Military simulations allow training of troops without the need for physical risk and allow them to refine their tactics and strategy before entering a live situation. These simulations need to be run by live operators that are effectively assisting in training and are typically reserved for officers quite high up in the chain of command. Live operators need to prepare a scenario beforehand - this includes choosing the geography and topography, time of day and weather conditions, etc. Then they need to lay out the situation before the exercise begins - this includes mission, equipment, personnel, possible obstacles and limitations. Finally, they need to actively engage with the officer in training during the entire simulation, as a truck driver, or helicopter pilot, or front-line soldier, or any number of positions with a narrow set of actions and knowledge of the scenario as a whole.
The reason for this is very simple: Hard coded patterns posing as true AI are simply too weak, repetitive and predictable to serve as a trainer for such important and live-saving skills. Fake AI might be a lackluster substitution for live human participants in video games, but certainly not even close to good enough for military-grade training simulators.
As you can probably imagine, these live operations are very expensive - not only the time spent pre- and during simulation, but also the design well in advance of the scenario. In order for the training to be effective, it has to be challenging, varied, adaptive, innovative, unexpected and unpredictable. There is a huge amount of effort and man-hours behind each scenario and every step of the way as it unfolds, and this entire cycle needs to be repeated and adapted for training different and specific aspects officers need and/or want to address.
Human live operation is the single largest obstacle to the efficient operation of military simulators. Decisive AI proposes to change that radically.
Imagine an AI that operates as an Intelligent Artificial Agent in tactical training simulators. When I say AI I do not mean hard coded patterns thought out by a human programmer developing the imitation behaviour of a tired, repetitive, predictable and limited piece of code (since that would defeat the purpose). I mean the opposite: fresh, adaptive, unpredictable, ever challenging and unlimited variations of behaviour developed by the computer through machine learning, reinforcement learning and other AI methods.
Training an AI with clear objectives in a virtual environment is not new - Decisive AI and other companies are already doing this for the video game industry. The principle is the same for simulators: the objectives the AI has to achieve need to be clear and editable, as well as its resources; the “natural” rules of the environment need to be well defined and impossible to break by all parties; the visibility, perception, and overview need to be defined and addressed in order to avoid unsolicited omniscience.
Part of the beauty of this approach is that once the AI is well trained and capable of giving a tactical officer a good challenge in the war room, it continues to learn and adapt its strategy. The more exposed it is to the expertise (or lack thereof) of human trainees, the more its experience grows, together with its capacity to produce countless scenarios and address myriad situations unmatched by any team of human live operators and scenario designers.
The result is as evident as it is exciting: better allocation of precious military resources making the use of simulators efficient, while enhancing the effectiveness of virtual training with more scenarios, unpredictability, variety, and adaptability than ever.
Just to reiterate, since this is so amazing it was probably too shocking to absorb: it is not just the significant recurring live operation costs that this value proposition addresses nor the impact on being able to run these simulators efficiently - the enhancement, improvement, and sheer volume of training scenarios is key to the contribution of this solution too.
So - not just reductions in quantity of time and money, but increase in quality and variety of the training itself. Well, AI might not be able to fight by our side in the battlefield yet, but it certainly is starting to do so behind the scenes.