By: Maya Lakhani
Edited by: Emilia MacDonald
Artificial Intelligence was created to ensure efficiency and public safety for all while reducing the effect of human biases. In reality, it can do exactly the opposite. AI is built on human-generated data, meaning that AI may pick up biases present in the given information. Therefore, AI can be inefficient in preventing prejudices in societal institutions like the criminal justice system. The struggle between leveraging Artificial Intelligence for its benefits and mitigating its biases remains an ongoing dilemma. A recent study suggests that AI and other smart technologies could potentially lower crime rates by 30 to 40 percent and improve emergency response times by 20 to 35 percent. However, at what cost? Is it eliminating crime primarily for majority groups while increasing it for minorities? This is where the struggle persists, and the risks must be taken into account. The justice system needs to find a way to ensure efficiency while remaining fair to all people. Without strict oversight, AI-driven policing can reinforce existing biases, misidentify threats, and disproportionately impact marginalized communities, ultimately undermining the very safety and fairness it aims to enhance.
Predictive policing now uses AI to anticipate crimes for greater efficiency. Predictive policing involves software that analyzes data and utilizes algorithms to predict criminal activity, aiming to optimize the allocation of law enforcement resources (NAACP, 2024). A key component of this system is Big Data and machine learning (ML) algorithms, which play a crucial role in modern law enforcement. While statistical analysis in policing has existed for decades, such as intelligence-led policing in the 1990s, Big Data now enables law enforcement agencies to process vast amounts of information to detect crime patterns, trends, and correlations (Storbeck, 2022). Machine learning enhances this process by identifying patterns in data and improving over time. Unlike simple algorithms, ML models can adapt based on experience. They operate in two ways: supervised learning, where they are trained with labeled data, and unsupervised learning, where they independently find patterns in raw data (Storbeck, 2022).
The goal of AI is to ensure that public safety is being enhanced. However, evidence finds that the growth of AI in this field is leading to racial bias that violates human rights and undermines trust in law enforcement (NAACP, 2024). This is caused by the fact that predictive policing decisions are based on the analysis of past crime reports and law enforcement, yet these decisions are formed from historical data that reflects systemic inequalities, thus perpetuating racist biases. Black communities, in particular, face disproportionate negative impacts due to excessive policing and discriminatory legal practices, which affect the data and reinforce existing injustices (NAACP, 2024). For example, French police forces heavily rely on AI in predictive policing, using a compilation of filed complaints, historical crime data, and geo-locations of crimes from the past seven to ten years (Storbeck, 2022). Future models are also shown to now incorporate meteorological data and national statistics to refine predictions, which can then illustrate the areas with a higher likelihood of crime (Storbeck, 2022).
As mentioned previously, a major concern with AI in policing is its potential for bias. Looking more closely, this bias can stem from two main sources, which are algorithmic bias and Big Data bias. Algorithmic bias arises from the unconscious or conscious prejudices of developers, who may unintentionally embed their views into the system (Storbeck, 2022). Meanwhile, Big Data bias occurs when historical crime data disproportionately reflects over-policed minority communities, reinforcing systemic discrimination (Storbeck, 2022). AI systems are also prone to false positives and positive feedback loops, where biased training data leads to repeated targeting of specific racial or socioeconomic groups (Storbeck, 2022). This bias is even further exceeded by the “black box” nature of AI, meaning that decision-making processes are opaque, making it difficult for policymakers and the public to understand or challenge AI-driven law enforcement decisions (Storbeck, 2022). Without a reasonable understanding of decisions, it makes it difficult to fully trust and rely on the system’s output.
In North American contexts, the US court system has been using an AI program called COMPAS. However, a recent study by ProPublica found that COMPAS not only failed to accurately predict criminal behaviour but also disproportionately labeled Black defendants as high-risk (Storbeck, 2022). According to their analysis, Black defendants were twice as likely as White defendants to be incorrectly classified as likely to reoffend (Storbeck, 2022). Additionally, Black individuals are five times more likely to be stopped without just cause, further inflating the data that AI systems use to assess criminal risk. These factors illustrate why platforms like COMPAS are flawed and thus remain a topic of immense controversy.
Incorporating Artificial Intelligence into law enforcement presents both opportunities and challenges. While AI has the potential to improve efficiency and resource allocation, its reliance on historical data and algorithmic processes risks reinforcing systemic biases and disproportionately impacting marginalized communities. It is essential to create a balance between innovation and accountability to ensure that AI enhances public safety for all, rather than deepening existing inequalities. The justice system must prioritize fairness, ensuring that technological advancements do not come at the cost of equity and human rights.
References:
Image Source: Okan Çalışkan – publicdomainpictures.net. (2022). Law Justice Court Free Photo. Needpix.com. https://www.needpix.com/photo/1477689/
Arrests by offense, age, and race | Office of Juvenile Justice and Delinquency Prevention. (n.d.). Ojjdp.ojp.gov. https://ojjdp.ojp.gov/statistical-briefing-book/crime/faqs/ucr_table_2?table_in=2
Deloitte. (n.d.). Surveillance and Predictive Policing Through AI | Deloitte. Www.deloitte.com. https://www.deloitte.com/an/en/Industries/government-public/perspectives/urban-future-with-a-purpose/surveillance-and-predictive-policing-through-ai.html
NAACP. (2024, February 15). Artificial Intelligence in Predictive Policing Issue Brief | NAACP. Naacp.org. https://naacp.org/resources/artificial-intelligence-predictive-policing-issue-brief
Storbeck, M. (2022). Artificial intelligence and predictive policing: risks and challenges Recommendation paper 2. European Crime Prevention Network. https://eucpn.org/sites/default/files/document/files/PP%20%282%29.pdf