To win the data war, defence forces look to leverage specialised skills in digital automation and artificial intelligence.
Organisations of all kinds are examining ways to use digital automation, under names including artificial intelligence, machine learning or bots. But defence agencies have among the strongest motivations to adopt this group of technologies, known collectively as intelligent automation, given the ever-increasing volumes of data they need to analyse accurately, safely and quickly, in the interests of national security.
“Intelligent automation is absolutely essential for the military,” says Ian McDonald, Director of Technology Enablement in Defence and National Security, KPMG in Australia. “They cannot operate the capabilities they currently have to their full potential without it.”
Take ballistic missile defence, a vital part of many countries' military activity. As many as a dozen satellite and sensor systems may be used in detecting hostile missile launches, with a further dozen systems involved in destroying such missiles. A defence agency could have just 8-10 minutes to decide whether a launch represents a threat, share findings with allies and decide what to do. The use of countermeasures has to happen quickly, given that missiles could impact 16 minutes after launch.
“At the moment, there are humans in the loop between those systems,” says McDonald. “You cannot respond within 16 minutes, unless you are already in a state of high alert.” Intelligent automation can rapidly integrate all the data from satellites and sensors and present findings, allowing personnel to decide what action to take.
Intelligent automation, which David Kirk, Managing Director of Advisory for KPMG in the US, defines as “the automation of mission delivery and business processes by leveraging digital technologies to support tasks undertaken by knowledge workers,” can be split into three levels of sophistication. The first level, robotic process automation, can make routine tasks more efficient in areas including finance, accounting, human resources and compliance. KPMG firms have worked with defence clients to automate payments between agencies, including securely accessing systems and obtaining and processing the data required.
The second level of intelligent automation, learning cognitive automation, often involves the use of natural language processing and chat-bots. Kirke Everson, Managing Director of the Intelligent Automation Group at KPMG in the US, says this has potential to automate routine communication such as helpdesk ticketing and directory inquiries. “Chatbots can understand sentiment and intent, they take care of multi-thread questions depending on the datasets plugged into them and they tend to learn over time based on interactions with humans,” he says.
`Virtual agents' that use multiple channels including mobile and desktop computers, telephone and voice-activated speakers, are already used by some companies and civilian government organisations. They could allow military personnel to get answers to straightforward questions on travel or housing, pulling together data from different agencies, leaving humans in call centers to handle more complex queries. “It's moving from a user interface to a conversational interface,” says Payam Mousavi, Director of Public Sector Advisory Services for KPMG in the US. “It makes accessing that information a lot easier and faster.”
The third and highest level of intelligent automation, reasoning cognitive automation, is able to derive insights from large amounts of data, including unstructured material such as written text, audio and video. In administrative work this has potential in recruitment and training, where existing staff records could be used to propose which specialisms a new recruit could be trained in.
But there is stronger potential for reasoning cognitive automation in improving battlefield decision-making. Currently, when a scout tank at the front of a group crossing hostile territory is running low on shells or fuel, it requires radio communication with the group commander to order another tank to take over the lead role. However, this process could be automated by analysing data generated by the tanks. “What you can do is speed everything up on the battlefield,” says Ian McDonald.
Integrating all the data from an area into a battle management system can improve safety, by accurately keeping track of allied forces and friendly non-military personnel such as aid workers. “Bringing all that data together is where intelligent automation comes in,” McDonald says, given the large quantities that need to integrated quickly. This is particularly important for the most advanced drones, which can collect terabytes of data – including visual, radio and radar – in missions lasting more than 24 hours. Intelligent automation can allow this data to be analysed in near real-time, highlighting what looks important within minutes rather than days later.
With higher levels of intelligent automation, systems need to be trained with data that includes inputs and the conclusions drawn, allowing them to model which inputs should lead to which conclusions. This requires reliable sets of training data, some of which can be gathered from the routine use of military aircraft, ships and land vehicles. It can also be generated by digital simulations, war-gaming and physical military exercises, including how personnel behave in them.
Data used to train an intelligent automation system has to be accurate and unbiased for its conclusions to be useful, and ensuring this is the case requires the right skills. “It's important you have someone with a data science background,” says Kirke Everson, as well as involving people with on-the-ground expertise, "someone who understands the use-cases and objectives of what you are trying to do.”
Payam Mousavi adds that a range of expertise helps to minimise bias: “There is some governance, some control and an approval process for what learning goes into the system – and multiple experts training the machine, rather than one person.”
Many defence agencies lack these at present, as militaries have tended to attract people who are good at operating machines and vehicles rather than computers, while defence-minded people with IT skills are often sucked into cybersecurity work. “There's a global shortage,” says Ian McDonald.
Kirke Everson adds that it is vital for intelligent automation projects to use people with a strong understanding of the technology: “That's where we're helping clients the most, to say you need a business architect, a robot process automation developer and three domain experts to implement this solution, and we can help you do that.” KPMG in the US recruits graduates of university courses that include machine learning and artificial intelligence, to strengthen its technical resources.
Intelligent automation projects also need strong cybersecurity, and defence agencies set high requirements, including controls on authentication and segmenting of duties. Intelligent automation techniques can themselves be used in building stronger cybersecurity, by analysing data on previous attacks for patterns.
As the need for skilled humans in its development suggests, intelligent automation is less about raising a robot army and more about making defence personnel more effective. For an officer in the field this can mean less time spent on routine administration and more on mission-critical work. “It can provide a force multiplier,” says Everson.
Ian McDonald says the number of unmanned vehicles is expected to increase, including driverless tanks and fighter aircraft, as this will speed decision-making, reduce costs and remove personnel from harm's way. But he adds: “You won't ever remove all the humans – you can't.” Indeed, defence agencies are likely to create new roles for people who can monitor and understand automated systems, to determine they run smoothly and safely.
“Agencies feel much more comfortable in having a human in the loop,” adds Kirke Everson. “You still have a human saying I agree with this, or I don't agree with this.”
Intelligent automation will involve significant changes in the way defence agencies work, and it can make sense to start with pilot projects. Everson recommends building a business case for such change through small proof of concepts that can show a good return on investment. In the US, the flexible and quick Other Transaction Authority (OTA) process designed for prototyping can be appropriate for these. It can make sense to start with projects that are trained on structured data, such as that held within databases or spreadsheets, as they are easier to evaluate.
Simulations, war-gaming and physical military exercises provide a good place to trial intelligent automation, as well as gathering data for it: “Build tomorrow's battlefields, tactics and strategies, and prove it that way,” McDonald says. If an intelligent automation system can save the price – both financial and human – of a single aircraft or ship, it will easily justify its cost.