In 2018, AI played a crucial role in solving one of the most notable cold cases of the century. Decades after committing the crimes, the Golden State Killer, Joseph James DeAngelo Jr., pleaded guilty to 26 counts of murder, being left with no choice after Investigators used AI to enhance their DNA matching platform.

The case above was solved using the help of AI, merely a few years after the tech became available, at a time when the general public still thought of it as a fantasy for the distant future.

Many years later, the tech has exponentially developed, evolving from being merely a tool assisting the forensic team to handling cases that no human can.

Tapes from the past that no audio engineer could clean, photographs that are to blurred to tell the face of a suspect or patterns that extends over so long to make no sense to a human; all are now made relevant using voice recognition, pattern recognition and forensic AI, pillars of the technology in which we are going to focus in depth.

How The ‘Impossible’ Was Solved & Prevented

The Golden State Killer case is not the only occasion where Machine Learning has helped police and Forensic teams solve what was long considered a buried mystery. But how did it get to the point of digging back decades in the past and pinpointing one guilty individual among millions?

According to available data and popular research on the subject, law enforcement has been utilizing reasoning algorithms that mimic human experts since the 1980s.

While not yet close to the capabilities of AI today, the technology was envisioned at the time only as a computer system. Keep in mind that 40 years ago, the word ‘computing’ held the same futuristic notion as Artificial Intelligence does today.

The late 1990s saw the rise of ‘Neural Networks,’ which used the same principles as machine learning today to identify complex patterns and deliver results in minutes that would take weeks for an entire team of detectives. It was not long before other technologies were blended in, and cases of crimes detected and solved through automatic license plate readers, fingerprint or facial recognition systems became the norm.

Following the early 2000s, technology in criminal detention became so popular that it merged with pop culture, with movies and cult TV series like CSI showcasing, and at times overhyping, the capabilities of algorithms.

Today, reality has far outstretched the imagination of those TV show writers, and it’s easy to see why when you consider AI that it’s correcting the ‘mistakes’ or slips of human detectives from an era past.

The Tools of Today – Voice, Image & Pattern Recognition

It’s fair to say that by the time of writing in 2025, law enforcement is keeping up with technology, and not vice versa. Yet, unlike many industries where AI is discussed as the great disruptor for both good and bad, machine learning is only enhancing the tools already at our disposal.
Voice recognition based on tone and frequencies is not new; yet, AI has pushed the boundary to the point of cutting through even the most sophisticated attempts by criminals to alter their voices in phone calls by matching complex speech patterns. The typical kidnapper phone call is suddenly not an option anymore, nor was it ever, if the system can analyze audio files from decades back and point to the suspect. The systems are already in place, with examples such as Nuance Forensics leading the industry, having deployed over 35 million voiceprints to its customers.

Facial recognition follows the same innovation path, with the latest use cases used in thousands of abduction-related cold cases. John Cole, one of the leading investigators in these cases, told Forbes that “No single effort like this has resulted in that amount of identifications in such a short period of time,” putting in perspective what can be achieved.

Partners in Law – A Potential Future

Will they replace humans in a Minority Report dystopian scenario? Most likely, no, and surely not in the foreseeable future. However, just as speed cameras made it redundant for police officers patrolling highways with radar, it’s unwise to think that a similar scenario won’t befall forensic teams or law enforcement overall. 

Guided by the example of successful AI applications in mimicking human behavior with companions offered by companies such as Candy AI, models could be trained to act as assistants to families of victims, or as long-term investigators of all suspects in the case. Over time, the ever-present companions would collect new data and evidence, signaling to law enforcement whether there was a trail worth following in a long-lost case. That would be a step up from what they were initially developed for, mostly for AI sexting, but the tech doesn’t ask about its foundation, it just spirals.

Not only on the victim and suspect side, but police officers could have the necessary assistance to make quick, evidence-based decisions. Just recently, Motorola announced its new tool, appropriately named Assist and dubbed as an ‘AI Sidekick’ that upgrades standard body cameras to tools fully aware of the environment, continually scanning for crime and relying on historical data.

Offices would not need to radio in for potential license checks, as the system automatically scans every piece of evidence and delivers a complete assessment of the situation. Furthermore, that assistant would automatically detect criminal patterns, hidden weapons, and keywords such as ‘shots fired’ to alert nearby patrols. Now imagine this concept taken to the limit with a full-on companion partner for a police officer and investigator that not only studies the environment but also their human partners’ behavior, understanding their logic and emotions, and knowing exactly where and how to react to offer support.

Again, the legal and ethical implication would act a filter to dictate who is the law enforcers, the human or the AI – but past experiences taught the public that at times a voice of pure logic or reason is needed even in real time to avoid tragic scenarios, at times inflicted by representatives of the law itself, such the scenario when Amazon withdrew its facial recognition technology from police use in support of the Black Lives Matter initiative.

No Case Is Too Cold Anymore

The National Artificial Intelligence Research and Development Strategic Plan, published in May 2016 and updated in 2023, among other points, included the following on the use of AI in Security and Law Enforcement, all of which are needed for the technology to solve cold cases.

• Detect patterns and anomalous behavior,
• Analyze case law history, 
• Assist with the discovery process,
• Summaries evidence.

Pattern recognition is truly the gemstone of Forensic AI, enabling the technology to transcend time and delve deep into the past. As long as the evidence is reliable, which is partly done by the AI, the system can connect those dots in ways that the lifetime of a human detective wouldn’t suffice. Statistical models, neural networks, and machine learning algorithms form the basis of modern pattern recognition systems that can collect and analyze all types of data in a criminal case, including speech, voice, DNA, images, physical evidence, context, and motives, to drive an accessible solution. What is impressive is that the extent of that data is not limited to one department and a few people over a period, but it can span decades of previous case history and evidence.

A trial by Avon and Somerset Police, using an Australian-developed AI tool called Söze, tested the digital investigation platform on 27 complex cases. The results showed it was able to review all the evidential material in 30 hours. Estimation shows that it would have taken up to 81 years for a human to review the same material manually! It’s hard to imagine that anyone will return to the old ways of doing things after such crushing results, especially when major companies like Palantir offer adaptable solutions like Gotham at a massive scale.

Challenges and the Ethical Use of AI

AI requires a massive amount of data, coming from every source, to make an accurate decision. In the context of law enforcement and AI forensics, this data needs to be overwhelmingly vast as the stakes of making the wrong decision are high.

Porcha Woodruff, an eight-month pregnant woman in Detroit, was wrongfully arrested on charges of carjacking and robbery. The facial recognition system had identified her as guilty, and after a lengthy interrogation, she was finally released, raising the question of what it takes to ensure this doesn’t happen again. What makes the situation even more problematic is that she is not the only person to be wrongly accused by automated systems; furthermore, is it correct to use these systems, even if we assume the accuracy is 100%, in scenarios where crime and political interest overlap?

Following the latest news of protesters being arrested worldwide, wouldn’t the use of a system that always knows your location and actions turn out to be problematic? Privacy is not even the primary concern, as the extent to which this tech can be damaging in the wrong hands far surpasses the implications of what the companies that develop these systems do with the data – another worst-case scenario for which companies like Clearview have already been fined and investigated.

Clearview CEO simply justified by arguing that the system that matched with 99.6% accuracy a person’s photo with all the images of them online was to be ‘trusted’ as the only authorized use was for official investigations.

It’s evident that the challenges and ethical implications are interwoven, opening up the case for what happens when the culprit to be fought is AI itself.

Using AI To Fight AI

The best defense is at times an attack, even if the tool used against it is the weapon itself. In a world where anyone can train their AI models and in which companies creating those AI can’t or do not do enough to put restraints on users, wouldn’t a more advanced tech used by law enforcement be the only solution to preventing AI crimes?

Going a step further from the previous section, the government might not only need better technology to catch individual criminals or organizations, but also full access and partial control over the companies developing the technology, so that it’s always one step ahead of any potential danger.

The problem of how to get more advanced tech so that you can keep in check that the people inventing the tech can easily blur into a chicken and egg scenario, the solution of which will only come if Laws evolved side by side with AI, as catching up is already proving hard.

Lesson From the Past

The two things that the Algorithm have that humans naturally run out of are time and memory, both tied together by the many ways we transcend them through writing, cataloguing, creating, and, in this context, preserving case law history.

Where AI kicks in is the next step, beyond preserving, it’s making the wrong for the past right, and also preparing us for preventing them in the future. On one side, there’s success to be celebrated and justice served; on the other hand, it’s up to the law to keep itself in check on where the line between right and wrong in terms of AI blurs.