Can Machines Listen to You? Using Artificial Intelligence to Monitor Inmate Phone Calls
Rory E. McBryde
American Jails Magazine has addressed the possibility of how AI can impact U.S. correctional facilities and law enforcement in its May/June 2023 issue with the article “Revolutionizing Jail Operations with Artificial Intelligence.” The author, Jonathan Ede, CJM, MSCJ, questioned whether there were AI partners that could help address manpower issues. This article provides an answer and practical solution using one form of AI that can quickly and efficiently replace (or at least greatly enhance) manpower with technology.
Monitoring inmate phone calls is an important aspect of security for a correctional facility and is often used by law enforcement to obtain evidence and information critical to investigations across every type of crime in the United States.
It’s also a time consuming and expensive process to pay an officer to sit and manually review calls. What if technology could do it for you? Now, with the expanding capabilities of Artificial Intelligence (AI), and specifically the subsets of AI called Natural Language Processing (NLP) and Large Language Models (LLM), the technology now exists for law enforcement to review large volumes of inmate phone calls in a fraction of the time and cost of manual listening.
Searching For Keywords
I’m a career prosecutor and I never used to have an interest in technology. But I also hate inefficiency and I enjoy solving intractable problems. A few years ago, I became interested in NLP because of a co-worker’s case. The defendant had made over seven thousand phone calls in 9 months, an insurmountable number of calls to review. My frustrated co-worker made an off-hand comment to me: “Why can’t we just hit “Control+F” and search for keywords in their calls?”
In my career, I’ve spent numerous hours listening to calls (often while multitasking with other work). It’s necessary, because I’ve found crucial evidence and successfully convicted defendants in serious cases with jail call evidence. But it’s also an agonizing process to listen to hours of calls just to find a few seconds of relevant evidence.
So, the question intrigued me. Why couldn’t we just search for keywords in audio? That day I spent my lunch hour researching the issue and discovered that we actually could search through audio with NLP.
You have already used this type of AI if you’ve ever spoken to Apple’s “Siri” or Amazon’s “Alexa.” They can hear, understand, and respond to spoken commands. NLP is the field of artificial intelligence that focuses on teaching computers how to understand human language. Essentially, NLP allows machines to recognize and interpret spoken or written words just like humans can.
When I found out about the possibility of computers searching audio for me, I started researching companies that provided this tool for law enforcement. I was disappointed. There were almost no vendors I could find in 2019 for this type of tool. There was one company that provided the service to review every call in the jail, but it was not quickly deployable, and I couldn’t use it personally.
I decided to try and make my own simple tool that law enforcement could utilize to search jail calls for evidence. I soon found a programmer to assist me with building the tool and our first available version was a simple “Transcribe and Search” software that could transcribe numerous audio recordings and
you could search for keywords across all your transcripts. With continuing development and the recent growth of LLMs like ChatGPT, which can interact with and generate conversational text, there’s now an additional level of analysis available for law enforcement.
Artificial Intelligence Evidence Analysis with LLMs
It’s important to note that AI is not a magic wand. You can’t pick it up and have it magically solve all your problems in a particular area. AI is not omniscient nor without failures, and you can’t use it to replace human beings in most settings. But it can be an amazing tool to speed up repetitive and impractical tasks like reviewing thousands of jail calls quickly.
Most people know about ChatGPT and generative AI chatbots from the news recently. Generally, an LLM is an algorithm that uses deep learning techniques and massively large data sets to understand, summarize, generate, and predict new content. You don’t need to know the exact details of how they work, but one fantastic feature of LLMs is they can take sets of data (or in this case a set of transcriptions from jail calls) and summarize and analyze them for you, flagging calls with potential evidence.
A combination of NLP transcriptions with LLM analysis can transcribe, summarize, and analyze hours’ worth of jail calls in a matter of minutes. There are still limitations based upon audio quality, obscure slang, or poor articulation of the speaker. But this technology is still very young and is improving constantly and use among law enforcement in the US is continuing to expand.
The speed and efficiency of having AI review calls, as well as having transcripts of calls already completed, can greatly assist with the speed of jail call review and report writing. Officers can copy and paste relevant portions from transcripts (after accuracy is confirmed) rather than listening and notating each word themselves.
Beyond NLP: AI and Machine Learning in Law Enforcement
NLP is just the tip of the iceberg when it comes to harnessing the potential of AI and Machine Learning in law enforcement. Several other subsets of these technologies can be integrated to augment various aspects of policing, including:
• Audio Detection: AI-powered audio detection systems can identify specific sounds, such as gunshots, alerting law enforcement to potential incidents in real-time.
• Video Review: Machine Learning algorithms can analyze vast amounts of video footage to identify faces, objects, and activities. This not only aids in criminal identification but also enhances situational awareness.
• Data Analysis: Predictive analytics and data mining techniques can be employed to identify patterns and trends in crime data. This assists law enforcement in allocating resources strategically and proactively addressing emerging threats.
• Predictive Policing: Machine Learning models can predict potential crime hotspots based on historical data, enabling law enforcement to deploy resources preventively.
There are an increasing number of companies providing these services to law enforcement across the nation. I predict all these various uses of AI will be standard in most jurisdictions in the near future when accuracy and quality improve, and costs go down.
The Future of AI and Jail Calls
The particular subset of AI that I work in has become more available from additional vendors since I started working in the field of AI for law enforcement. I believe this tech has the potential to completely transform how corrections and law enforcement monitor jail phone calls in the future and will be widely implemented soon. Using NLP and LLMs to analyze large sets of jail calls faster than humanly possible and at a fraction of the cost of manual review will greatly assist law enforcement investigations in the future. There are a multitude of possibilities that the coming AI revolution will bring to American law enforcement, and if recognized and implemented correctly, we can utilize it to effectively free up human resources for more important tasks that only humans can do.
_________________________________
Rory E. McBryde is an Assistant Prosecuting Attorney for the Berrien County Prosecuting Attorney’s Office in Saint Joseph, Michigan, assigned to adult felony prosecutions. He has 10 years of prosecutorial experience in California and Michigan. He is also the founder of Wiretaptech.com, a cloud-based software that uses AI to transcribe and analyze jail calls. For more information, he can be reached at (907) 406-0337 or via email at wiretap.llc@gmail.com