Mission critical is defined as “vital to the functioning of an organization.”
Mission critical software is at the heart of every major technological system in the world: from electrical grids to nuclear power plants, from the international electronic payment infrastructure to international air traffic control, from the locally connected home to the globally connected internet of things. As everything becomes software, the resilience and security of this software becomes ever more important.
Artificial intelligence and machine learning have a major role to play in securing mission critical systems. However, realizing this potential requires technology and techniques with six key characteristics:
- It should work in real, real-time, (sub 10 milliseconds) rather than “near-real time. Most existing AI and machine learning technologies require databases and/or “big data” file systems (e.g. Hadoop Distributed File System, HDFS) to work in production. However, databases and big data systems were not built to support millions or billions of complex queries on millisecond time scales required to support mission critical systems.
- It should seamlessly integrate with existing systems. Existing and future mission critical systems run on varied operating systems and platforms. Successfully implementing AI and machine learning technologies on these platforms requires software solutions flexible enough to work “out of the box,” rather than requiring expensive custom hardware like AI-specific processors and/or in-memory machines.
- It should be resilient. Critical systems operate continuously, the tools that secure them need to provide 100% availability. However, many existing AI and machine learning techniques are niche software and/or are built on top of open source libraries and technologies that have yet to be tested in high throughput, continuous mission critical systems.
- It should dynamically scale. Mission critical systems create petabytes of data. Securing these systems requires AI and machine learning technologies that can ingest billions of events and transactions in real-time by seamlessly scaling.
- It should adaptively learn over time. The elements of mission critical systems each exhibit unique behavior, from individual spending behavior on payment networks, to the devices and and sensors that make up our critical infrastructure. Successfully securing and increasing the efficiency of these systems requires AI and machine learning technologies that characterize the behavior of every individual entity. Continuously characterizing individual behavior will enable these technologies to learn and adapt to novel or evolving behavioral patterns, from cybercriminals defrauding card holders in payments, to terrorists attacking electrical grids. Current approaches, like neural networks and data mining, cannot provide this level of resolution or rapidly adapt to novel or changing conditions because: (a) they can only characterize previously identified trends or behavioral patterns because they rely exclusively on historical data, and (b) they apply the logic they extract to every entity they are attempting to characterize.
- It should be data agnostic. Not only do mission critical systems produce billions of data points daily, but this data comes in many different formats and sources. Therefore, AI and machine learning technologies supporting mission critical systems must work with any data, in any format (structured or unstructured), at any volume, and from any source, from files, logs and transactions to machine sensor and network devices.