I recently had an informative conversation with members of PSA’s emerging technology committee around artificial intelligence (AI) and analytics and the role they (will) play in security systems now and in the future. Kevin Henderson is chief operating officer at Lone Star Communications of Arkansas and Josh Akre is the performance engineering manager at Northland Controls of California. Both are in tune with the needs of their end users and well-versed in the technologies needed to meet them.

The conversation covered how quickly the technology is evolving, many practical use cases both are encountering, as well as a deep explanation of how the technology works… for a layperson like myself. The true takeaway was how rapidly security technology will evolve and that, in some cases, the applications are driving the demand. 

“There will be a transitory period for a few years, then it will be stigmatizing not to have this technology, Henderson said”

As they both explained, end users aren’t asking, “Can you use AI to do this,” but rather, “Can this be done?” Consumers now have the smartest technology of all directly in their pockets: a smartphone. This increases expectations of what enterprise technology can accomplish. Therefore, AI is necessary to continue to meet the needs of the market. 

Likewise, I learned the difference between AI and analytics and that there is confusion in the marketplace between the two. Akre and Henderson explained that with analytics, a determined dataset is used that cannot evolve over time. While this data enables a camera to identify a car is red or what a weapon is, it isn’t constantly learning and evolving the way an AI-enabled device is. An AI-enabled device can learn over time things like the difference between a broom and a gun, and, ultimately, deliver a higher interval of confidence to its operator, cutting down on the need for human intervention. 

Lone Star Communications is working to develop solutions for hospitals that can anticipate a patient’s moves before they are in motion. In one use case, this allows hospital staff to stop falls before they happen. Another use case would be weapons detection in schools before an event even begins. These data sets take time to create, though. The technology must be in use for a while to observe behaviors and, ultimately, make better decisions. 

“90 percent of the time people pull their weapon before they go into a building,” Henderson said. “We are trying to teach it to detect that and spur a lock down. As an integrator, if you can own this piece and you can own the data, you aren’t replaceable. The market is demanding AI-enabled technology. This acceleration is very similar to the analog to IP transition as many of these new devices will be able to leverage existing infrastructure.” 

Akre explained the COVID-19 pandemic rapidly accelerated AI capabilities. “For every single camera it requires a lot of collaboration with the end user to get it configured. It’s a lot to scale with hundreds of cameras,” he said. “It’s exciting to see how companies will respond to this over the next few years.” 

But where does the object recognition data come from?  Manufacturers are selling libraries of what their devices can identify. It’s the manufacturers secret sauce. They continue to grow that database as the real-life use gets sent to the cloud and continues to refine the product capabilities over time. 

It was challenging in the past to get the library of data because the devices must be in use to capture it. Without wide adoption of the products, this doesn’t happen. Pre-pandemic, companies were trying to go to market with libraries that weren’t as refined as they are now. With the data improvements, devices are moving on to next-level capabilities like false alarm detections.  

“There are layers: AI, machine learning and deep learning,” Henderson explained. “In deep learning you are trying to get it to function like a human brain. We want it to almost be able to reason. We’ve created more data in the past 10 years alone than has ever existed ever. You must teach the system what it’s supposed to be learning and how to learn it and you have to keep inputting the data.” 

We concluded the conversation around end user demand for AI-enabled devices. They explained the adoption has moved slower due in part to the man hours needed to configure and deploy devices. Likewise, in some instances, they’ve installed new systems for clients who didn’t even realize the AI-capabilities. It was a nice surprise after the fact, rather than the main feature they acquired the system for. 

The future is exciting to think about with AI technology for security systems integrators, though. From incorporating security systems into overall smart building applications to detecting, identifying and verifying a customer at a bank and having their account information pulled up before they get to the teller, the possibilities are endless.