The term “Artificial Intelligence” consists of a couple of words; artificial and intelligence. Oxford dictionary explains “artificial” as manmade or something produced by humans and “intelligence” as the ability to learn, understand and think in a logical way. Hence, Artificial Intelligence means manmade things, or computer algorithms/programs that help machines to think, learn, act and make life easier. The term was first coined by John McCarthy in 1956, though AI’s journey started way before that. It has been around for quite a while yet considered as an ambiguous subject in computer science. It has also been related to robotics or some form of superman like figure that can think and act on its own.
Though, the use of AI has increased many folds globally, security managers generally do not have the desired level of understanding of its applications and still consider it a farfetched thing rather than a currently available technology that stands poised to revolutionize almost all industries, let alone the security industry.
This could be partially attributed to the confusion that surrounds AI as to what it can achieve and what it cannot. In order to bridge the gap between what AI is actually capable of achieving and what the world thinks it can, it must be understood that some advanced applications of AI require extremely expensive hardware and exorbitantly high processing power, besides encompassing serious ethical issues like facial recognition or AI powered body scanners without permission.
For facilities warranting a comprehensive Physical Protection System (PPS), a significantly large portion of their budget is spent on keeping the desired work force (guards) in order to secure the facility, however, owing to paucity of this budget or even experienced guards, security managers are mostly found struggling. That’s precisely what makes AI so relevant and indispensable in days to come. This article is aimed at providing an insight to the security managers about this technology and the role it can possibly play in enhancing the security of any organization/facility.
Aim
To analyze the role of AI in Physical Security with a view to highlight trends and opportunities associated with this fast growing technology in order to enhance security managers’ awareness.
Artificial Intelligence (AI)
AI is a branch of computer science in which machines are programmed and given cognitive ability to think so that they can mimic actions like humans and animals.
Simply put, AI refers to machines that mimic the functions or actions that are generally related to or associated with humans, e.g., learning and problem solving. For better understanding of security managers, three levels of AI are briefly summarized below:-
Artificial Narrow Intelligence (ANI)
ANI is also termed as weak AI is the AI in which the machine can perform a specific task better than a human. In Narrow AI, it is important to understand that the machine can perform only one task at a time e.g., playing chess, analyzing some data set etc. The tasks can be performed in real time but it cannot perform more than a single task, however, with perfection e.g., Deep Blue by IBM, which defeated chess grandmaster Garry Kasparov in May 1997, or Google’s DeepMind’s AlphaGo, which defeated 18 time world champion Lee Sedol at Go in 2016.
ANI is the one that surrounds us all today e.g. RankBrain by Google, Siri by Apple, IBM’s Watsoon, facial recognition software etc. These applications might seem strong but they are no-where close to have human like intelligence.
Artificial General Intelligence (AGI)
An AGI is considered to be such a machine that is programmed to be able to think, analyze as good as humans would and capable of executing a range of tasks at a time rather than tackling a single task as in case of ANI. As of today, AGI does not exist and has only featured in science-fiction stories.
Artificial Super Intelligence (ASI)
Super intelligence is a hypothetical term that is supposed to be possessing intelligence far surpassing the intelligence of the brightest and most gifted human minds on earth. Dr. Nick Bostrom, the director of “Future of Humanity Institute” at the “University of Oxford”, explains super intelligence as “any intellect that greatly exceeds the cognitive performance of humans in virtually all domains of interest”.
Machine Learning (ML) – A Subset of Artificial Intelligence
A well written code in any programming language for a certain purpose actually tells the machine to follow that set of instructions in a sequence, however, in Machine Learning; the historical data (data already inputted in the machine) helps it to analyze patterns in the data and make decisions. In simple words, it is that part of AI that provides intelligence to machines so that it can learn with experiences automatically without having been programmed.
Types of ML
Supervised learning: Learning in which the machines learn from already known data sets to predict the output.
Unsupervised learning: Learning in which the machines learn without any supervision or any training to predict the output.
Reinforcement learning: Learning in which machine is given training by giving some instructions, and seeing its actions, it gets a reward as feedback. Using these feedbacks, machine improves the performance.
Deep Learning (DL) – A Subset of Machine Learning
It provides the ability to a machine to perform tasks that are like humans without any human involvement. Deep learning is implemented through neural networks which adapt and learn from huge amount of data. The major challenge in implementing Deep Learning is that it requires immense computational power.
In the training phase, you can show a machine a number of different images and it will learn to find out the differences between them. Once the neural network has learnt about the data, it can then intelligently infer to interpret new data based on what it has already learnt. e.g., if it has been shown a number of horses, the system will automatically know if the new image is a horse. In the succeeding paragraphs, AI and its relation with Physical Security will be briefly highlighted.
Artificial Intelligence (AI) and Physical Security (PS)
A physical protection system (PPS) uses people, procedures and equipment / technology (software and hardware both) for the safety and protection of any asset/facility. Despite methodically designed and evaluated, it will still have some gaps in it.
The same PPS can be improved by integrating technology offered by Artificial Intelligence (AI) which has the ability to detect the specifics out of the ordinary, and raise flags to the exceptions which then alerts timely human intervention, or automated actions such as the summoning of emergency services etc. It will change the security dynamics in future by reducing the number of people at the scene; introducing intelligent and knowledgeable systems, bringing large sizes of data from multiple sensors, analyzing the patterns in the data thereby limiting crime and assisting guards deployed to protect an asset to expose impending risks.
Let us have a closer look at the role of AI in physical security and how can it assist in enhancing the already placed PPS in any facility.
Artificial Intelligence and Access Control
Even if an efficient PPS is in place, access control at any entry/point remains vulnerable owing to human involvement who can compromise either intentionally (whenever there is a need to) or unintentionally (due to fault). However, once AI is applied to existing PPS (access control mechanism to be precise), it can quickly identify unusual activity such as off-hours access and/or access from unauthorized locations; combine it with other threat indicators to quickly identify threats and raise an alert once there is a need. Additionally, in fluid situations where a threat to security is detected warranting on spot decisions, it can quickly make practical recommendations and execute complete/zonal lockdowns or emergency shutdowns or both, if required.
Artificial Intelligence and Crime Detection
Artificial Intelligence is designed to learn various changing and repetitive tasks using complex algorithms. This continuous learning process helps it take systematically better decisions capable of further strengthening facility’s security. This feature also makes it possible for it to distinguish and detect specific unauthorized personnel out of the numerous persons been recorded.
Whenever the system notices an unusual event having monitored various breaches, it triggers an alarm. It then learns the pattern of what led to the warning so that it becomes self-aware. When a similar situation occurs, it repeats the same event.
Artificial Intelligence and the Control Room/Security Operations Centre (SOC)
AI can regenerate the cumbersome, manually or even automatically generated reports in the SOC making them easier to understand for business owners who want to know how well their facility is secured and what are the vulnerabilities. Besides, SOC team often uses new technologies for security analytics, impact analysis, risk analysis etc. for its clientele which has always been relied upon word of mouth, news and social media monitoring and intelligence reports from different sources. However, with the advancement in technology, a host of artificial intelligence tools are now available with the sole purpose to automate, improve and eventually perform those manual activities. Besides, these tools also provide opportunities and functionalities that could not have been possible earlier using manual analysis alone.
Artificial Intelligence and Facility’s Surveillance Systems
According to Carnegie Endowment for Global Peace, “at least 75 out of 176 countries globally are actively using AI technologies for surveillance purposes with China in the driving seat and the figure is likely to increase in years to come”. This is because:-
The standard CCTV systems can assist in analyzing a crime only after it has been committed i.e., they act reactively whereas, on the other hand, with AI technology, if a crime is being committed, it can be seen in real time and appropriate actions taken (also recommended by the AI driven technology).
Likewise, though, the facility may have a large number of cameras deployed all around it but cannot be effectively monitored 24/7 owing to human limitation. It is a well-known fact that human attention can span to about 20 minutes once engaged in a routine task and it then starts decreasing paving way for errors. AI can overcome this limitation by analyzing input from a large number of cameras, sensors, footage and detecting anomalies in real time, alerting operators to issues to prevent any unlawful entry to any facility so that criminal acts can be prevented, or dealt with as quickly as they occur, rather than beating the bush after the incident has happened.
Besides, if a crime has already been committed, it becomes cumbersome for an operator/human to watch hours and hours of video and identify the problem. Imagine a scenario where one has to use internet without the help of any search engine; searching through one page at a time to look for something. This is how most video surveillance search these days, however, with AI technology, it is possible to watch/scan hundreds of hours of video in few minutes to identify the suspects, or victims, thereby saving time and giving as less reaction time as possible to perpetrators. That is precisely where AI comes in.
Artificial Intelligence and Facial recognition
Facial recognition is identifying an individual from an image/set of images or videos. Intelligent, AI powered facial recognition software can search huge databases instantly in real time, compare them with the existing ones and produce results within seconds, However, there are serious ethical implications associated with the development of AI facial recognition e.g. collection of data without consent.
Artificial Intelligence in larger gatherings/crowds
Though, CCTV cameras provide a means for security personal to monitor crowds, however, these may not be able to cover complete area in consideration. The problem is further exacerbated by the short/limited attention span and lack of expertise of operators. AI-enabled cameras and sensors keep a track on the surroundings/large gatherings at various access points in parks, shopping malls, functions. This is particularly useful once there is shortage of manpower besides covering a large area with utmost accuracy and precision even under limited/low lighting that would not have been possible with naked eye. With the help of these cameras/sensors, certain explosives can even be detected besides being able to discern items with certain shapes like firearms etc.
Future Trends
Integration for Better Results: Wireless cameras (running on facility’s internet connection), motion sensors, predictive AI and human intervention can be integrated to detect any unauthorized entry in a facility. As soon as the motion detector senses motion, deep learning algorithms come in to play and determine whether it’s a human, animal or some other object that is moving. Simultaneously, the facial recognition application starts determining whether it’s the resident, a family member or a stranger.
Predictive AI technology tries to determine person’s actions/behavior through the use of wireless cameras. If the facial recognition application does not identify the individual as resident or a family member and/or the predictive AI technology also reports suspicious actions, the system raises alert or even calls police. This all is done in under a minute and can be very useful in any facility’s security management.
Access Control for Larger Crowd: These days, there are personal screening machines capable of removing long lines; screening one person per second without having them to empty their pockets or even stop through advanced facial recognition application by capturing video footage while the algorithms at the back end simultaneously start matching visitor’s face with already stored data and raise alert if the match is not found. If the incoming visitor is not found in the data base (unverified), the onsite security guards can physically check the person by taking him aside. Proponents of these advanced machines even claim to detect concealed weapons as well. The bottom line remains, “the larger the database, the better”.
Searches based on Physical Appearance: Possibility of searches based on physical appearance / description or facial recognition or both combined has also been proposed by one the leading companies which allows operators to go through hours of video within no time and builds a set of evidence from multiple video sources. Key features include:-
Extremely quick search (person/vehicle).
Self-learning video analytics.
Integrated with indigenously developed software.
Unusual Activity Detection (UAD) and Unusual Motion Detection (UMD): An AI technology that learns what a typical activity looks like and then detects the unusualness (activity/motion) in it. This also helps the operators to go through hours and hours of footage in minutes and take corrective measures.
Surveying / Securing Wide Areas: In order to secure facilities with large areas like farms, oil and gas facilities and water utility sites. AI driven physical protection systems can be used which integrate cameras (both wireless and thermal), sensors and video analytics that provide a single view of the complete area under consideration. Any/all intrusion patterns, including but not limited to anyone (person or vehicle) entering the area or carrying out suspicious activities are detected, recorded over a wireless network and alerts concerned security managers/security operations centers/control rooms.
Conclusion
Undoubtedly, the future belongs to AI. Traditional view of physical security is changing and the future includes more than just the guard force, barriers and batons. More reliance will be on intelligent, sophisticated yet simple systems, though, the humans aspect is unlikely to be totally eliminated. Various countries around the globe use HD AI enabled cameras, advanced facial recognition software and innumerable sensors gathering heaps of data with events to be analyzed by complex algorithms in real time, thereby indicating potential risks and raising flags. With AI, out of ordinary acts can also be monitored, something which may not be possible through operators alone.
