This article originally appeared in the March 2021 issue of Security Business magazine. When sharing, don’t forget to mention @SecBusinessMag on Twitter and Security Business magazine on LinkedIn.
Back in 2017, when use and hype of the Artificial Intelligence buzzword really hit the security industry, we wrote: “The uphill climb for today’s video surveillance and other manufacturers, who are tasked with integrating and then marketing AI to security integrators and end-users, is how to dispel the pre-conceived perceptions of what Artificial Intelligence really is.”
Nearly four years later, AI has gained a more stable foothold in the security industry; however, the struggle to advance from the early adopter phase to the broad adoption phase is still taking shape.
This exclusive trends roundtable sponsored by Tyco American Dynamics takes a closer look at this struggle, as well as other aspects of AI technology adoption, with Johnson Controls/Tyco Security Products Manager Jammy DeSousa, as well as veteran integrator Eric Yunag, VP of Technology & Innovation for Convergint Technologies.
What indications have you seen that we are now past the early adopter phase when it comes to AI technology, and if so, when will we see markedly broader adoption?
Yunag: I think one of the clearest signals that we have reached a tipping point is the interest from major cloud and technology companies investing in “platform” strategies in the computer vision space – such as Amazon AWS with Panorama, Microsoft Azure with Custom Vision, and NVIDIA with Metropolis.
These companies clearly recognize the enormous impact these technologies will have on the consumption of their services and are evolving their product strategy and investments accordingly. The speed of development, diversity of hardware deployment options, and scalability of the solutions enabled by those platforms will pave the way for sustained market adoption.
Last year also offered another signal in the form of accelerated digital transformation initiatives across a number of vertical market segments that included legacy security, fire, and life safety systems. In many mission-critical, enterprise environments, these technologies had simply not been proven enough to be relied on, but that has begun to change.
DeSousa: More and more leading companies are investing in AI, such as IBM, Amazon, JCI, etc., because they understand the potential. AI is such a broad “tool” that can be used in almost every business case to make it more efficient, making more accurate decisions faster. The industry will only see broader adoption when more companies use AI and learn to communicate its benefits. The companies who can best communicate AI’s value in their products and solutions will gain quicker adoption.
How does AI differ from traditional intelligent video analytics?
Yunag: Primarily this comes down to the virtually limitless flexibility of AI and what it can be trained to recognize vs. the essentially static nature of traditional analytics. AI – or better stated, computer vision – can leverage ground truth and sample data to train machine learning models to recognize very specific scene elements or, more importantly, behavioral elements. Ultimately, the value of many machine learning models becomes less about the video itself and more about data and trends identified by data, which businesses can then leverage to more effectively optimize outcomes, manage behaviors, or reduce risks.
What criteria should integrators and end-users apply when evaluating AI software and technology?
Yunag: First, it is important to consider a software company’s philosophical approach to the ethical use of AI as it relates to computer vision. There is a continually evolving landscape of laws and regulations that need to be considered before scaling any significant deployment. It is also critical to understand very specifically what the application and use-case is for these technologies. Broader is not better. The more specific the use case, the higher the likelihood of success will be and more impactful the results.
There are several specific computer vision solutions that are effective because they were developed and trained to serve a particular industry or a specific environment. The results could be different if used in another environment where conditions are significantly different.
Lastly, I believe that the best solutions allow for some element of “human in the loop” – even the best AI models can be improved with periodic human validation and correction to allow for continuous improvement and optimization.
DeSousa: Accuracy, performance and user experience. Companies need to trust AI in their systems, so having a high accuracy and
performance rate is key. 80% accuracy is good, but 90-plus is better and more trustworthy. If an application is centered around real-time video monitoring (airport, bank, healthcare facility) has a BOLO (be on the lookout), knowing about that intruder, and knowing where he or she is or was last as quickly as possible could be a matter of life and death.
Explain the role that a VMS plays in AI deployments.
Yunag: It is likely that Video Management Systems will remain a core component of enterprise surveillance environments into the foreseeable future. Beyond the historic forensic requirements, there is a significant amount of data (information) stored in traditional VMS, which has the potential to provide additional value as a source of data for model training and enhanced, intelligent search.
A number of innovative computer vision companies are coming to market that take advantage of the in-place VMS infrastructure to connect to a single enterprise system, and aggregate historical data to provide real-time analytics and intelligence for investigative purposes and better situational intelligence.
DeSousa: Video management systems use a multitude of analytic features within the software on either the cameras or recorders. AI can take a camera that simply provides a viewpoint on the entrance of a building and turn it into a highly intelligent behavioral sensor. Facial recognition, tracking persons of interest, and automatically redacting faces for privacy is the tip of the iceberg of what AI could bring to a VMS. As technology advances, a VMS without AI will not only become obsolete, but ineffective compared to a VMS that deploys AI features.
What are the ideal target markets for this technology?
Yunag: These technologies are evolving at an incredible pace, and new use-cases are possible seemingly every day. As a result, the ideal market and customer applications remain very broad. Any environment where digital transformation is a priority, and real-time data and behavioral analytics are valuable, are prime areas for pilots and proof of concepts.
Digital transformation is a power driver in nearly all segments, and connecting the physical and digital world via computer vision is one of the most powerful ways this is accomplished. I believe security integrators have a unique role to play in that transformation, given our experience deploying and maintaining cameras as mission-critical sensors.
Another area of current interest is the broad desire for computer vision applications that enable or accelerate ‘return to work.’ Use-cases like mask detection, social distancing, and crowd detection are all areas of high interest. While these use-cases can be applied in many markets, care must be taken to effectively implement these solutions given the potentially significant differences in environments – manufacturing floor vs. retail checkout, as an example.
DeSousa: Integrators should target any customer who would find value in adding analytic technology to their system. This would also have to be a customer who has the necessary funds to be able to invest in AI. For example, a bank that wants to market that they are the most secure bank in the Midwest would find it extremely marketable to say they use the best in analytic and AI technology to keep
their investments secure.
What is the best way for integrator salespeople to make a case for AI for customers?
Yunag: I would encourage security integration salespeople to think bigger. Integrators need to start thinking and educating their clients about computer vision and AI platforms as force multipliers for security, but also for other business outcomes. As use-cases grow, the total addressable market grows, and educating the market on how to leverage visual intelligence for broader organizational use will be well worth the additional effort and engagement.
DeSousa: With recent events, having an analytic technology that can track persons of interest, or redact faces automatically, or ensure mask compliance could be of value. The starting point is to understand the customer’s problem, and the cost of these problems or loss of productivity. This will allow the best pairing of technology with return on investment or increase in productivity. Ensure that the cost of the solution does not outweigh the value you are trying to offer.
Are there any misconceptions about the technology, and is it meeting end-user expectations? Also, explain the importance of managing expectations and not overpromising from an integrator's perspective.
Yunag: There is a common misconception that AI can solve anything – “just use AI.” Because of this, end-user expectations when sourcing a solution labeled as “AI” are very high. Every use-case is different, and therefore, each recommendation should be carefully evaluated to ensure objectives are met.
When choosing which technology may be the right deployment for a customer, it is essential to understand how it will be used (very specifically), what data it needs to provide, and how it will help the company make better or faster decisions moving forward.
As integrators, it is our responsibility to manage customer expectations by clearly defining the business objectives first. This can be further supported by developing a proof of concept that demonstrates the capabilities in a real environment. Subsequent pilots can then be launched to validate the solution’s effectiveness before investments in a broad deployment.
In a time of breakneck innovation, “AI” companies are coming to market seemingly every day, and it is unlikely that all will be successful. The integrator’s job is to vet the right partners, understand the market fit, and ultimately how they can uniquely deliver value to customers.
DeSousa: The biggest misconception is that AI makes decisions by itself and learns by itself. AI needs to be taught in order to function. The benefit of AI is not that it can learn on its own – the benefit is that once fully trained, it can be leveraged for automation of forensic and real-time tasks. In many cases, it can perform tasks more efficiently and accurately than a human in unstructured environments. How well the AI is trained determines its effectiveness over that of the human eye and instinct.
The importance of managing expectations and not over promising comes from how well integrators educate customers. With AI being relatively fresh to the market, being able to effectively educate customers will reduce the risk of them interpreting AI for more than what it really is. Set expectations and show how AI will work for them in use cases as accurately as possible.
Learn more about Tyco American Dynamics and request more information at www.securityinfowatch.com/10212816.
