Recently, The Cove invited questions on AI and QT which would be answered by SMEs from Army's Robotic & Autonomous Systems Implementation & Coordination Office (RICO). Your questions were interesting and thoughtful, and RICO has provided insightful responses to these. Find them all below!
Question: Regarding Artificial Intelligence (AI) – how do you expect to align human military expert expectations with the often black-box systems that are provided by machine learning systems?
Answer: This is a really great question. There are a few aspects that we should decouple here, the first being the type of systems used and the second being the function of command.
There are many types of models with varying levels of opaqueness. Explainability is a concept broadly focused on how opaque these models are, or conversely, how we can understand the method by which the model arrived at a particular answer. A general rule of thumb is that the simpler the model, the more explainable it is (although this doesn’t always hold).
Explainability research seeks to provide a mechanism to understand how models arrive at their output, with emerging techniques successfully explaining quite complex models.
The second aspect here is the function of command. We often hear that an AI system will replace people in a range of roles and jobs, although we’d argue it is more nuanced to think about AI giving back capacity to a person to complete more human-centric tasks. The phrase we use within the RICO is transforming our people from transactional work to more meaningful and analytical roles.
The important point to note is that AI doesn’t absolve a person of their command function, it should enhance. The human retains the authorities and and permissions for action, as well being the custodian of the AI to be used.
While systems remain somewhat opaque, a commander could think about the decision support provided as a reference point or anchor to augment the decisions they are responsible for. As the machine learning systems become more explainable and the human commander gains more trust in the outputs, then we will begin to see the future promised by these systems. The output of these systems is another piece of context for a commander to consider.
Question: Can/should we use artificially intelligent tutoring in the ADF?
Answer: The idea that an AI can assist the ADF in tutoring is very interesting. There are many opportunities, although these must be contextualised by the objective of the tutoring. This could range from on-the-job support to upskill our workforce, through to a more prevalent AI-support agent that conducts low-level tasks to free up human capacity.
An AI tutor may provide personalised feedback to decease the time taken to learn a new concept. Equally, these outputs must be trusted to ensure that they are meaningful and do not reinforce potentially negative or incorrect ideas as the desired solution.
One way to quickly leverage such an opportunity is to integrate the technology into areas where organisationally we clearly understand the desired output and what right looks like. As we become more familiar with these technologies in such roles, we can look to integrate the technologies into more uncertain areas where the desired output is less well defined.
It’s important to remember that the human (commander) must retain responsibility for the systems used and cannot simply apply these technologies without oversight. If leveraged appropriately, the upshot here is an ability to tutor many more people at a scale beyond what may have been previously possible with only our workforce.
Question: Is it possible to use AI at a level that is useful to the military, while having transparency regarding the factors that the AI is using to reach results?
Answer: This is a really great question (and quite challenging to provide a concrete answer). AI model transparency could be considered as the accumulation of interpretability, explainability, and predictability. Each of these tenets provides a distinct lens to view AI technologies through; interpretability focused on understanding system behaviour through output observation, explainability of the logic coherently connecting inputs to effects, and predictability of the estimation of future system states.
Let us pose a narrative to consider this problem through. We start with a system by which a commander receives a particular piece of information at a certain level of confidence. The commander in this situation is unable to query the validity of this information or understand the means through which it was provided. Plausibly, this is the intelligence that a commander has received today for a mission that they are responsible for, based on information and assessments that they may not have the ability to access directly and enumerate all pieces of information. Similar examples are systemic throughout all decision-making processes.
The difference between the current system and an AI-enabled system is the way in which the information has been processed. While not a fool proof system, the human-centric approach is well understood and offers accountability for decisions and subsequent actions.
An important aspect is to integrate technologies at the speed of relevance, where they provide capability but are also provisioned in such a way that a commander can employ the outputs of these systems with confidence. Within RICO we see AI as an input to a system; we must remain focused on the outputs of the system and the context for its operation. The application of a transparent AI should augment human-centric processes and procedures to improve command decision making.
Question: What actually IS quantum technology (QT)? Can you dumb it down so that non experts can understand? What are some practical applications of QT for the ADF? Are any already in use?
Answer: As stated in the Army Quantum Technology Roadmap, Quantum technologies are a suite of emerging technologies that exploit the fundamental laws of nature to offer unprecedented capabilities in sensing, imaging, communications, and computing. They are diverse, complex, generally early in technical readiness and demand new ways of thinking about the employment and exploitation of technology.
Their true capabilities, limitations, most disruptive applications and associated countermeasures are still being discovered. Army is currently focussed on three areas of quantum technology: sensing, computing, and secure communications through key distribution and quantum-resilient cryptography.
Sensors employed in the ADF that most of us are familiar with rely on the exploitation of the electromagnetic spectrum, that is the spectrum of frequency ranging from radio through to visual light and on to x-rays. There are other sensors we use every day that detect things like acceleration and magnetism, accelerometers are in your phone so that the screen turns the right way up, and hall effect sensors can be found in the ABS of your car. Quantum sensors take the detection of acceleration and magnetism to new levels of sensitivity and offer the potential for sensors to never need calibration (a critical factor for precision navigation and timing). Army has worked with industry and academia to explore applications where sensors could detect concealed equipment through its gravitational or magnetic signature in both surface and subsurface applications. In subterranean exploitation and urban environments this could enable the detection of tunnels and bunkers by their absence of mass or weapons caches due to their magnetism. When developed fully, quantum accelerometers have the potential to fundamentally change the inertial guidance systems in complex weapons such that they can work in fully GPS denied environments with absolute precision.
Most people are familiar with general purpose computing, it proliferates our daily lives and is supporting you reading this article now. To appreciate quantum computing, a fundamental understanding of early special purpose computing is needed. Where early classical special purpose computing uses bit or binary storage of information, current quantum computers exploit quantum superposition and entanglement to represent and manipulate information in a fundamentally denser and more efficient way.
Note though, that quantum computers have a slower clock rate than classical computers so operations that are already sufficiently fast on classical computers won’t be improved by quantum computing. Discrete problems which currently take days or even years may however be treated by quantum computing in seconds. This is why Army has sought out computationally challenging problems such as logistics optimisation through the Quantum Technology Challenges.
Secure communication is fundamental to military operations. The breaking of enigma is regarded by many historians as the single greatest victory of the allies in WWII and is an example of classic cryptography. The underlying principle is that a key has to be shared between those communicating, currently there is no way of knowing if a key has been intercepted. With quantum key distribution (QKD), the key can’t be looked at without it being changed, this relies on superposition or entanglement of qubits and can be hard to understand.
The harsh truth, as expressed by Neil deGrasse Tyson, is ‘the universe is under no obligation to make sense to you’, and to explain quantum communication through the simplification of polarisation of light generally produces more confusion and is beyond the scope of this article. It is Army’s current assessment that QKD enabled networks will in all likelihood be limited to a few high priority links (for example, at the operational and strategic levels) due to technical constraints.
Quantum technology presents both strategic risk and opportunity for Army in an accelerated warfare environment. Its true capabilities, limitations, and most disruptive applications are still being discovered. Army is challenging Australian academia and industry through the Quantum Technology and Next Generation Challenges to push the bounds of what we currently understand and leveraging the significant investment in a world class research and industry capability.
We hope these answers have helped and look forward to continuing to support Army’s journey of innovation.
The Cove would like to thank RICO for taking the time to respond to these questions.