The military is at the forefront of technology. From the creation of the internet via DARPA to its development of the Dark Web as an anonymous web, one branch of the American military has been there through it all. Indeed, much of modern warfare can be carried out halfway around the world from the front lines – all that’s required is an internet connection.

It should come as no surprise that the military is one of the biggest artificial intelligence enthusiasts currently working in the field. Indeed, the military has invested heavily in several vessels of this emerging technology, particularly in robots, drones, and future weapons.

Areas of Interest: Artificial Neural Networks, Deep Learning, and Weapons

The U.S. military’s interest in artificial intelligence isn’t simply to avail of it to improve efficiency. On a trip through Silicon Valley in autumn 2017, Secretary of Defense James Mattis noted how important it was for the Department of Defense to better integrate AI into its policies. But what does this mean in a military context?

Artificial intelligence evolves every day as computers take over tasks once performed only by humans and then learn from those tasks to become “more intelligent.” Recently, significant focus has been placed on software that doesn’t just learn but can ‘think’ independently. The field is referred to as Deep Learning, and it includes a set of techniques to teach computers to learn via Artificial Neural Networks.

Artificial Neural Networks (ANNs) are a type of software designed to mirror the human cerebral cortex loosely. Presently, the software is much simpler than the human brain and focuses seriously on only one task at hand.

ANNs are run through processing units that include three layers: input, output, and hidden. If you want to compare these layers to the human brain, you can say that input layers would be likened to the retina: your eye takes in information to send to the brain.  Output layers are similar to the visual cortex in the brain. Hidden layers process the signals from the input (retina) and send them to output (the visual cortex).

A simple network will have a single hidden layer – a single translation piece. However, it’s not possible to build ANNs with two, three, or four hidden layers. Added layers allow great complexity and Deep Learning.

Deep Learning imitates the workings of the cerebral cortex, but more importantly, it works with unstructured or unlabelled data in an unsupervised fashion. Working with unstructured data requires the software to make inferences and see past immediate characteristics to better analyze and collate data, therefore, learning from data that is more complex to produce observations that were previously impossible.

A New National Defense Strategy

AI isn’t something the military is vaguely considering as an added tool: it could transform and dominate all military strategy in the not-so-distant future.

A 2017 report from the Department of Defense outlined some of the key ways the DoD might not just employ but weaponize this technology.

According to the DoD’s science research division, the U.S. military will soon use robotic systems, drones, and hacking and cybersecurity software based on a foundation of artificial intelligence.

Let’s break them down here.

Military Robots

The employment of artificial intelligence in military robots provides them with the potential to carry out missions on their own without gross human intervention.

Indeed, the premise of artificial intelligence is to allow for a program to learn from a huge collection of data and streamline the decision-making process, reaching the right decision nearly instantaneously compared to the minutes, hours, and days of research and thought that go into the process today.

The use of robots isn’t new: they serve all kinds of mundane functions including working as packhorses, transporting gear on the front lines.

Military Drones

Drones have been in use in the military in some form since 1918 when the U.S. used an unmanned aerial vehicle dubbed the Kettering Bug as a cruise missile in combat. Since then, drones have evolved to become smaller and faster, and the technology has become more widely available. Today, 86 countries around the world have military drones, and the consumer drone market has exploded.

Artificial intelligence will only sharpen the use of drones – and it’s already in use. The U.S. military is using Google’s TensorFlow AI system in its drone program. The system allows the military to use machine learning and artificial intelligence to cope with the ludicrous amount of footage collected by U.S. drones.

Initially, TensorFlow was picked up to analyse the video and find interesting sites to be flagged for review by humans. Using AI saves an incalculable amount of analyst time and frees them up for targeted searches and research, allowing analysts to do twice as much work.

Google says that the use of its software by the military is for non-offensive uses only. In other words, the software isn’t being used to target individuals for assassinations. Indeed, Alphabet has been careful to point out that is recognizes the concerns of using civilian products in a military context and notes that it is constantly discussed as polices and safeguards are developed both internally and externally.

Pushback: Problems and Criticism in Military Artificial Intelligence

Military applications of technologies naturally come with pushback: using a robot to vacuum the floor is very different from using a robot to carry out a lethal task.

Still, there are clear advantages of using AI. Artificial intelligence isn’t limited by the human knowledge used to develop the software; it can learn through countless iterations of scenarios. As AI programs learn, they can even transfer their knowledge onto humans, providing training in scenarios that are incredibly uncommon.

Despite the benefits, there has been significant pushback on the current and potential use of AI in the military in part because the military is using software and technology from the private sector. The biggest criticism has been in the problem with lethal action.

Banning Autonomous Killing Machines

In 2017, 116 specialists from 26 countries petitioned the United Nations to ban of killer robots. While semi-autonomous weaponry has been in use since World War II, the combination of lethal weapons with artificial intelligence worries many critics. Still, even the most careful critics haven’t outlined what weapons and uses of those weapons should be banned, opting instead for the ban of “morally wrong” lethal autonomous weapons systems.

Weaknesses in Technology

Regardless of what side of military technology one stands on, there are some clear disadvantages to relying on artificial intelligence in the military. AI will only be as good as the data it receives: faulty data leads to poor outcomes and could cost the lives of civilians and result in a tragedy that cannot be undone.

Additionally, AI and Deep Learning aren’t yet ready to take on roles that require a deep level of contextual knowledge. Missions requiring this kind of experience would not yet be possible. Finally, the opaqueness of the technology may lead some to feel uncomfortable deploying it in new scenarios.

Overturning Current Key Processes

One of the greatest internal limitations lies in present military operations. Currently, the military deploys a verification and validation process designed to ensure every decision is carefully made. Verification and validation is well suited to frozen software; it is deeply challenged by AI, which learns and adapts on its own.

Adopting AI would mean adjusting or even scrapping verification and validation processes. Such a dramatic change could come with serious mistakes that have fata consequences. Tainted or false data might be spotted by analysis combined deep contextual knowledge, but it might be mishandled by an AI system, particularly on the first go-round.

Implications of Military Artificial Intelligence

There are several implications of the advent of military artificial intelligence that are often glossed over in favor of more emotional factors like saving soldiers’ lives.

The replacement of frozen software with AI could potentially slash military spending because the software wouldn’t need to be rebuilt and replaced periodically. An AI system would be more nimble and better able to adapt to changing environments and data requirements.

AI could also be used to solve logistics challenges that consistently plague the military.

At the same time, artificial intelligence could also make the military more vulnerable than it is presently. Because the software isn’t limited by human knowledge, recruiting the best analysts and engineers in the world isn’t required for developing an AI program with the potential to do serious damage to security systems.

Military and AI: The Future Is Here (In 20 Years)

Military artificial intelligence is already used in rather benign ways: its primary function is in sorting data rather than carrying out battles or engaging in warfare. What is more, the kind of AI required to perform the lethal functions that worry critics could still be 20 years away from a prototype much less a fully operational program.

Whatever your take, the military is an early adopter of new technologies because they make military operations easier, cheaper, and more secure while also saving lives.

What do you think of the potential use of military AI? Share your thoughts in the comments below.