All AI Is Not Created Equal

July 29, 2019

Stepping up your security surveillance system might be on your priority list, especially if you’re concerned about the increasing prevalence of active shooter incidents against places of business, worship, and education. If you’re considering implementing AI-enabled security, as many people are, you need to understand that there are big differences in types of artificial intelligence on the market. There are specific things about AI that make some extremely effective for security purposes, and others useless. Given the high risk involved if an AI system fails to work, it is especially important to understand what you’re buying before making the investment. Here’s what you need to know.

AI is only as good as what it’s built to do

Artificial Intelligence is a hot buzzword right now. Every technology company wants it, and coders all over the world are learning how to create it. It’s become an ecosystem in and of itself. Defendry could have completely outsourced AI development to a team of engineers oversees and had a system built for pennies on the dollar. Our mission, however, is to save lives, and so rather than gamble with a high risk of error from inexpensive development, we have assembled a team of expert engineers who inherently understand what that takes.

Our AI development is led by Dr. Sean Huver, Defendry’s Chief Science Officer. Huver is a research scientist and veteran of numerous United States Department of Defense DARPADefense Advanced Research Projects Agency (DARPA) is responsible for the development of emerging technologies for use by the U.S. military. projects. Upon discovering the gaping need for faster response to retail crimes, Huver created a company named Deep Science to develop the proprietary AI detection software that is now owned exclusively by Defendry. The system he built was the first human-in-the-loop AI active threat detection platform on the market.

Most artificial intelligence surveillance systems are built to simply recognize shapes and send alerts to authorities. This is problematic, as there are a number of scenarios in which the shape that’s recognized as a threat actually isn’t one, and other times the system misses an active threat because the AI doesn’t recognize the shape of the specific weapon in the frame. 

Defendy’s AI has been designed to learn all kinds of threat-related patterns, and will naturally learn what all manner of threats look like. For instance, it can tell the difference between a gun and a drill (many AI systems cannot), and it can distinguish very intricate details such as different models of guns, the different ways guns are held, etc. “Military-grade AI” means it was designed to learn these things specifically. The more our AI is used, the smarter it becomes, and all of our customers become safer.

AI has to be trained in something specific, in order to do something specific

Artificial Intelligence is, by nature, a learning system. That means it needs to be built to learn specific things, and it needs to be trained, just like a human brain. Defendry’s AI has been trained using surveillance videos of actual crimes, as well as proprietary staged crime scenarios. These videos teach our AI to understand and respond to multiple ways that an active shooter could attack. Further, our AI training was directed by renowned Police Officer Tony Sanborn, who trains and manages the Scottsdale, AZ K-9 unit attack dogs. 

Officer Sanborn has helped train our AI in thousands of potential scenarios and in varying environments where an active shooter threat might happen. Variables like type of weapon, angle of attack, and even time of day are taken into account. This means our AI system is trained to accurately recognize virtually every imaginable type of gun in just about every possible situation. 

When Defendry is installed in a location,  if a threat is detected, Defendry’s 24/7 monitoring initiates a series of auto-responses ranging from locking doors to sounding alarms to calling police (only after a human verifies that the threat is real, which prevents false alarms). Other AI might be useful for lessening the severity of attacks or identifying bad guys faster. Our AI is built to help prevent active shooter incidents from happening in the first place by locking the shooter out.

AI must be able to identify non-threats 

We mentioned above that some AI systems are designed to recognize just shapes, and not specific types of weapons (including models). This can be very problematic. Imagine if your security system were to see an employee walking into the building holding a drill or a stapler, inaccurately identified it as a gun, and automatically initiated an emergency response protocol? The result could be a very costly liability, not to mention traumatic for the person accused, and potentially even a PR nightmare for the business. 

Defendry’s security-trained AI, plus our human verification, drastically reduces the incidence of false positives and eliminates false alarms. The most important thing an AI-enabled, active shooter defense system must have is precision. That means recognizing every kind of possible threat, and confidently excluding false threats.

Artificial Intelligence is an amazing breakthrough, with endless applications. Enabling better security through AI is a no-brainer for anyone who’s tasked with protecting the people inside a business, place of worship, school, or other building. Know that before you make the investment, all AI is not created equal. When it comes to protecting lives, you need an active shooter defense system that was built just for that purpose.

To learn more about Defendry and how we can help you, talk to us.