Source: ifsecglobal.com
The world is becoming more automated – from collaborative robots through to computer programs which can sift through thousands of documents almost instantaneously, organisations can now save time and money in new ways. The technology can now be used for necessary but tedious, time-consuming tasks that would take humans much longer and be more prone to error. However, there are aspects of automation which are misunderstood and misrepresented – I talk here about artificial intelligence (AI), where the hype is spreading faster than the aptitude of the technology.
Subsets of AI – like machine learning or deep learning are often referred to as AI, when they are not. In fact, they are closer to intelligent automation than artificial intelligence. Intelligent automation (IA) can help organisations by using existing data and automating analysis based on that data, ultimately helping to improve operations and workflow, as well as reducing redundant responses. But neither technology is truly “intelligent” in the sense that they cannot think or act like humans. We are many years away from that.
Artificial Intelligence – will it reach its true potential?
Artificial intelligence is certainly a buzzword in security, yet many capabilities are misinterpreted, undefined or misunderstood. Misunderstanding the capabilities of AI will often lead to unrealistic expectations.
In data science, AI refers to a fully functional artificial brain that is self-aware, intelligent and that can learn, reason and understand. While advancements in what is referred to as AI technologies have come a long way and will continue to do so, the reality of AI, however, is very different from an intelligent computer that can learn and make decisions like a human.
In practice and as it relates to the physical security industry, AI is a technology that runs a series of algorithms, searches through large databases or does calculations swiftly to provide deeper insights. The results can help users make decisions more quickly and efficiently depending on the application. General examples of applications that fall under “AI” would be facial recognition, object detection or people counting.
However, the broad nature of the term means that often, expectations and hype exceed its capabilities, and causes disappointment. Currently, only its subsets are tangible, such as machine learning techniques that include neural networks and deep learning. For example, deep learning uses task-specific algorithms to help train a computer to properly classify inputs. To do this, programmers essentially teach a computer by inputting a very large amount of data with corresponding labels, improving the technology’s ability to recognise new inputs. In a real-life scenario, deep learning is being used very effectively in automatic number plate recognition (ANPR).
It is possible to train the system by feeding it raw images of number plates, alongside parameters for it to work within, so it knows it can only class said images with possible outputs. This then lets the system take an image of the back of a car it has not seen before and identify the characters on the plate, along with other useful information such as location, colour and model. For a human, this would be a tedious and time-consuming task – but it’s ideal for computers with minimal human supervision.
Another vertical where machine learning can offer significant value is retail, thanks to its ability to monitor and identify trends. For example, such technology can help stores determine retail conversion rates or the number of people visiting a location versus purchasing. A high-accuracy deep learning algorithm can track the number of visitors and combine this with sales data to present valuable information to a human operator.
While highly advantageous for well understood applications, current AI technology has its limitations. Specific use cases and algorithms can certainly help organisations achieve greater operational efficiency, but it cannot teach itself completely new tasks or automatically make sense of data that it hasn’t been first taught. In addition, it can be difficult for users to interpret how an AI technology, such as deep learning, came to a decision or output.
AI does not give meaning to something on its own, but it can allow users to make more knowledgeable decisions and perform tasks more efficiently. If investigators run a face captured on video through a facial recognition database, for example, it’s important to know that faces returned as matches are not guaranteed positive matches but rather meet the probability requirements previously programmed into the algorithm. In other words, it can be a very valuable and necessary start for users to eliminate some mundane legwork and facilitate more informed investigations or decisions, but should only ever be used as a tool to aid human decision-making.
This is well illustrated by Met Police trials of facial recognition where all potential matches are sent to a trained officer to perform a secondary verification. The trials take advantage of the fact the technology can look at many more faces over a much longer period of time without getting tired than any police officer. Yet there is an acceptance that the trained human must have the final say as they have wider decision-making capabilities than any machine on earth today.
Defining Intelligent Automation
Intelligent automation also allows users to eliminate the donkey work, helping them make quicker, more informed decisions. It can automate some of those decisions too, as it integrates both automation and data together to recommend an outcome. One way to look at IA is that it uses an organisation’s existing data from different technologies and enables large-scale analysis to automate operations and improve productivity.
In security, IA can amalgamate and automate differing datasets, such as thermometer readings, video, incidents, facial recognition, number plate recognition, map-based data and other records. This correlation means humans are then able to assess specific problems or situations, instead of being presented with calculations of different databases but no combined conclusion. Like most automation platforms, IA is at its most potent when deployed for specific solutions.
For businesses to get the most out of intelligent automation, the technology should have a clearly defined environment where the emphasis is on the human input with machines doing the heavy lifting and not on the machines making decisions. With IA, humans review and approve machine decisions to help better drive outcomes.
In terms of deployment, critical national infrastructure is a strong example. These buildings have multiple systems, including temperature sensors, airflow sensors and a centralised security system. Intelligent automation can be used to automatically pull video footage, send a map of where an incident is located, and sound an alarm in the event that both the temperature spikes greatly and the airflow sensor states “danger,” suggesting the possibility of a fire or chemical spill. The technology can help enable an efficient response by initiating a specific standard operating procedure (SOP) when necessary – such as unlocking specific doors, notifying management, etc.
Whilst generating time and financial savings, IA can help create a culture of innovation. Automating processes frees up employees, allowing them to concentrate on more interesting and stimulating skilled tasks – learning more and yielding more value in turn.
Accuracy needs to be the priority, not marketing
There’s no denying that using buzzwords like AI makes a product sound sexy – even if it’s not entirely accurate. From toothbrushes to security programs, we regularly see the term misused and it’s time for that to stop. There is great potential for automation to benefit businesses by making more of data – thereby uncovering insights and informing decisions which was previously not possible.