2020, we were able to look at those predictions and determine the crime hotspots in Chicago, based off which districts had the greatest number of predicted crimes. Our neural net outputs a predicted number of crimes of that type of crime in that specific district for 2020.
Breaking outputs out by district helps narrow the focus for police. The data outputted shows police the opportunity so it is known where certain hotspots are and what types of crimes will occur in that hotspot. From this police can make conclusions on why these numbers might have been outputted. Such as in smaller districts there are going to be less crimes because of the size. In areas with high populations there are going to be more crimes due to there being a higher probability of criminals being in that area. There is also going to be a correlation between the amount of public transportation and the amount of crimes, due to several factors one being a quick way for someone to leave the scene of the crime.
Utopian/Dystopian
In an utopian society there would be no more unsolved cases, we will be able to give cops the most accurate information to be able to ensure the most effectiveness. Whether that be stopping a crime before it happens or being in the area of the crime after in happens in hopes to catch the perpetrator. Being able to correctly inform police on where crimes will happen will hopefully eliminate crimes all together. In our utopian society, it would be known that before you even commit a crime or right after you commit a crime that a police officer would be there to arrest you, therefore creating an environment where people wouldn’t even risk committing crimes because they would know they were going to get caught. This would also decrease the amount of cops, there would only be the amount of cops necessary for the night that is most active. This is because cops will be so efficient that were will be no need for wasted cops. Instead of decreasing cops it might also lead to an increase of cops working high profile cases, because individuals out patrolling will be so effective that there would be an abundance of time for cops to focus on needed work.
In a dystopian society cops would use this algorithm to predict crimes, leading to arrest for crimes that are just suspected to happen coming from the movie Minority Report, where in the movie there was Agatha who was able to predict when murders were going to happen. When she would get a “vision” of a murder, a task force would be sent out to stop the murder from ever even happening where people are arrested for even the possibility of committing a crime. This could lead to a great deal of falsely accused individuals or people who might not have actually gone through with it. It would also raise the question if it is against the Amendments to arrest people without the correct evident. Falsely accusing individuals would also lead to a great amount of people being incarcerated possibly breaking our economy or leading to an influx of children being raised without one parents or both parents.
It is also important to note that the dataset being used, is very skewed and shows areas that are heavily populated with African Americans being the areas where the most crimes occur. This can cause an increase of paroles in African American areas and an increase in petty crimes found in those areas, leading to more African Americans being charged, having to pay crimes, unable to pay crimes and therefore having to spend time in jail. Leading to even higher rates of African Americans in jail, which could lead to societal problems, where people question if this is the new form of enslavement.
Going off of the idea that certain areas will be targeted, especially minority communities, in a dystopian future there is also the idea that this will be used not just in geographical make-up, but go further into DNA make-up and tie in how once genetic make-up will lead to them being a criminal. Therefore leading to people of color yet again being the back bone of racial profiling. This will be because the police officers will already be looking for crimes in their neighborhoods based off of past skewed data, they there will be a correlation between the people being imprisoned having the correlation of genetic makeup (minority decent), leading to all of the minority population being imprisoned. Which could lead to a break in the foundation of the government in America.
History behind policing
Before noting the ethical concerns it is important to discuss how the system got to the place that is today. The genesis of the modern police organization specifically in the South is the “Slave Patrol”. The first formal slave patrol was created in the Carolina colonies in 1704. Slave patrols had three primary functions: (1) to chase down, apprehend, and return slaves to their owners; (2) to provide a form of organized terror to deter slave revolts; and, (3) to maintain a form of discipline for slave-workers who were subject to summary justice, outside of the law, if they violated any plantation rules.
Fast-forwarding to the end of historical slavery, there was an even greater desire to patrol African Americans, due to a fear of losing authority and power. Some of these slave patrols transformed into police departments attempting to enforce Jim Crow Law’s. Whereas others went a different path, with an attempt to terrorize Blacks, leading to groups such as the Ku Klux Klan. Basically, these tactics were an act to terrorize and patrol African Americans, which ultimately preserving the interests of whites. Leading to this opposition that there are “bad apples” in the police force, which lead to cases like Trayvon Martin and Michael Brown. However, when looking at the history, police forces are upholding their assignment to terrorize African Americans, regulate Black activity and keep authority where it has always been.
Ethical Concerns
Some of the ethical concerns that we face in this project steam from the foundation of the society we live in being inherently racist. Throughout this project we were tasked in reading The Color of Law by Richard Rothstein, which talks about how we formed large populations of African Americans in certain areas due to white only communities and black only communities. This then leads to a practice where African Americans are forced into certain areas as a form of corralling and limiting people of color. This leads us to our first concern, with the socio-economics involved in policing, we are disturbed with the possibility with equipping police officers who already, some could say unknowingly, target minorities we will be increasing the terror and violence that African Americans face.
Going along with the question if police should even be equipped with such knowledge challenges people being able to have freedom and security. Having police paroling areas because there is the possibility of a crime can lead to some backlash. Especially if a cop tries to say “The black box algorithm told me a murder was going to happen here, I didn’t mean to shoot him I thought he was a killer”. This is something we already see today, where police perceive someone as a threat for just walking the street. Giving them a reason to be scared because they know a crime might be occurring is something that could lead to further unnecessary police violence.
As stated previously, all data, especially historic data, is skewed. This means that as of right now the data that can be seen will make areas that are heavily populated by Black people are targeted by police, meaning there are going to be the individuals stopped for petty crimes, or being caught doing something that white people are doing more frequently, but not being caught because of the lack of police officers being dispatched to those communities. Having skewed data that shows the targeting of African Americans, could be viewed as racial profiling when used along with our algorithm.
Conclusion
In this project, our team explored how AI could change the world of policing, we found a dataset of crimes in the Chicago area from 2001 to 2017, used that dataset to learn where crimes are going to happen next and (dispatch the appropriate amount of police officers depending on the type of crime that is going to occur). We implemented a neural net that imports keras to run, through our implementation and analysis of our results were able to see that AI models are not perfect and are limited by the model setup and the data inputted to them. One future consideration is to send out a survey to get unreported crimes, so that we are able to get a more accurate depiction of the crimes going on, not just the ones that have police involved. If it is even possible, we would also like to make the data we used less bias. This will ensure that there is less racial profiling and targeting of minorities.
When it comes to policing and AI, AI is more likely to be used as a suggestion since there are a plentiful of ethical implications that need to be looked upon before using the program. As well, without a higher accuracy and unbiased data information outputted is unreliable.
See our program and data here!
