Within the UK Military Cyber domain, much is discussed about the topics of non-kinetic effects, Cyber security and cryptography. Rightly so, these subjects draw a lot of attention as they raise the immediate questions of the ethics of automating processes within the ‘kill-chain’ and the protecting of our digitally held information. As important as these subjects are, they appear to be shrouding other highly relevant subjects in the field, namely the use of computers to do what they were created to do, that is computation.
The fields of Artificial Intelligence (AI) and Machine Learning (ML) are expanding in the commercial and academic centres with what seems like daily breakthroughs being announced in the media, from automated accountants to driverless cars. This article suggests that we as the military should be looking at how to integrate AI into our processes so as not only to make our decision making immeasurably better, but to guard against the risk of falling behind the inevitable wave of technology that is sweeping our world.
The subjects of AI and ML overlap in many ways but can differ slightly in their approaches to problem solving. Broadly speaking, AI may be defined as solving a problem algorithmically where the method of solving the problem is known and so a computer will apply the algorithm to solve said problem much quicker than a human mind would be able to, many times, without tiring. In the case of ML, the way in which the problem is solved is not necessarily known in the first instance. A program is ‘trained’ by being shown the correct solutions to a set of problems and by ‘learning’ how to achieve the correct solution, teaches itself an algorithm so that it may then solve subsequent similar problems. Additionally, any further solutions that are found by this method improve future results just as a human will increase their chances of success in an exam by revising more questions.
The following examples will work to highlight just how powerful computation has now become and the benefits that may be granted when they are used to their potential. The most used single page web application to date is Google Maps. Using powerful point to point route finding algorithms like the Djkistra Algorithm it is able to tell a user how long it will take to get from one end of a continent to another in the matter of seconds whether they are travelling by car, bicycle or public transport. A possible application relevant to Air Manoeuvre might be rapid tactical route planning for a C130 low level airdrop mission, where the point of departure and location of an airdrop are known. The program would then select the best possible route to take given any number of factors including distance, time, risk, etc.
Another method called natural language processing enables an AI agent to read a host of documents and make inferences about their contents which would take a human doing the same job several orders of magnitude more time than it would the computer. The practicalities of this has already been demonstrated by the ‘lawtech’ firm Linklaters who have created a program called Verifi which can sift through 14 UK and European regulatory registers to check client names for banks and process thousands of names overnight. A junior lawyer would take an average of 12 minutes to search each customer name. A possible application in the military context of this method would be to ‘feed’ an AI agent a host of administrative JSPs to learn so that a service person may submit a routine request that would usually go to their unit admin cell (Expense claim, leave request or an RFI on unit numbers or competencies to name but a few examples) and get a timely, correct answer any time of day.
These two examples just scratch the surface of what the power of computation can provide but what is important to note is that we cannot (yet, if ever) expect these so-called AI agents to act totally autonomously. Just like an Air Force with highly technical equipment requires highly technical people, a military with integrated AI processes requires people to understand the systems that they are working with so that they may be monitored, tweaked and improved over time. The benefit of these systems that would require supervision just as a junior technician does is that we need not concern ourselves with the career progression or how many hours a day a computer may work. The computer will not be posted to another unit and take valuable knowledge with it gained over the course of a tour as the junior technician might. The computer will stay to do the job that it was created for, to compute. Arguably most importantly in these times of tight governmental purse strings is that the computer will not require a salary. Of course, it will cost money in the first case to invest in the technology but there exists a crossover where the outlay cost of automation becomes worth it.
The range of tasks that may be solved by a computer is limited only by the imagination of the person sat at the keyboard. In the field of AI, if a process may be written down explicitly it may then be translated into pseudo-code, algorithms applied and experimented with and finally written into real code to be used as a program to solve problems and if needed be subsequently tweaked to make the problem solving even more efficient. In the field of ML the process need not even be fully known, all that is required is a suitable solution set that is known to be correct and the program may learn how to solve a problem itself. This is not all to say that these techniques are easy to handle. A whole range of background knowledge in Mathematics and Computer Science is required to make such tools, but once wielded, these tools may well provide computational power unknown to a military any time prior to now and would allow the creation of maintainable efficiencies across the MOD as a whole.
The views expressed within individual posts and media are those of the author and do not reflect any official position or that of the author’s employees or employer. Concerns regarding content should be addressed to the wavellroom through the contact form