# From data to decision: The promise and peril of AI in policing > In our previous blogs (here and here), we have reported on how AI has become an integral part of the military. But the military is not the only government… - URL: https://datafund.io/blog/from-data-to-decision-the-promise-and-peril-of-ai-in-policing/ - Published: 2024-10-03 - Author: Datafund - Tags: ai, privacy, ai-and-policing, datafund, latest ![Source: Datafund, Dall-E](/assets/blog/from-data-to-decision-the-promise-and-peril-of-ai-in-policing/01.png) *Source: Datafund, Dall-E* ## A day to do 80 years of work In our previous blogs ([here](https://blog.datafund.net/bringing-ai-to-the-battlefield-brief-overview-and-trends-part-1-5b4f966fdf06) and [here](https://blog.datafund.net/bringing-ai-to-the-battlefield-what-could-go-wrong-part-2-32f050824835)), we have reported on how AI has become an integral part of the military. But the military is not the only government subsystem to benefit from this technological breakthrough. An interesting story recently came from the UK, where the [Avon and Somerset Police are trialling an AI system](https://news.sky.com/story/ai-tool-that-can-do-81-years-of-detective-work-in-30-hours-trialled-by-police-13220891) to help them solve decades old cold cases. And the benefits are truly amazing. The system, called Soze, was able to sift through the evidence in 27 complex cases in just 30 hours. A gargantuan task that would have over 80 years for a human to do. So today we’ll look at how police forces are leveraging the tech in their work and where are the potentially really dangerous pitfalls that could usher in a grim Minority Report-style future. Let’s begin. ## Reheating the cold cases ![Source: Datafund, Dall-E](/assets/blog/from-data-to-decision-the-promise-and-peril-of-ai-in-policing/02.png) *Source: Datafund, Dall-E* Using AI systems to help police do their work is relatively novel and most of it is in experimental stages, so field and research data is not that abundant. That doesn’t mean its potential uses cannot already be discerned on the horizon. One area of police work that will likely be impacted by AI the most is big data analysis and reopening old cases. Just like armies, police forces around the globe are gathering increasing amounts of data — from smart cameras, facial recognition, biometrics, forensics… that need to be processed for relevant information and gathering insights. Not only that, they usually have at least some backlog of cold cases that have been sitting there for ages that nobody ever got around to look at again. The reason for this is that each cold case usually contains several hundreds of pages of evidence material. In the old days, investigators would have to manually review them all and look for connections. It’s just too much for people to process. But for an AI, it’s an area in which it excels, as the case from Somerset and Avon police shows. And they are just one example. [The police in the Netherlands](https://www.bestpractice.ai/ai-case-study-best-practice/netherlands_police_prioritises_which_cold_cases_merit_further_investigation_using_dna_new_testing_techniques_through_ai_analysis_) are implementing AI in their workflow in a similar fashion to help them gain new information from old data. They are *“digitising their cold case forensic data to enable them to use machine learning to prioritise which ones may be worth re-investigating”*. The scope of the project is something that human investigators would no doubt find daunting. Over 1,500 reports and 30 million pages of cold case material are being transformed into an AI-friendly form. Once the digitalisation is complete, a machine learning algorithm will begin combing through the records and decide which cases have the most promising evidence, reducing case processing time from weeks to a single day. The US Department of Justice is also exploring AI-based text analytics tools and their potential value to cold case review. [Initial reports](https://forensiccoe.org/private/63656b211b684) show that these tools can, among other things, reduce time searching for documents, come up with additional testing opportunities and strengthen investigative objectivity. ## Preventing crimes before they happen… Cold cases are one aspect in which AI is transforming police work at the moment, although its potential impact is much larger (in its [recent report](https://www.europol.europa.eu/cms/sites/default/files/documents/AI-and-policing.pdf), Interpol goes into more detail about current and possible uses of AI). In the future it could or is currently influencing: - Street patrolling - Criminal investigations (e.g. finding patterns, correlating different types of data) - Fraud prevention - Surveillance (e.g tapping into mobile networks, facial recognition) - Managerial functions - Bureaucracy - Officers’ well-being (e.g. reducing their exposure to harmful content) In 2018 the [US National Institute of Justice](https://nij.ojp.gov/topics/articles/using-artificial-intelligence-address-criminal-justice-needs) already supported four large areas of AI research: - Public safety video and image analysis - DNA analysis (this is an area of special interest in cold case reviews) - Gunshot detection (this helps identify weapons by analysing the acoustic properties of gunfire) - Crime forecasting The latter, also called predictive policing, is currently the core focus point of AI exploration by police departments. According to the European Crime Prevention Network’s recommendation paper, the Netherlands, Germany, Austria, France, Estonia and Romania are already making use of predictive policing. Other EU Member States, such as Luxembourg, Portugal and Spain are still assessing the possibility of implementing it. On the global level, China, USA, India, Canada, Brazil, South Korea, Japan and Australia are also using AI for crime prediction. **Predictive policing** refers to the *“application of data analytics, artificial intelligence, and machine learning techniques to anticipate potential criminal activities before they occur. By leveraging historical crime data, demographic information, and real-time inputs, predictive policing aims to identify crime hotspots, allocate police resources efficiently, and ultimately prevent crime”* ([Olaoye, 2024](https://www.researchgate.net/publication/383565286_Predictive_Policing_and_Crime_Prevention)). As [Interpol](https://www.europol.europa.eu/cms/sites/default/files/documents/AI-and-policing.pdf) describes it, predictive policing is primarily implemented through rule-based machine learning models. These scour historical data for patterns so they can produce indicators and liken them with the likelihood of a crime occurring, and then generate risk scores as predictive outputs. This type of policing can focus on an area or individual, meaning they can be used, for example, to predict crime at events or identify individuals that are more likely to commit them. The main advantage of predictive policing and its underlying AI is that it can continuously sift through vast amounts of historical and real-time data and make a prediction. In this way, predictive policing can detect crime hotspots, improve resource allocation, underpin decision making with more data and improve with more and more data. ## …could be a risky endeavour Predicting crimes before they happen is inherently risky. As [Richard A. Berk points out in his article](https://www.annualreviews.org/docserver/fulltext/criminol/4/1/annurev-criminol-051520-012342.pdf?expires=1727860452&id=id&accname=guest&checksum=1F0E77E6517CB2BE05C6D4C36537E460), the issue with predictive policing is whether it *”leads to improvements in accuracy beyond current procedures”*. According to him, *“there have been some successes, but so far we have no general conclusions one way or the other”*. Some studies and examples suggest a higher success rate, but this is far from a definitive conclusion across the board. The problems of AI-supported police work were also one of the main topics at this year’s[ Computers, Privacy and Data Protection (CPDP.ai) conference](https://thesecuritydistillery.org/all-articles/legal-ethical-and-governance-challenges-of-ai-for-law-enforcement-voices-from-cpdpai-2024). Experts pointed out that the main problem is the difficulty of **challenging AI-processed evidence** and algorithmic decisions because the basis for the decision can be extremely vague. In addition, the decisions of AI systems can be influenced by **algorithmic bias** and lead to excessive policing of disadvantaged communities. There’s **automation bias** as well, which can influence human officers to rely too heavily on AI, leading to errors in decision-making. The **legal framework** surrounding AI in law enforcement is also underdeveloped. In the case of the EU, it’s the lack of clear regulation in the AI Act regarding the classification of high-risk AI systems. This ambiguity raises concerns about accountability, particularly when AI tools are used to process evidence. When it comes to the use of technology, there is an interesting divide between police leadership and street units. While supervisors are happy to adopt new technology, officers on the street tend to be more reluctant and *“find ways not to use the technology and not to register data if they consider it impractical”*. In his article, Berk points to a related problem, citing a study from 2001. The latter found that *“the road from forecasts to implementation can be rocky. A lot depends on how the information is transmitted to patrol officers and how it conforms to existing police officer’s priorities. Accurate forecasts alone are no guarantee of improved policing performance”*. ## Conclusion It’s really interesting to see how widespread and diverse the use of AI is becoming. At the same time it’s raising red flags because it is being integrated into society’s critical functions. When it comes to policing, AI’s decision-making processes and people responsible for it must always be monitored closely since personal freedoms and rights are on the line and the stakes are high. If you liked this content, keep following us for more on: ## [X](https://x.com/DataFundProject) ## [Linkedin](https://www.linkedin.com/company/datafund) ## or get in touch via: ## info@datafund.io --- [From data to decision: The promise and peril of AI in policing](https://medium.com/datafund/from-data-to-decision-the-promise-and-peril-of-ai-in-policing-1058ab6dc43c) was originally published in [Datafund](https://medium.com/datafund) on Medium, where people are continuing the conversation by highlighting and responding to this story. --- *First published on [Medium](https://medium.com/datafund/from-data-to-decision-the-promise-and-peril-of-ai-in-policing-1058ab6dc43c) on 2024-10-03.*