Today’s featured expert, Innodata’s own Justin Haynes, has extensive experience in the Federal Government and the private technology sector. Justin spent 15 years in U.S. Marine Corps and Federal Government working in special operations and intelligence, where he trained in technology and data science and cross-trained with the NSA, CIA, and FBI. Following this, he worked for 10 years as a local law enforcement officer. More recently, Justin held a variety of roles in tech startups, focusing on data analytics and data fusion projects, before joining Innodata as a senior business development executive. One of his key areas of interest is using technology to improve operations and outcomes in the Federal Government and law enforcement. Innodata recently sat down with Justin to get his perspective on the current role of technology in law enforcement and how he would like to see it evolve. 

Innodata: Thanks for sitting down with us. Based on your background, you have a unique inside-and-out view of technology in the U.S. federal government. How would you describe the government’s current relationship with technology, especially AI and ML? 

 JH: Since 9/11, the Department of Defense and myriad federal agencies have received a ton of funding, leaving a void in state and local governments. In fact, other than major cities that are critical to homeland security, state and local governments were left way behind in funding and technology. Surprisingly, even cities as large as Denver and St. Louis are in a different era…they still think in terms of fax machines rather than AI/ML. 

I’ll give you an example: a district attorney’s investigator once told me that he needed a fax. Since the fax machine was down, I offered to scan the document and email it to him. He insisted that a scanned, emailed copy wouldn’t work, because they needed the original document, so it had to be a fax. He did not understand the tech at all. And this was only about three years ago. 

New purchases of tech go unused or underutilized due to unfamiliarity. Funds get wasted on purchasing new tech that is not needed. For example, one department purchased a software platform for document management. The platform allowed for collaboration, shared drives, etc. The department paid more than $150K/year in licensing fees for this, when they could have had the same capabilities for free with something like Google Drive. They are wildly behind the times, through no fault of their own. And this was in Denver, CO, not a small, rural area! 

In many departments, basic technology is almost seen as witchcraft, and AI can come off as so daunting that it doesn’t even come up as a possible solution. So, we have a long way to go in terms of cultivating an understanding of current technology. 

Innodata: In your opinion, where could AI and ML add the most value in law enforcement? What are some use cases that could significantly reduce officer workload, increase effectiveness, and reduce costs? 

JH: I see enormous potential in data extraction/redaction and document/image intelligence. In a document intelligence platform, for example, the redaction capability for structured data would be huge for law enforcement…and this is just one micro-use case. One time a lawyer from a defense attorney’s office was looking to hire an expensive forensic company to scan some child pornography videos so that his team wouldn’t have to. He also needed to organize some unstructured data like invoices. This kind of task could be done easily, on-site, with a doc intelligence platform.  

Another great AI application is image annotation for crime investigation and prevention. The number one crime in Colorado is motor vehicle theft. It’s a nexus crime, where cars are stolen in order commit other crimes. Currently, automatic license plate readers scan plates as cars drive by and identify stolen vehicles. They provide vehicle status info, but not much else. An ML use case here would be getting an image of the driver, using facial recognition, and comparing this to a list of known/wanted offenders. Right now, the process is much more manual and random, where you have to scroll through lists or search for one specific vehicle or plate at a time. 

Another use case for image annotators is predictive analysis using indications and warnings. What does it look like when someone’s about to steal a car? We can capture the image by going backwards in time. What does it look like before a ‘smash and grab’ at a marijuana dispensary or jewelry store, where people smash windows, grab stuff, and take off? The precursors look almost the same every single time. There is a whole methodology around indications, warnings, and precursor analysis that could be done really effectively with AI. 

Finally, at the forefront of every law enforcement agency these days is PR and reputation management. PR platforms would be great for monitoring social media and a department’s reputation. What are people saying about my department? Are there unreported officer offenses that are blowing up on social media? A friend of mine was involved in an encounter where he had to use deadly force. It was justified, and he was cleared of all charges. But because of one careless post on social media, a PR nightmare unfolded.  A PR platform would have identified this immediately and the dept could have removed it, apologized, and taken immediate action. Reputation management needs constant, persistent monitoring of social media. And we need to stay aware of publicly available information because it stays out there even after users think it has been deleted. Many police departments are not careful about this. When combined with a human element for sentiment analysis, this would be an excellent job for AI. 

Innodata: What do you see as the biggest challenges to AI adoption in law enforcement, and how could they be overcome? 

JH: Right now, I would say the three biggest hurdles are the lack of familiarity and understanding of AI, the real and perceived error rates, and the cost or perceived cost of implementing AI. 

For the lack of familiarity, I think a lot of internal education is needed to bring departments up to speed. This should ideally be done by people who speak the languages of both law enforcement and AI, who can translate and interpret between the two jargons. 

For error rates, both real and perceived, I believe that a human element in AI will be critical at this point. ML and AI are becoming more palatable in law enforcement because there are humans at the back end double-checking the AI functions. A human-in-the-loop arrangement makes these departments more likely to use AI. 

For cost control, one solution is for police departments or small municipalities to get together to share AI solutions and costs. An AI/ML system could be a shared asset or a joint regional asset. This is already being done successfully with SaaS, people, and vehicles. It helps keep costs down.  

Innodata: Do you have any final thoughts on AI in law enforcement as a whole? 

JH: AI and ML could address many of the issues we face in law enforcement today. With widespread internal education, thoughtful implementation, and human-in-the-loop frameworks, AI could vastly improve the efficiency and effectiveness of police departments. They can help predict, prevent, and reduce crime, all the while maintaining a positive relationship with the citizens they are sworn to protect.