The US Special Operations Command is drafting its first own artificial intelligence strategy

What is the concept of artificial intelligence and when did it appear?

As it has many meanings and sense, a universally accepted definition for artificial intelligence (AI) is hard to identify. It seems that the term AI was used for the first time in 1956 by John McCarthy, a reputed computer scientist from the US Stanford University who, among others, also invented one of the oldest programming languages – LISP – and founded the field of AI. He wrote in is “A proposal for the Dartmouth Summer Research Project on Artificial Intelligence – 1966” study that the research will be based on the presumption every aspect of learning or any other characteristic of intelligence can, in principle, be defined so precisely that a machine could be built to simulate it.

In any case, after World War II, several researchers started to work on intelligent machines independently. English mathematician Alan Turing would have been the first. He held a lecture about in 1947, and is apparently the first who decided that AI would be best researched by programming computers rather than building machines. By the end of the 50s there were many AI researchers, and most were basing their work on programming computers. One of the AI research communities was created and financed by DARPA, approximately 60 years ago.

DARPA defines AI as the programmed ability to process information. The agency considers that three waves of AI technology happened until know. The first was named “Handcrafted Knowledge”, as computers could only study the implications of rules which engineers were providing them on a certain area, which was very good in making logical judgements without being capable in any way of perceiving the outside world, studying, abstraction or managing uncertainty.

The second wave, “Statistical Learning”, was marked by the debut of using automated Machine Learning (ML) techniques, very efficient in voice and facial recognition. This made computers very good at perceiving the real world, but also in studying and adapting to various situations based on provided data, but with very limited capabilities of making logical judgments or abstraction. Automated learning as an AI appliance which offers systems the capacity to learn and automatically improve from the experience they gain, without being programmed in this sense. Automated learning techniques contain Reinforcement Learning, Deep Learning and Artificial Neural Networks.

The first two waves’ limitations led, according to DARPA researchers, to the need for a third AI technological wave, “Contextual Adaptation”, in which the systems themselves will build basic explicative models which will allow them to characterize real world phenomena. They will be able to use the explicative model to make logical judgments and even to take decisions on those phenomena, while also being capable of abstraction. Until then, however, there is still a long way to go and much work to be done, according to DARPA specialists.

Despite all this, it is more and more obvious that AI is everywhere around, from autonomous vehicles and drones to virtual assistants and search engine recommendation, translation, financial transactions software, or recruiting labour force in the field of human resources. It is also increasingly used in the areas of security and defence to: develop autonomous weapons systems and intelligent munitions, improve target acquisition, develop ISR capabilities, develop predictive analysis applications in intelligence, improving cyber defence capabilities and others.

The American 365 SOF AI strategy

Obtaining a competitive advantage in the upcoming military clashes will certainly presume the use of autonomous and intelligent weapons systems capable of completing, in the end, effort made by human fighters (special operations forces – SOF, for example) and improving their performance and efficiency on the battlefield.

American SOF understood this and went to implementing measures to protect their competitive advantage in developing and using AI and ML technologies, critical for defending the US’ security and national defence interests against strategic competitors and foreign adversaries.

The project of the first 365 AI strategy was unveiled by USSOCOM Chief Data Officer David Spirk Jr. during the Special Operations Forces Industry Conference (SOFIC 2019) held in Tampa, Florida, in May 2019.

The strategy will describe what the American SOF will do and how they will proceed to adopt and become comfortable with using AI and ML techniques, towards the purpose of taking faster decisions with a way higher degree of accuracy. They didn’t invent anything new, they took inspiration from industry, saw what works there, read Andrew Ng’s rules of how to lead your company in the era of artificial intelligence and modelled their own strategy based on that pilot project, as well as lessons learned in applications developed by the Department of Defense’s (DOD) Join Artificial Intelligence Center (JAIC), but also from results of the Maven Project (a Pentagon project destined to use AI and ML technologies for the purpose of differentiating people from objects in thousands of hours of drone footage).

In drafting the AI strategy, American SOF realized that everything revolves around data and took inspiration from the principles used by Jeff Bezos in Amazon since 2002, about liberating data, using open-source application programming interfaces instead of closed technologies or systems.

Furthermore, the US used the leading principles stated in government documents such as the Executive Order on Maintaining American Leadership in Artificial Intelligence drafted by the Presidential Administration in January 2019, the DoD’s 2018 Artificial Intelligence Strategy – drafted by JAIC specialists – and were able to develop what the USSOCOM calls the SOF AI Principles.

According to these principles, American SOF will identify the constraints related to expanding AI and ML technologies and will also identify purpose-driven applications from operators deployed in theatres of operations or zones of interests up to strategic commands. At the same time, they will allot the necessary resources and transform human resources into a professional unit in the digital field, capable of finding and requesting opportunities for human-machine teaming, which will act on the entire spectre of combat actions and functions. They aim for AI and ML technologies to be a regular occurrence within 10-20 years, and American FOS to be a spearhead in this regard within the DoD.

The basis of this first AI strategy is the “365 SOF Artificial Intelligence Strategy”, built around three lines of effort (AI-trained workforce, implementing AI applications, interinstitutional extension of AI implementation), six areas of interest (perception ad action, planning and executing manoeuvres, resilience of cyber communications and defence, talent recruitment, instruction and management, predictive maintenance, logistic planning and predictions, the management of acquisitions and contracts) and five anticipated collective results (cloud computing-based data and services, the widespread use of Agile management practices in developing classified and non-classified software products, accelerating digital technology purchases for deployed troops, attracting talents and instructing the workforce, transitioning to a sustainable digital future).

Current FOS applied AI projects

Perception and action – the fields of ISR, exploitation and dissemination, as well as captured enemy material are in the responsibility of an algorithmic team specialized in several fields, also known as the Maven project. As intelligent systems will carry out more such tasks, the released workforce (intelligence analysts, for example), will be retrained, but part of it will have to monitor the execution of the intelligence cycle. Identifying and providing the answer to message threats and fake news will be in the responsibility of Military Information Support Operations (MISO) capabilities.

Planning and executing the manoeuvre – virtual reality (VR) and artificial reality (AR) technologies will be used to operationalize combat capabilities and make SOF operators faster. USSOCOM realized the value of the feedback received from SOF operators instructed under VR and AR headsets, taking into account the experience accumulated by the best of them in the past 19 years. This feedback will be used to develop AI technologies which will be able to learn from the best operators and later contribute to the creation of the best human-machine teams in the combat zone.

Talent (human resources) recruitment, instruction and management) – AI will help to identify ideal candidates, evaluate the performance of operators, adjust instruction and training programs to maximize the effort of operators and make them faster. USSOCOM will make use of some technologies already used in sports or some areas of medicine.

Predictive maintenance, logistic planning and predictions – these are among the most interesting fields with the most progresses achieved, according to David Spirk, as data-based technology is used to predict maintenance problems before they appear. There are attempts to extend it towards the fleet of tilt rotor CV-22 Osprey insertion aerial platforms.

Instead of conclusions

The AI strategy of American SOF is the solid expression of their vision on how to adapt their own capabilities to the new challenges brought by the fourth industrial revolution to the security and defence of their country, which began to change the nature of security risks, vulnerabilities and threats. AI is omnipresent and is achieving significant progress, due to the exponential rise in computing power and the availability of huge quantities of data. In the end, all SOF operators will have to understand the way in which data can be used on the battlefield and perfect their basic abilities of working with intelligent systems.

Currently, the need for human control on AI factor is still obligatory, because these systems will need to reach way higher intelligence levels before they are used on the battlefield. A reduced reliability of autonomous and intelligent systems used in the fields of defence and security can lead to dramatic consequences, including the loss of human life. And, to convince myself, I asked one of my PAs the following: “Alexa, can you tell me how old I am?” And it answered: “You know yourself better than I do! I think you are Daniel, this is Daniel’s account”. Personally, I was expecting way more, as my personal data was available.

If only to achieve interoperability with other NATO partners and keep in line with technological evolutions, Romanian Army SOF should start to think how they will adapt to the new challenges, as Romania is in the process of drafting its first national AI strategy, with a 2019 deadline.

Translated by Ionut Preda