Human-Machine Teaming in Warfare
Hello, I’m Ylli Bajraktari, CEO of the Special Competitive Studies Project (SCSP). Welcome to the 2-2-2.
Save the Date:
Mark your Calendar! On September 16, 2022, we will be hosting our Global Emerging Technology Summit. The purpose of the Summit is to bring together government and private sector leaders from the United States and our staunchest allies and partners to ensure that emerging technologies help advance freedom, strengthen democracies, and protect the rules-based order.
SCSP Call for Engagement:
The call aims to answer the question of: “How can the United States and allies and partners better deter authoritarian aggression in the Western Pacific and Eastern Europe?” Specifically:
What low-cost techniques that could be implemented in 2-3 years might strengthen deterrence?
What strategies might the United States pursue that preserve our vital strategic interests more effectively than deterrence?
What does the United States get wrong about deterrence, including its relevance as a concept, how emerging technologies are changing the nature of deterrence, or how different cultures perceive deterrence differently?
From May 2 to July 29, 2022, SCSP will accept submissions in the form of a short paper or video answer to the prompt. Up to 3 entries will be selected to be published in the SCSP Newsletter, 2-2-2, as well as on the SCSP website, and SCSP will offer the authors of the three submissions $2,000 respectively. The authors of the selected submissions may also have an opportunity to brief the SCSP Board of Advisors.
Visit our website to learn more and upload your submission.
Human-Machine Teaming in Warfare
Edition 5
In this month’s edition of 2-2-2, Justin Lynch, Senior Director for Defense, Emma Morrison, Associate Director for Defense, and Research Assistants for Defense Fiona Pollack, Jessica Burrell, and Luke Vannurden, talk about military human-machine teaming.
Artificial intelligence (AI) will become ubiquitous in warfare. When many think of the intersection of AI and warfare, lethal autonomous weapons, or “killer robots,” are often the first concepts that come to mind. But many of the most important—and imminent—applications of artificial intelligence to warfare will not be fully autonomous systems, but human-machine teams: partnerships in which service members and intelligent machines work together to accomplish warfighting tasks. As nations begin integrating human-machine teaming (HMT) into their operational concepts and acquisition plans, it is important to understand what HMT is and how it will impact warfighting. In this newsletter, we will examine HMT efforts in the U.S. military, by U.S. allies, and in China and Russia.
What is human-machine teaming?
HMT is the combination of three elements: the human, the machine, and the interactions and interdependencies between them. A core concept of HMT is that humans and machines have comparative advantages and excel in different areas. Humans outperform machines on many sensory tasks, certain types of communication, high-context tasks requiring intuition, and many types of creative exploration. Machines often outperform humans at tasks that require processing extremely large volumes of data, a high degree of precision, memory, and consistent repetition. Augmenting human weaknesses with machine strengths (and vice versa), can create human-machine teams that outperform both humans and machines in many of their individual tasks.
How will human-machine teaming affect warfare?
HMT is at the center of several states’ vision for the future of warfighting and has the potential to considerably change warfare. Many of the earliest changes will be to cognitive tasks. A warfighter's mental bandwidth, as for every human, is limited. A decision to spend time solving one problem is a decision to not spend time on an equally critical task. The growth of HMT will enable individuals to break problems into their component pieces and, like Apple’s bicycle for the mind, task some to be optimized, automated, or performed at scale by a computer. This could directly improve outcomes as it has already in other fields, such as Amazon’s use of robotics and random stow in warehouses. It will also allow individuals to (re)focus their mental bandwidth towards gaining situational awareness, understanding enemy plans, developing courses of action, accomplishing far more than they would otherwise, and mastering the tasks that humans do best.
Another major change is that the balance of mass and effects delivery will shift towards machines. Today, in most cases, many warfighters collectively control one platform, such as a ship, as illustrated below by [1]. While that relationship is unlikely to vanish, at least two other human-machine relationships are developing that could begin to chip away at the dominant warfighter-platform relationship. One such mode is a small number of warfighters controlling many machines, such as DARPA’s OFFensive Swarm-Enabled Tactics (OFFSET) program, represented below by [2].
When using cheap, easy-to-manufacture machines, a small number of people may control a larger number of machines as a way to increase the mass and scale of military forces. Massed machines, guided by their human teammates, would potentially be able to overwhelm some traditional defenses, often at a relatively smaller cost in human casualties compared to more traditional offensive operations. Machines will also serve as the “eyes and ears” of their human teammates, helping them gain more information about their environment and taking risks in their place.
Another, less discussed relationship, is one wherein a small number of warfighters, skilled at software development, could create an application that optimizes the performance of many warfighters or machines down the line, as illustrated by [3] above. This relationship was already put to use by DIU when it commissioned the development of an algorithm to optimize aerial refueling operations, saving 350,000 pounds of fuel a week, a capability that is migrating to NATO. This HMT relationship has also been seen more recently as GIS Art for Artillery (ARTA), a Ukrainian military software used to distribute fire requests. The software enables multi-gun, multi-munition fire missions at a much faster rate than traditional targeting processes by allowing a machine to develop, interpret, and distribute guidance to operators.
For a warfighter, the ability to efficiently and responsibly delegate tasks to machines—a routine skill in missile defense—will be just as important as more traditional skills like land navigation and marksmanship. Military personnel will have to know when to trust or not trust their machines based on the machine's limitations, and vulnerability to cyber-attacks. This judgment will be as integral to their success as knowing their human teammates’ capabilities and personalities. We can expect that a special premium will be placed on warfighters who have the skills to program machines, and not just the skills to operate them.
Data-informed “coup d’oeil” is also becoming essential for military leaders. War’s inherent uncertainty and unpredictability often requires commanders to rely on intuition to understand events, and to decide their next course of action – hallmarks of Boyd’s OODA loop. At its most effective, coup d’oeil, the “ability to see things simply,” and “to identify the whole business of war completely,” presents as military genius. In Clausewitz’s time, coup d’oeil was embodied by Napoleon, a commander who could see an entire battlefield, and understand the terrain and units by looking at a map.
Today, many commanders and operators already benefit from data-informed decision-making. F-35 pilots ingest massive amounts of fused data, and operational-level commanders process many data streams every day. But this does not equate to turning data-informed decision-making into a national advantage. As operations become increasingly complex, and the amount of data flows grows, commanders at every echelon and in all domains will need to add data-informed decision-making to their intuitive decision-making.
An examination of the HMT efforts by the United States, our allies, and our adversaries, provides a sense of current trajectories in HMT and why the United States should lead in this area.
2 Perspectives: United States
One:
HMT has become an increasingly important part of the Department of Defense’s vision for the future of war. It represented a core part of the Third Offset strategy. Today, DoD has multiple programs underway to prepare for the HMT future. DARPA's "Adapting Cross-Domain Kill-Webs" (ACK) program is undergoing testing, and intends to provide decision-aid tools to allow mission commanders to dynamically combine assets across domains to create the most efficient "kill webs" or pairings of sensors and shooters in real-time.
The Office of Naval Research (ONR) has also initiated multiple programs on machine-aided military decision-making. Researchers in the ONR Program on Command Decision Making (CDM) are using concepts from AI and cognitive science to create "decision-support tools for warfighters” to support a variety of functions across mission types, including command and control (C2) assets for degraded environments.
A growing variety of private companies and defense contractors are also engaged in building advanced software-based machine tools to help U.S. military decision-making in conflict situations become more streamlined, informed, and robust. The Guardian AI platform from Rhombus Power, for example, uses machine learning to provide “strategic, operational, and tactical decisions at the speed of relevance.”
Two:
HMT in the U.S. military has already extended beyond cognitive tasks to work in physical spaces, and has programs to increase this over time. A team of engineers at MIT designed drones with the ability to breach doors and use LiDAR – light detection and ranging – to scan for armed adversaries before service members enter a building. In 2020, AI-enabled simulators in DARPA’s AlphaDogfight program bested human pilots at air combat maneuvering, illustrating the potential for machines to be integrated into squadrons as co-pilots. According to DARPA, AlphaDogfight represents a “crucible” for HMT and demonstrates the potential of “a collaborative relationship with an AI agent handling tactical tasks” while the pilot “focuses on higher-level strategy as a battle manager.” Eventually, U.S. service members will be able to team with whole swarms of semi-autonomous machines. DARPA's OFFSET program aims to allow small infantry units to utilize drone swarms to accomplish tactical goals in complex urban environments with what DARPA calls "human-swarm teaming.”
2 Perspectives: The United Kingdom and Australia
The United Kingdom and Australia are strategic allies to the United States, as evidenced most recently with the AUKUS agreement. They have also invested in developing and integrating HMT into their warfighting forces, and serve as notable examples for how to fuse current skills with emerging capabilities.
One:
Australia’s Robotics and Autonomous Systems Strategy, launched in 2018, articulates the Australian Army’s goal to introduce HMT into their ranks. Accordingly, Australia is increasingly investing in semi-autonomous platforms that will help augment the tactical capacity of human service members in combat. Boeing is building an unmanned combat drone for Australia, the MQ-28A Ghost Bat, which will accompany manned Australian fighter jets in order to “jam signals, conduct surveillance, and fire on assigned targets” while human pilots perform battle management and other vital tasks. In addition to the growing use of HMT in physical platforms, the Australian military is using AI to aid human decision-making. For example, Australia’s Artificial Intelligence for Decision-Making Initiative aims to find ways to use machine learning to inform and speed up military and national security decision-making. These efforts are especially significant as both the United States and Australia modernize and prepare for military operations in the Western Pacific with the third member of AUKUS, the United Kingdom.
Two:
The UK Ministry of Defence also plans to make HMT a core part of their operating concepts, arguing that human-machine teams will help “generate mass and tempo while reducing risk.” In 2022, the UK launched the Human Machine Teaming Project, as commissioned by the Future Force Development at Army Headquarters in London. Its goal is to develop a combat-ready Robotics and Autonomous Systems (RAS) – enhanced light brigade combat team by 2025, then broadly implement RAS technology throughout the division by 2035.
To accomplish this, Defence Equipment & Support and the British Army established an Expeditionary Robotics Centre of Expertise to bring together defense robotics and RAS projects under one roof. The Centre is set to test multiple RAS platforms. One example, Project Theseus, will have autonomous air and ground platforms deliver munitions, food, and fuel into the battlefield, ultimately reducing the number of warfighters entering fraught environments. The HMT project will also reportedly include the development of a “robotic platoon vehicle” and other platforms with semi-autonomous capabilities. HMT efforts in the UK will surely increase in the future: as reported in the Ministry of Defence’s 2018 report on the subject, “RAS will be a key means of generating mass,” resulting in “a high ratio of AI driven systems – both physical and virtual – to people” in the UK military.
2 Perspectives: China and Russia
One:
The Chinese government saw the risks and opportunities of the HMT in 2016 when AlphaGo, a computer program developed by Deep Mind defeated a Go World Champion Lee Sedol. This event appeared to have accelerated Chinese efforts towards future potential employment of AI in warfare. The impression made by AlphaGo’s performance in a game long tied to strategy led the Chinese government to invest in AI and HMT. In accordance with its strategy of “intelligentized warfare,” the People’s Liberation Army (PLA) plans to use HMT as an offset to U.S. power. One example of Chinese HMT comes from the military AI company StarSee. StarSee claims to have developed an AI-enabled, real-time system that uses image, audio, and video data to categorize an enemy platform, give a detailed overview of its capabilities, and help its user decide the optimal strategy to engage the platform. With this tool, PLA warfighters could conceivably streamline the decision-making processes in combat and mount faster, more effective attacks against their enemies. Another example is the PLA’s research into drone swarms. Designed to launch from naval carriers, drone swarms would attack other ships from many directions simultaneously, or probe air defenses, revealing their location for follow-on attacks by manned assets. This division of tasks would increase lethality while reducing risk to human warfighters.
Two:
In 2017, Russian President Vladimir Putin claimed the country that becomes the leader in AI “will become ruler of the world.” The Russian Air Force allegedly embraced this concept when it received the MiG-35 fighter jet in 2019. The jets reportedly have an “integrated artificial intelligence (AI) pilot assistant” named Rita, which makes audible recommendations when pilots approach boundary restrictions or engage in combat. Although a relatively simple development, more sophisticated AI technology is reportedly being integrated into Russian fighters. The newest Su-57 jets are expected to include an AI co-pilot, potentially removing the need for a second human pilot in the rear seat. The jet will also have systems that help control drones, and the ability to fly in an unmanned “drone mode” itself. Russia has also used loitering munitions in Ukraine. Systems such as the Lantset allow humans to designate an area and target type, and the machine to loiter, search, and attack.
These developments can allow machines to take care of certain tasks, such as neutralizing ground targets, while human operators accomplish other critical objectives. Russian defense experts claim that the new Su-57 jets may be paired with S-70 drones to “replace entire squadrons of piloted aircraft for reconnaissance and combat missions in the near future.” This formation would represent a unique combination of both intra-platform HMT—a human pilot and machine co-pilot collaborating on the jet—and inter-platform HMT—a manned platform collaborating with unmanned platforms—at the same time. HMT development would allow Russia’s fighter pilots to control more platforms, punch above their weight class in combat, and become a greater threat. These are Russian claims, but as their struggles in Ukraine suggest, it remains to be seen whether they will be as effective in practice as they appear impressive in presentation.
Questions that Preoccupy Us
Will HMT reduce human casualties?
Increase in operational risk appetite. As militaries are likely to increasingly rely on human-machine teams, the ratio of humans to machines is likely to decrease, with attritable machines taking more risks than their human counterparts. With this change, commanders can execute previously unacceptable maneuvers with the reassurance that combat losses will mostly consist of machines, not human lives. In other words, ground commanders can be expected to fight more determined defenses with higher risks to force, knowing that many of the casualties on both sides would be machines and not someone’s family member. This shift could potentially alter the assessment of proportionality and necessity in evaluating targets, as well the political risk-benefit calculations of national leaders.
Balance in HMT Capabilities. Because state’s have different levels of resources and experience with autonomous technology development, some states will significantly outperform others in regards to HMT. Due to this potential imbalance, if one state could overwhelm or bypass their adversary’s machines, they would likely continue on to attack the human elements of its adversary’s military, or even its population. This would result in a significant imbalance in casualties between the combatant states, where the advantaged side would sustain few casualties while the disadvantaged side would sustain many.
In conflicts with more symmetric HMT usage, the number of casualties sustained by the combatant states would depend on the circumstances of that specific conflict. In general, states are likely to increase the use of machines in order to minimize human casualties. During limited wars, or wars fought “to achieve specific political objectives, using limited forces and limited force,” states are more likely to seek a negotiated outcome after a large portion of their machines have been neutralized or destroyed, but before they have lost a large number of military and civilian personnel.
In total war, a state’s political ambitions are often higher, the percentage of their contributed resources is greater, and their military’s targeting practices are less discriminatory. In these conflicts, states are more inclined to continue fighting despite machine losses and human casualties. In a future where HMT is more prevalent in conflict, it is likely that machines will be designed and directed to attack other machines, and their operators. This is congruent with the ever-adapting nature of warfare. Early military aircraft served in reconnaissance roles, not unlike when unarmed drones were first introduced. However, militaries quickly developed air superiority fighters to shoot down other aircraft, bombers to attack infrastructure and populations, attack aircraft to provide close support and interdiction for ground forces, and escort fighters to make sure bombers reached their target. It is reasonable to expect a similar shift to occur for HMT.
How will HMT Change the Propensity for War?
Based on the scenarios above, HMT can be understood as increasing or decreasing the number of casualties an attacker inflicts on its adversaries, and increasing or decreasing the number of casualties an attacker sustains during operations. When the attacker’s HMT is superior to the defender’s HMT, it will inflict more casualties, and sustain fewer. When the attacker’s HMT is inferior to the defender’s HMT, the attacker is likely to inflict fewer casualties, and sustain more. In cases where the attacker and defender’s HMT is evenly matched, their casualties are likely to be evenly matched as well, but overall, will be a function of their political commitment to the war.
These casualty dynamics could have a considerable impact on states’ propensity to go to war. In situations when both attackers and defenders can expect fewer casualties, policy makers face fewer risks when they go to war. When HMT would cause a state to sustain fewer casualties while inflicting more on their adversary, the state would risk less when going to war. In cases where states would sustain more casualties, they would be more likely to be deterred.
How quickly will war change?
Sometimes, change takes place incredibly quickly. In 1914, large infantry formations marched towards battles of annihilation escorted by horse cavalry – a sight that would have felt familiar to soldiers from Napoleon’s army, who likely would have recognized most of their tactics. Yet, a soldier from August 1914 suddenly transported to the summer of 1918 would be astounded by the changed operating environment of trench warfare, massed artillery fire, tanks, and battles for air superiority. If such rapid changes have happened before, they can certainly happen again.
HMT has the potential to rapidly change warfare again. The emerging changes in the relationship between humans and machines has the potential to alter how wars are fought, and how they are won, from the tactical to the strategic level. The United States, United Kingdom, Australia, China, and Russia, are all making significant investments into HMT technology, and its role in their operational concepts. And the private sector’s research into HMT creates numerous opportunities for dual-use technologies to accelerate change.
No matter how quickly HMT changes warfare though, many of the challenges humans face will remain the same. Warfighters will still rely on their teammates and will still experience the fear of failure and the fear of death. Adapting more quickly than adversaries will remain critical. Achieving political outcomes will still require eliminating a human actor’s options or forcing them to change their risk calculus. And in the end, military and political leaders will still have to understand human nature, human decision-making, and outcomes created by humanity.