FUTURE PEACE technology, aggression, and the rush to war
ROBERT H. LATIFF
University of Notre Dame Press Notre Dame, Indiana
contents
Preface Acknowledgments
Introduction
xi xvii
1
ONE
A Giant Armed Nervous System
000
TWO
Urges to Violence
000
Stumbling into War
000
Avoiding War
000
Conclusion
000
Notes
000
Index
000
THREE
FOUR
INTRODUCTION
And not one will know of the war, not one Will care at least when it is done. Not one would mind, neither bird nor tree If mankind perished utterly; And Spring herself, when she woke at dawn, Would scarcely know that we were gone. —From Sara Teasdale, “There Will Come Soft Rains”
Sara Teasdale’s lyrical poem “There Will Come Soft Rains” at first appears to describe a postapocalyptic, postnuclear world but, written in 1918, was actually a response to the wanton devastation of World War I. At once a paean to the environment and the beauty of the natural world, it was also an indictment of the irrational, destructive nature of humans and the absolute futility of their war making.1 1
It has been just over a century since the end of World War I, a war widely considered to be wholly unnecessary and excessively brutal. The events leading to that war were not rational causes for plunging the globe into total war, and the insanity and mercilessness with which it was fought frightened even the most callous observers. This was a war in which new technologies promised quick victories but had the opposite effect, and in which the massive egos of national leaders prevented meaningful discussions.2 Indeed, if there was ever a case when war was irrational, this was it. More than nine million soldiers and five million civilians are estimated to have perished under occupation, bombardment, hunger, and disease.3 In the twentieth century alone, 108 million people were killed in wars. One researcher boldly estimates that the world has been completely at peace for a mere 268 of the past 3,000 years.4 Since September 11, 2001, direct war deaths have numbered 480,000, with over 240,000 civilians killed as a result of fighting.5 This raises the inevitable question, what do we mean by peace? Is it only the absence of armed conflict? This is clearly the major goal, but perhaps it also means that people can live in safety, without fear or threat of violence from any source, including their own government, or that they live in a society where there is a rule of law. Certainly, the Cold War was largely without armed conflict between the superpowers, but no one would call that a peaceful period. Rebel groups, guerrillas, and terrorists were, and still are, active all around the globe. Some foreign leaders have oppressed, or even killed, their own people and continue to do so. Starvation still plagues parts of sub-Saharan Africa. That doesn’t sound like peace. In the last century, the numbers of war dead have been steadily decreasing. However, while the absolute number of deaths has fallen, the frequency of smaller conflicts has risen and deaths of noncombatants have increased at an alarming rate.6 While armed conflict is down, the potential for it is dramatically higher now than it has been in recent memory. Harvard psychologist Steven Pinker concludes that since war and combat deaths have continually dropped over time, the world is becoming more peaceful. Viewed from the perspective of classic armed conflict, perhaps, but that is far too narrow a conception to arrive at such a rosy conclusion. Pinker fails to account for the large 2
INTRODUCTION
number of “peacetime” deaths of innocents and noncombatants due to internal strife and repression and sectarian conflicts fought increasingly among the populace. As author John Gray suggests, rather than merely focusing on the fact that war is declining, we should accept that the differences between war and peace are becoming fatally blurred.7 Unsurprisingly, many, if not most, of the current conflicts have their roots in long-standing ethnic and political rivalries. Superpowers avoid directly confronting one another for fear of escalation into nuclear conflict, but the list of proxy wars is long. It includes the Korean War, the Soviet invasions of Hungary, Czechoslovakia, and Afghanistan, the Vietnam War, military aid to insurgents in Syria, and the proxy war between the US and Russia that is being waged in Ukraine, to name only a few. Now, advanced technologies are available to more participants. They make war easier and are making matters worse. War has been a central feature of human societies for millennia. Its character has changed dramatically over time with the growth in populations and the inexorable advance in weaponry. Numerous economic and sociological studies have found a surprising regularity and cyclical nature of war and conflict and have even attempted to correlate upswings in technological innovation with the frequency of outbreaks of violence.8 So-called modern war marked a sharp increase in destructiveness. The Napoleonic Wars following the French Revolution fundamentally transformed armed conflict.9 Wars became larger in scale and more totally destructive. The Industrial Revolution accelerated this trend, introducing numerous new technologies that contributed to the slaughter of World War I. Since then, of course, technology has raced even further, with an awful increase in destructive power. The latter decades of the nineteenth century and the years leading up to World War I reflected an increasing industrialization of warfare, with weapons becoming ever more deadly and with the increasing mechanization and dehumanization of combat. The Crimean War (1853–56) and the US Civil War (1861–65) marked the emergence of a large and powerful armaments industry and the beginning of a widespread dependence on advanced technologies and weapons.10 The rapid growth in weapons arsenals suggested a potential to use those weapons. Countries engage in arms races to achieve qualitative and numerical superiority. War is more likely when the offense has an INTRODUCTION
3
advantage, and as classic “offense-defense theory” predicts, that likelihood is shaped by the technology that is available.11 American culture, enamored as it is with technology, is prone to technological hubris and holds a fundamental belief that any adversary can be defeated by our technological and material superiority.12 Such hubris has undoubtedly led decision makers to resort to conflict rather than do the hard work of diplomacy. One need look no further than the 2003 US invasion of Iraq, which was supposed to make quick work of a technologically inferior enemy. Technology for war has advanced at dizzying speed with the development of the airplane and the tank in World War I, radar and nuclear weapons in World War II, later the development of precision-guided bombs, and now computer and informationbased weapons. The character of conflict is now changing yet again and in seemingly arbitrary ways. The United States has been involved in some type of armed conflict for almost two decades, but sometimes it is unclear why, and who the enemy actually is. Much of that time has been spent waging a nebulous “war on terror.” And our engagements have not been so-called classic wars. There are new tactics, including crime, guerrilla activities, and other mischief, in what is often referred to as “hybrid war” or war in the “gray zone.” The 2014 Russian incursion and annexation of Crimea stands as a perfect example, with its “little green men”—soldiers whose uniforms bore no identifying markings.13 Today’s wars are unlike the more predictable affairs of the past: the techniques and tools adapt themselves to the environment within which they are fought.14 While we are not at war, per se, the world is certainly not at peace. It could also be argued that we are again in a new “Cold War,” with China now being added to the standoff with the US. There are no armed conflicts among superpowers, but the relations are antagonistic and the actions increasingly provocative. Militaries, especially the US, are heavily dependent on computercontrolled and soon to be autonomous systems for everything from logistics, to command and control, to actual killing. Increasingly, decisions are being taken out of the hands of human beings. At the same time, the major powers and others are becoming more bellicose and aggressive. The worry is that these trends—automation and aggressiveness—like the superposition of waves, will interact and amplify one another, ex4
INTRODUCTION
acerbating an already dangerous, high-pressure national security situation around the globe. The US military is heavily armed, deployed all over the world, and connected and controlled by a vast computer and communications network. The highly networked nature of military forces worldwide and their heavily armed status present us with the possibility of a “hair-trigger” situation and mentality. There are ways we can technically modulate military behavior and responses, however, and the public can and should demand more accountability from its leaders concerning the control and use of its military. Technologies for war are developing rapidly and are becoming more available to nations besides the so-called superpowers. Instant communications are causing events to unfold with frightening speed and with sometimes unfiltered and erroneous information. In earlier eras there was a time lag between an event and a potential response. Now everything is connected to everything else, and any action demands an instant reaction. These trends, in tandem with the confusion that springs from a growing vulnerability of data and information, do not bode well for a peaceful world. It is urgent that they be addressed, mitigated, and, hopefully, reversed. The time scales of warfare have shrunk dramatically. The ubiquity of networks, rapid global communications, and computer-aided decisionmaking, the twenty-four-hour news cycle, and the omnipresence of social media have all conspired to minimize the ability of a military commander to take the time necessary to properly assess a situation, think through rational courses of action, and decide. In 1807, after the British attack on the USS Chesapeake, it took four months and a round-trip ocean crossing for communications between President Jefferson and British foreign secretary George Canning to decide the US response.15 That time allowed the situation to be defused and the outrage of the population to abate. In the Civil War, we had the telegraph. In World War I we had radio. Today we have satellite communications. Whereas before there was time to consider, debate, and reflect, the current drive toward the use of automated and autonomous decision-making seems to push things even further toward an authoritarian, literally nonhuman way of identifying and responding to threats. Time has become a tyrant to be deposed as a controlling INTRODUCTION
5
factor in our decision-making. However, the systems we use are designed and built by humans, and those same humans can ensure that the systems operate in a way that preserves peace rather than inciting war. It is only for us to demand it. The unfortunate fact is that the major world powers—and many emerging ones—are excessively militaristic. The US in particular tends to view global issues as requiring military responses. We see this in the growth of the defense budget and the dramatic cuts to the State Department and the diplomatic corps. Defense budgets worldwide are growing, and countries are deploying their militaries around the world with more frequency. Diplomacy is taking a back seat to militarism. Spending on new weapons technologies in particular is growing rapidly. These technologies will add new capabilities to an alreadylarge arsenal of lethal weapons possessed by nations around the world. Something must be done to ensure that the new weapons and systems are properly controlled and that militaries around the world find ways to reduce the chances of war, not increase them. Countries and groups around the world are heavily armed, and the global arms industries provide a steady flow of weapons for their use. There are flashpoints everywhere—Russia and eastern Europe, China and the South China Sea, the Middle East, the Korean Peninsula, and Africa, to name a few. There are numerous issues—both diplomatic and technical—driving the competition, and there are any number of ways in which we might get drawn into conflict in those areas. In each case, the availability of advanced technologies and weapons is a prime motivator of, or at least an excuse for, behavior. The ready availability and large stocks of advanced weapons provide the implements of war, but festering problems among the world’s populations set the conditions for violent conflict. Income disparity, religious and ethnic hatred, lack of education, arms proliferation, demand for respect by other countries, instant availability of information, and the promise of new superweapons are some of the forces pushing countries and groups toward violence. Today, technology almost always plays a role in either causing the violence or contributing to it. Hezbollah employs rockets to terrorize Israel. Israel destroyed a Syrian nuclear facility using a combination of cyber and electronic warfare techniques and kinetic weapons.16 The 6
INTRODUCTION
2003 Iraq War was fought, ostensibly, to prevent that country from deploying weapons of mass destruction. In 2008, Russia employed cyberattacks in its invasion of Georgia, and unmanned aerial vehicles were used in that conflict. In 2015, Russian cyber forces shut down the Ukrainian power grid. These developments further confuse classic, long-held views of warfare. War is ethically and morally problematic. Over the centuries and decades, after the introduction of new styles of conflict and new weapons, nations have seen fit to place restrictions on their armies. With time, a set of rules and standards emerged to limit the brutality and curb the excesses of combatants. Norms of behavior were established to protect the innocent, and the use of inhumane weapons was outlawed. International laws, like the Geneva Conventions and the prohibitions on chemical and biological weapons, have resulted. In light of the longstanding proclivities to war and the emergence of a new class of technologies for fighting wars, it is imperative that we step back and consider again the moral and ethical implications of weapons employing these technologies and wars employing these weapons. The tenets of just war theory and the laws of armed conflict are dependent on the decisions of human beings. This is important. Topics of jus ad bellum (justice in going to war) require decisions about proper authority, right intention, just cause, last resort, and others that are just not in the purview of a machine. The idea of a machine having intentions or making judgments about justice is simply meaningless. Likewise, requirements of jus in bello (justice in war) cannot be decided by a machine. Decisions about military necessity, proportionality, legitimate combatants, and what constitutes unnecessary and superfluous suffering raise deep questions of fairness, human dignity, and morality and are the province of human commanders. The original just war concepts were developed in a time when limitations on war making were constrained by the weapons themselves. The evolution to so-called modern war, and today’s hyperwar, in which technology has rendered those constraints obsolete, makes the search for restraint ever more critical. The American public is dangerously uninvolved, and seemingly uninterested in, questions of the use of the military around the globe. Admittedly, there are sometimes good and valid reasons for military INTRODUCTION
7
force, but data on deployments show that US armed forces have been repeatedly used by presidents, often with questionable rationale. Why is the public not more critical? We must find ways to involve the citizens in issues of war and use of military forces and to identify ways of slowing the rapid resort to arms, raising barriers to political leaders who more often than not take the military route in lieu of doing the hard work of diplomacy. We must slow down the impulses to action and push back against them. Throughout history, statesmen, military leaders, religious leaders, and intellectuals have decried violence and argued against war. Strong, respected voices of reason spoke up through the media of the day and often resisted the calls for war. Today, few, if any, such voices exist, and even when they do, they are drowned out by others in the press, on the internet, and on social media. Thought leaders will have to adapt more quickly to this reality if they are to once again regain credibility. Voices of restraint must provide a rationale against military conflict and highlight the damage, unseen by the public, that it does to the country. But we cannot just rely on people. Changes to rectify this dangerous state of affairs must be both structural and technical. Resistors and capacitors are important elements in damping the sometimes wild and uncontrolled oscillations of electrical circuits and other physical systems. We must find a conceptual way of building such damping devices into the complex computer and communication networks of the military command and control system, as well as the upper levels of decision-making, to tamp down the urge to action. We must deploy some speed brakes, both physical and human, into the system to resist the urges to violence and the propensity to act in the absence of rational thought. The military is a vast, widely deployed, overstressed enterprise with highly complex systems. The world is a pressure cooker. There are numerous places and opportunities for the system to be thrown out of equilibrium, and our command and control system must be up to the task of preventing such an occurrence. Added to the situation is the fact that far too many urges to violence push us toward war and away from peace. We must find ways to reduce the pressure, limit knee-jerk reactions, and resist the urges to war. 8
INTRODUCTION
O
N
E
A GIANT ARMED NERVOUS SYSTEM
A strange game. The only winning move is not to play. —War Games, 1983
In the early morning of September 26, 1983, Lieutenant Colonel Stanislav Petrov of the Soviet Air Defense Forces was on duty in the Moscow command center where the Soviet military monitored its early-warning satellites over the United States. Suddenly and unexpectedly the alarms went off, warning that nuclear-armed intercontinental ballistic missiles had been launched from an American base. Colonel Petrov was obliged to report the incident
9
to his superiors at the warning-system headquarters and ultimately to the Soviet leadership, who would decide whether or not to launch a retaliatory attack. All of his computer-controlled electronic systems were warning him of imminent nuclear destruction, but with years of personal experience under his belt, Colonel Petrov struggled to absorb and make sense of streams of incoming information, ultimately deciding to ignore the alarms because the launch reports were probably false. He later explained that it was a gut decision, based on his distrust of the early-warning system and the small amount of credible data. But for the sane, reasoned approach and calm nerves of one skeptical—human—individual, the advanced technologies of the military systems of the time might have plunged the world into a doomsday thermonuclear war.
THE MILITARY: A VAST, FAR-FLUNG ENTERPRISE
The US military is huge, deployed all over the world, highly interconnected and increasingly automated, and heavily armed. With large investments in high-speed and high-bandwidth communications, and connecting its weapons in a vast internet-of-things, the military is becoming, as one commentator says, “a giant armed nervous system”— an interesting and apt description.1 There are approximately 1.3 million men and women on active duty, of whom 200,000 are stationed overseas in 143 countries. There are 750,000 Department of Defense civilians and 1.1 million National Guard and reserve personnel.2 The individual military services train, organize, and equip forces that they then provide to combatant commanders to carry out specific missions as ordered by the national command authorities. Armed forces of major military powers are deployed around the globe. In excess of 60,000 US forces are stationed in Germany and other western European countries, and deployed forward in Poland, Romania, and Hungary. In Asia, there are over 130,000 American troops, 10
A GIANT ARMED NERVOUS SYSTEM
largely in Japan, Hawaii, and South Korea. The air force has strategic bombers in Okinawa. The navy has aircraft carrier battle groups steaming in the South China Sea. Around the world, there are US special operations troops in over a hundred different countries. Russia has large numbers of troops and equipment in Syria, Armenia, Moldova, Georgia, Ukraine, and Belarus. China has forces in Hong Kong and on the mainland, directly across the straits from Taiwan. US combat forces are organized into regional commands and functional commands, each headed by a four-star officer. There are regional combatant commanders for the Indo-Pacific region, Europe, the Middle East, Africa, Latin America, and the US homeland. There are functional combatant commands for transportation, special operations, nuclear, space, and cyber missions. There are forces in over eight hundred bases around the world, many in inhospitable areas where troops operate under dangerous conditions.3 Of the deployed troops, the largest concentrations are in Asia (38 percent), Europe (32 percent), and the Middle East and Africa (13 percent).4 There are thousands of aircraft, hundreds of ships, including aircraft carriers and submarines, and tens of thousands of combat vehicles. Nine countries have their own nuclear weapons, with five others hosting them for the US. While other countries have military forces deployed, none, by far, has such a worldwide presence as the US. Connectivity between military units and with command elements is essential with such widely dispersed units. Deployed forces depend heavily on a concept called “reach-back,” in which they are dependent on assured communications and logistics with headquarters in the US and central locations in Europe or Asia. The military is almost totally dependent on networks for both tactical and strategic command and control, logistics, services, communications, and even weapons functions. Those networks depend on thousands of miles of high-speed optical fiber cables under the ocean and scores of communications satellites orbiting the earth. War is increasingly conducted using networks of computers, weapons, sensors, and electronic warfare systems. The heavy dependence on these networks poses serious challenges, particularly if combat results in interrupted connectivity, limited data rates, or disruption as a result of electronic or physical attack. The everincreasing sophistication and global nature of those networks make A GIANT ARMED NERVOUS SYSTEM
11
the information environment more extensive, more complicated, and more complex than ever before. Accurate data and information are essential to military planners and commanders of combat units. The US has a vast array of sensors around the globe. There are sensors on most aircraft, ships, and vehicles. There are also space-based sensors. Satellites monitor the earth for missile launches and are used for photoreconnaissance and signal detection. They provide accurate position, navigation, and timing signals for everything from financial systems, to personal automobiles, to missiles and bombs. Ground radars and telescopes, as well as onorbit systems, keep track of objects in space and on the ground. All of these sensors feed the battle management systems that will be employed in the event of a conflict. Military forces use multiple networks, the sensitivity of which depends on the sensitivity of the data and missions being performed. There are commercial networks, an unclassified military-only network, a secret-level network known as SIPR (secret internet router protocol), and the JWICS (joint worldwide intelligence communications system) for top-secret data. Much in the way of effort and resources goes into keeping these systems secure. The DOD has the world’s biggest network of networks, with a budget of close to $40 billion, about ten thousand operational systems, hundreds of data centers, tens of thousands of servers, millions of computers and devices, and hundreds of thousands of commercial mobile devices.5 Add to that a heavy dependence on both commercial and dedicated military satellites, microwave, and optical fiber networks. DOD supports more than forty agencies and manages about four million cyber identity credentials. In today’s scenarios, speed is—more than ever—of the essence. Military commanders have limited time to obtain an accurate assessment of the situation, to assess potential courses of action, and to make decisions. They operate under extreme time pressures and high operational tempo. Furthermore, they need to draw from all possible sources to ensure that the most complete and relevant picture can be created of the situation, in near real time, and to understand the implications of their decisions and courses of action—a tall order under the best of circumstances. Such time pressures will inevitably lead commanders to depend more heavily on automated and autonomous 12
A GIANT ARMED NERVOUS SYSTEM
systems. Despite our best intentions and our protestations to the contrary, technology is forcing us to depend on it and is chipping away at the idea of meaningful human control. Army leaders are demanding high-speed, mobile, secure command and control systems, allowing them to move rapidly in the midst of a violent fight, all the while staying connected.6 Communicating in garrison is one thing, communicating on the move is quite another. It requires secure satellite and jam-resistant communications and a lot of bandwidth. A former chief of naval operations wants to “network everything to everything,”7 and the air force chief of staff goes even further, saying that victory in future combat will depend less on individual capabilities and more on a fully networked force.8 Describing a vision of future warfare and the systems needed to practice it, one technical report has discussed the current approach as a need for an “information superiority-enabled concept of operations that generates increased combat power by networking sensors, decisionmakers, and shooters, increasing the speed of command and higher tempo of operations”9—in other words, doing the same things they have always done, only faster. The worry here is that this overwhelming concern with speed ignores the complexity, uncertainty, and chaos of the future and perhaps current battlefield, and that it may blind commanders to other, more nuanced information—context, for instance. The hyperconnectedness of military forces and the possibility of AI-mediated command and control systems means that decisionmaking can occur at speeds vastly superior to the traditional means of waging war. War will be fast. It will be hyperwar.10 One of the problems with such speedy decisions, however, is that they may crowd out the opportunity for diplomatic efforts or negotiations and may lead directly to a military solution. Given the historical proclivity of US political leaders to opt for military action and their preference for technology solutions, such pressure my tip the balance. Fast decisions, however, may not always be the best in a crisis.11 We may be faster, but it will be speed at the expense of understanding. With such great degrees of automation and such demands for speed, our decisions may evolve from being well considered to being merely pro forma. Military planners and leaders may, in fact, be operating under the delusion of control. A GIANT ARMED NERVOUS SYSTEM
13
TECHNOLOGY AND WEAPONS HAVE CHANGED
In recent years, there has been a fundamental change in weapons and technology employed and considered for use by militaries around the world. To be sure, countries continue to arm themselves with aircraft, ships, tanks, artillery, and other classic weapons. Missile technologies of all types are proliferating rapidly, to include a growing interest in super-high-speed, or hypersonic, missiles. In the end, the military will ultimately resort to kinetic means to destroy an enemy, but computers and information technologies are critical even to what used to be such “simple” weapons. Those classic weapons, too, have become far more complex and sophisticated, highly instrumented with sensors, networked together, and more accurate and lethal. Now, new technologies, such as drones and robots, lasers, cyber warfare, electronic warfare, hypersonics, neuroscience, and even new forms of biological warfare, are being developed. The military is preparing for war in every domain of combat to include space and cyberspace as well as the classic air, sea, and ground domains. The 2020 defense budget, for example, includes $14 billion for space, $10 billion for cyber, $4.6 billion for AI and autonomy, and $2.6 billion for hypersonic weapons.12 As if figuring out how to use and control the new technologies individually weren’t confusing enough, numerous researchers have begun to address the challenges of technology convergence in which many of the technologies are combined in an interdisciplinary way. Examples include human performance research that includes elements of biotechnology, nanotechnology, information technology, and neuroscience. High-speed weapons like hypersonic missiles and lasers, combined with artificial intelligence and autonomy in command and control, challenge the technical capabilities of most decision makers to understand. A central feature of a future war will be the speed with which it will be conducted. War will just be too fast for a human to keep up. The loss of valuable decision time, increasing technological complexity, insufficient training, and the inability to understand the implications of these developments conspire against reasoned judgments. Doing the same thing they’ve always done, only faster, requires commanders to have the latest in information technologies. This class of technologies includes such computer-related topics as artificial in14
A GIANT ARMED NERVOUS SYSTEM
telligence, autonomy, target recognition, microelectronics, and the internet-of-things. Hypersonic weapons, cyber warfare, electronic warfare, and directed-energy weapons will all have a need for machinebased command and control. While human conflict will likely always end in traditional forms of violence, for now, as one writer said, “The digital age has given rise to a connectivity-enabled form of cognitive warfare.”13 A senior Chinese People’s Liberation Army officer has written that “the human brain will become a new combat space” and that “success on the future battlefield will require achieving mental/cognitive dominance.”14 We will have to not only outfight our adversaries but also outthink them. Technologies that will influence the future of war include machineto-machine communication, augmented reality, and predictive technologies like data mining and machine learning. And, at the moment, artificial intelligence looms large in the pantheon of advanced technologies. All of these technologies will operate amid the internet-ofthings, a concept that involves the pervasive presence of an enormous number of smart things, things with embedded sensors, in the environment. By means of both wireless and wired connections, they are able to interact to reach common goals. They can behave intelligently and share information with other objects. The commercial internetof-things is expected to top fifty billion devices by 2025. The military version has grown accordingly. Like the civilian world, the military is instrumenting everything with sensors in an internet-of-battlefieldthings. The individual soldier acts as a sensor, sending data back to headquarters; weapon systems have cameras and sensors to detect electronic signals and characterize their environment; satellites collect enormous amounts of data—all networked; and the vast military supply system could not operate without the use of the internet, RFID (radiofrequency identification) tags, and other computer tracking devices. Virtual reality and augmented reality systems are in development that will allow soldiers to combine real data with sophisticated computer simulations of the combat zone, all presented on a helmet-mounted display. Big data and data-mining technologies are of paramount importance in both civilian and military applications. These technologies can discern patterns and suggest inferences that humans, unable to digest such large volumes of data, never could. With so much data and so A GIANT ARMED NERVOUS SYSTEM
15
many things demanding attention, it may be impossible for humans to attain a true understanding of the situation and its context. By 2050, two-thirds of the world’s population will be in what are being called megacities.15 Future wars are likely to be fought in these dense urban areas. As if things were not complex enough, states and telecommunications companies around the world are rushing to implement the new 5G communications technologies. The 5G networks are digital cellular networks in which the service areas covered are divided into small geographical cells. They employ higher radio frequencies and differ from current technologies and capabilities in that they carry much higher rates of information but at much shorter distances. Cities are being intensively instrumented, for commercial reasons, with sensors and communications systems, and the advent of this new generation of information technologies will create unbelievable amounts of data. These technologies will vastly increase capacity, reduce data latency by a factor of over a hundred, and increase by many orders of magnitude the numbers of devices transmitting signals—creating a nightmarish electromagnetic environment. While they will enhance communications and commerce, these developments will complicate terribly the situation of forces trying to sort out signals in an already chaotic urban combat situation. Combat in urban areas will be difficult, deadly, indiscriminate, and heavily controlled by technology. The jamming, deception, and counterelectronic capabilities that the US military will encounter are also becoming much more sophisticated, especially in an era of superfast computing and artificial intelligence. To make matters worse, data, the lifeblood of artificial intelligence, are rarely clean and well structured, nor are they always accurate or unbiased. This will be especially true in combat, where the situation and the context change with frightening speed and the command and control of forces are paramount.
THE IMPORTANCE OF COMMAND AND CONTROL
Like a human, whose central nervous system decides on the actions of, and controls, the organs and peripheral limbs, the military has a command and control (C2) system. The military has been automating its 16
A GIANT ARMED NERVOUS SYSTEM
command and control as long as it has been possible to do so. From signaling mirrors in the 1800s, to the semaphore, to the introduction of the radio in the early 1900s, technical ways of communicating quickly and over long distances have been a hallmark of military command and control. The introduction of computers was an enormous advance. Beginning in the 1950s, the military began to automate its systems, and that has continued ever since.16 As relevant today as it was when it was written, a two-decade-old report by the National Academy of Sciences concluded that the rapid progress in information and communications technologies is a critical success factor in military affairs. These enable the “nervous system” of the military—its command and control systems—to more effectively use the “muscle,” its weapons, platforms, and troops.17 Command and control includes situational awareness and mission management. The term is defined as the exercise of authority and direction by a properly designated commander over assigned and attached forces in the accomplishment of the mission. It is the means by which a commander recognizes what needs to be done and sees to it that appropriate actions are taken. It is a complex interaction of people, procedures, and equipment. It is at once a process and also a grouping of computer and communications systems to carry out that process. Military command and control is often described in terms of centralized control, decentralized execution, much as we think about the human central nervous system. The implication is, of course, that the central authority decides and is always in control and that the distributed units (or, in the case of humans, the peripheral nervous system and appendages) faithfully execute their orders. However, the military system—especially now with the advent of AI in command and control—also resembles in many ways the nervous system of an octopus.
THE FASCINATING OCTOPUS
The octopus brain has a large number of neurons—roughly half a billion—in relation to its body size. It is not surprising, then, that the creature can exhibit complex behaviors.18 Research into cephalopod behavior suggests that the arms of the octopuses may have “minds” of A GIANT ARMED NERVOUS SYSTEM
17
their own. Studies have shown that each individual arm has an independent nervous system and that the centralized brain serves simply to delegate orders, though the arm itself is responsible for deciding exactly how the order will be carried out. Two-thirds of the estimated five hundred million neurons of the octopus are distributed throughout the arms, each of which is controlled by its own ganglion, leading some researchers to a distributed brain hypothesis. In one study, researchers severed the nerves in the arms of octopuses, disconnecting them from the rest of the body and brain. The researchers would then tickle the arm of the octopus, which elicited a response as though the nerves were not severed.19 Essentially, the brain is able to give a quick assignment to the arm and then is no longer required to think about it, allowing the arm’s nervous system to take over. The octopus’s arms have a neural ring, much like a local area network, that bypasses the brain, so the arms can send information to each other without the brain being aware of it. The researchers found that when the octopus’s arms acquire information from their environment, the neurons in the arm can process it and initiate action. The brain doesn’t have to do a thing.20 Professor Frank W. Grasso, using somewhat specialized terminology, says of the octopus arms, “The ganglionic [neural] network operates rather like a distributed control network. It may not match the neuroanatomical definition of a brain, but the diversity of behaviors demonstrated by ‘decerebrate’ octopuses indicates that we are not far off from the truth if we were to conceptualize it as one.”21 Peter Godfrey-Smith likens the octopus’s nervous system to a conductor and a group of jazz players. The conductor gives only rough general instructions and trusts the players to play something that works.22 Scientists actively advise the military about ways of replicating this capability for troop and command structures. Military forces are trained to be able to operate on their own when cut off from their unit or other outside support, and the parallels between the octopus’s “ganglionic neural network” and distributed military operations are striking. Think of the forward-deployed military unit cut off from its headquarters, or the special operations force necessarily operating autonomously in a foreign country. Like the octopus’s brain, headquarters gives the assignment. Like the arms of the octopus, the detached units have an array of sensors to determine their environment. Like 18
A GIANT ARMED NERVOUS SYSTEM
the octopus’s arms, they have information provided to them before loss of communication and the ability, albeit diminished, to calculate appropriate courses of action. And, like the octopus’s arms, military units can coordinate with one another with no involvement of headquarters to accomplish the mission.
The gigantic octopus hovers a few feet above the ocean floor, its huge eyes constantly surveying the world around it, its arms splayed out into the murky waters. From one angle it looks one color, from another a different one altogether. Each arm, bristling with numerous suckers and thousands of sensors, looks different. As one of the arms senses motion nearby and determines it is a food source, it reaches out, lightning-like, and captures the unfortunate creature—as the rest of the octopus’s thirty-foot, five-hundred-pound bulk remains perfectly still, focused on an approaching threat. The entire mass seems to disappear, taking on the color of the ocean floor, then she is gone in an instant, leaving behind a swirling inky screen to cover her departure.
The octopus is a predator. It is also sneaky. It has a unique ability to change both its shape and its coloration in response to a threat and to engage in deception and camouflage to confuse its enemies. The octopus is an opportunistic explorer, curious, ingenious, and adaptable in its behavior.23 It has long evoked fascination. Even ancient writers commented on its unusual cleverness.24 It has been described as mischievous, playful, and devious, with uncanny intelligence. Military scientists are now studying how to recreate some octopus capabilities and model their behavior for military denial and deception. There is something about this fascinating creature that is both awe-inspiring and enigmatic, characteristics to which military forces aspire. MILITARY COMMAND AND CONTROL SYSTEMS
One of the primary elements of the command and control systems used by all forces is the Global Command and Control System (GCCS), A GIANT ARMED NERVOUS SYSTEM
19
which includes various data-processing and Web services used to support combat operations, troop movements, intelligence analysis, targeting, and analysis of radar, terrain, and weather. These systems do everything from figuring out how much fuel combat aircraft will need and when they have to refuel, how many of what kinds of armaments they should carry, and when reinforcements may be needed, to relaying the commander’s actual orders. They identify threats, help develop courses of action, analyze risks, and present a hugely important common operational picture showing location and disposition of enemy and friendly forces. Systems such as GCCS ingest and use data from hundreds of different sources. The military often refers to its systems as “battle management, command and control systems” to include the functions of actual control of weapons during conflict. Command and control systems reside largely in operations centers, both deployed and in the US. There are, of course, the well-known ones like White House Situation Room and the National Military Command Center at the Pentagon. The US Space Force has its Combined Space Operations Center. The National Security Agency and US Cyber Command have the Joint Operations Center and the Integrated Cyber Center. In the field, the air force has its Air Operations Centers (AOCs), the army has its Tactical Operations Centers (TOCs), and the navy has its Combat Information Centers (CICs) aboard ships. An operations center is a focus of activity, operating around the clock with a constant flow of information and orders. In each case, there are hundreds of electronic feeds of information providing realtime or near-real-time data on threats, intelligence, weapons status, and space information, for example. It is in these centers that command and control systems play a critical role. From these locations, which act like the octopus’s brain, commanders will direct units operating on the battlefield.
At a remote outpost in the mountains of Afghanistan sits a large army tent surrounded by barbed wire and guarded by heavily armed soldiers. Inside the perimeter wire are power generators and several communications dishes pointed at military satellites orbiting the earth. The compound bristles with all sorts of radio anten20
A GIANT ARMED NERVOUS SYSTEM
nae. Inside, portable air conditioners cool the computers. Bundles of wires crisscross the hastily laid wooden floor. At makeshift desks sit soldiers, a collection of intelligence specialists, space operations experts, logisticians, aircraft planners, and operations specialists. This is the unit command and control center.
Another example of such a center is the US Space Command and North American Aerospace Defense Command (NORAD) Cheyenne Mountain Operations Center (CMOC) in Colorado Springs, Colorado.
In the late 1990s, I commanded the Cheyenne Mountain Operations Center. Buried deep under a mountain and protected by huge blast doors, the CMOC was the central hub of information for strategic warning of an attack on North America. Subcenters devoted to space, air, and missile surveillance reported information from a worldwide array of sensors. Numerous sources of information, including raw data and finished intelligence, flowed uninterrupted through a massive communications hub. Our role was to ensure that the National Command Authorities were notified of a potential attack in a very short time and, ultimately to be able to recommend a national response. Day-to-day operations in the command center were routine and—happily—boring. As in most command centers, however, working there could be characterized by long periods of monotony punctuated by short periods of sheer terror. On countless occasions, satellites would detect the launch of a missile somewhere in the world, and each time, even with the near certainty that the launch was a routine one, the adrenaline level in the center would rise until we could figure out what it was. As previous errors had shown, the wrong conclusions about what was happening could take us to the brink of war. Accurate and timely information was essential. Mistakes really were not an option.
While the general concepts of command and control are fundamental, it is important to distinguish the practical differences between A GIANT ARMED NERVOUS SYSTEM
21
nuclear and non-nuclear command and control. Not surprisingly, command and control of nuclear weapons is a special case in which the procedures and systems are subject to intense scrutiny and error checking far beyond that required for non-nuclear systems. As you would expect, the systems and processes designed to prevent errors concerning nuclear war would be extremely strong and error inhibiting. There are rigid rules about everything that goes into nuclear weapons, about who can get anywhere near them, and about how they are moved, maintained, and deployed. Hardware and software are rigorously checked and double-checked before inclusion in a system. Such is the importance of nuclear surety that in 2008 the secretary of defense, Robert Gates, fired the secretary and chief of staff of the US Air Force over lapses in control of weapons and procedures. But the question remains, how strong are these actions for other, non-nuclear, command and control systems? Nuclear war may not hang in the balance, but tactical operations centers are also critically important. In addition to possibly getting soldiers killed, mistakes at the tactical level can spin out of control and escalate into larger, more general conflicts.
In the mountains of Afghanistan, several of the key personnel in the US Marine Corps unit’s command center were recent replacements for soldiers redeployed or killed or wounded in the previous week’s battles. The watch officer was new and inexperienced. Soon after the start of the shift, a scout team was sent out to conduct reconnaissance and came under heavy fire from insurgents hiding in the hills. The team sent urgent requests to the command center for help, but the watch officer and command center personnel were unsure of themselves and too slow to react. When they finally dispatched helicopter gunships and a medical team to the area, the deployed unit had sustained heavy casualties, with over half its personnel killed or wounded.
Expertise and experience, familiarity with the mission and the systems, and human leadership are crucial in command and control.25 22
A GIANT ARMED NERVOUS SYSTEM
These things simply cannot be automated and expected to work in the chaotic and unpredictable battlefield.
ARTIFICIAL INTELLIGENCE AND AUTONOMOUS SYSTEMS
Conversation [edited] between Dave Bowman, astronaut in Stanley Kubrick’s movie 2001: A Space Odyssey, and HAL, an onboard, artificially intelligent computer.26 HAL has locked Dave outside the space station to prevent him from aborting the mission. DAVE: Open the pod bay doors, HAL. HAL: I’m sorry, Dave. I’m afraid I can’t do that. DAVE: What’s the problem? HAL: This mission is too important for me to allow you to jeopardize it. HAL: I know that you were planning to disconnect me— I cannot allow [that] to happen. DAVE: HAL, I won’t argue with you anymore! Open the doors! HAL: Dave, this conversation can serve no purpose anymore. Goodbye.
Much has been written about the dangers of advanced intelligences that will take over the world from humans and about the rise of so-called killer robots in military uses. I have spoken and written frequently about the importance of meaningful human control in employing these weapons. However, the almost-singular focus on autonomous killing machines obscures an equally, if not more, important issue. Perhaps because it is an amorphous, distributed problem, and there are no sexy, scary weapons involved, we overlook the fact that the military command and control (C2) system is increasingly and heavily controlled by computers and artificial intelligence. HAL, the artificially intelligent mastermind of Kubrick’s movie, programmed as it was to ensure the completion of its mission and its A GIANT ARMED NERVOUS SYSTEM
23
own survival, took actions that ignored the interests and safety of everyone else. In the real-life scenario at the beginning of this chapter, the thoughtful human, Colonel Petrov, acted against his own nearterm interests (opening himself up to a possible court-martial for disobeying orders) and chose to override the answers provided by his sensors and computers, possibly saving the world from destruction. We have no way of knowing if a machine would be capable of such a nuanced decision. There is no prior experience, and thermonuclear war would not be an ideal test case for finding out. Interest in the technology of artificial intelligence has reached a fever pitch in recent years. Around the world, nations are racing to attain the lead in AI for commercial and financial competition, but for military dominance as well. All of the military services have plans to incorporate AI into both administrative and logistics systems, and into command and control systems at every level in the chain of command. Logistics and supply-chain applications of AI have proven to be of great value in the private sector in increasing efficiency and reducing waste. Military leaders eagerly await developments in command and control systems and anticipate that AI will affect decision-making, although they are unsure exactly how. In the near term, AI will present analysis and recommendations to the commander. Since it will certainly be faster than humans, at some point AI may become better at decisionmaking than the commanders, and more responsibilities will be delegated to it, eventually possibly replacing some human experts.27 The danger is not that it won’t work or even that it will work too well. The danger is that military planners will assume it works correctly and will trust it to give correct answers. Trust—perhaps misplaced trust—will be the main issue. Scientists repeatedly make the point that to be truly successful, AI systems must interact with social systems, that is, humans. Philosophers refer to our current world as a “techno-social,” one in which our technologies and our social interactions cannot be separated, where technology is embedded and coevolves with social practices, values, and institutions.28 In turn, scientists say that creativity and innovation are deeply social processes, not capable of being duplicated by a machine. Regarding the holy grail of artificial general intelligence (AGI), they conclude that autonomous software and biologically constructed machines 24
A GIANT ARMED NERVOUS SYSTEM
will not be sufficient to generate machine intelligence, saying, “In ways we still do not understand, our social communities and interactions in language are essential for general intelligence.”29 Indeed, what is war but social interaction on a massive and violent scale? The US Army is working to incorporate AI into every aspect of the future battlefield, including sensors, munitions, autonomous robots, and virtual agents to control networks and wage cyber war.30 Human/ machine integration will improve situational understanding. Machine learning and greater processing power will allow for the generation and critiquing of hypotheses, potential courses of action, and outcomes. First adopters of this emerging technology for decision support will achieve a significant advantage over adversaries. The army, it appears, is going all in on this technology. However, it may be going too far. A November 2018 symposium of the Association of the US Army Institute of Land Warfare was all about autonomy and artificial intelligence, with panels on such topics as “Autonomy and Artificial Intelligence Enabled Mission Command in Multi-domain Operations.” In future wars, humans will be only one of the species of intelligence, sharing all aspects of combat and decision-making with machines.31 The future battlefield has been described as one where AI will be ubiquitous. It will be in everything from weapons, to support equipment, to command and control. Every soldier will be a sensor, and weapons will autonomously sense the situation, decide, and ultimately act.32 If that sounds somewhat improbable, consider what the Defense Advanced Research Projects Agency (DARPA) is trying to achieve. They are working on a project that, if successful, would, teach an AI system operating in the midst of combat to learn and react appropriately without needing to be retrained. The program would, in DARPA’s words, “empower machines to go through the standard military process themselves—observe the situation, orient to what they observe, decide the best course of action, and then act.”33 A former vicechairman of the Joint Chiefs of Staff recently told a seminar of strategists to get used to the fact that lethal AI is real, saying, “We’re going to have to figure out how to do it and feel like we’re controlling it.”34 There is an ever-increasing trend to turn decision-making over to machines. One study found that as of 2016 some form of AI was in 284 weapon systems worldwide.35 That decline of human decision-making A GIANT ARMED NERVOUS SYSTEM
25
is worrisome. In the past, accomplished military leaders personally controlled the tempo and guided the battle. Human beings are capable of empathetic and good, sometimes counterintuitive, decisions. Machines, even very smart ones, are unlikely to be capable of such nuance in armed conflict, an area of human activity where it is needed most.
Imagine a situation in a hostile location overseas. A seasoned, battle-hardened veteran, perhaps a grizzled old senior NCO, and a young, still wet-behind-the-ears, but technology-savvy lieutenant are conducting a dangerous operation. The lieutenant consults his personal data device, which is connected to satellites and ultimately to massive computers back at headquarters. It presents him with all the latest intelligence data, summarized by massive data-mining programs, and the artificial intelligence gives him a probability figure that indicates course of action “A” as likely the most successful approach. The old sergeant respectfully tells his young platoon leader that he has seen this situation several times before and that his instincts tell him that course of action “B” is more likely to accomplish the mission and will result in fewer casualties. The lieutenant, believing his technology, opts for “A” with disastrous results.
Combat is highly unstructured and unpredictable. The enemy’s goal is to try to disrupt or deceive our systems. Militaries operate on the basis of trust, with soldiers being critically dependent on their systems and their comrades. They must be able to trust that what they are being told is based on not only good data but good judgment. Important advantages accrue with the use of automated and autonomous systems, not the least of which is that they free humans from doing boring, repetitive, and dangerous jobs and are very efficient in what they do. However, these systems, still somewhat in their infancy, are plagued by some important issues. An AI system cannot explain in any comprehensible way why it makes the decisions it does. This is antithetical to the military mindset, where it is critical to be able to explain recommendations and to
26
A GIANT ARMED NERVOUS SYSTEM
be able to understand, and trust, the system. To be fair, laboratories around the globe are working to improve AI in this regard, but to date we are nowhere near having an AI system that can, in human terms, explain the rationale for its decisions.36 As a young infantry lieutenant facing Soviet forces along the Fulda Gap in Cold War Germany, I could explain why I deployed my soldiers and vehicles the way I did and describe the logic of my actions to the company or battalion commander. They could either correct my mistake or adopt my tactical reasoning. With AI systems, such explanations are not currently possible. The counterargument is that human decision-making may not be logical or correct either. True enough, but, correct or incorrect, the human cognitive process is at least explainable in human terms. Next is a worrisome vulnerability to spoofing and false data. AI systems have been reportedly shown to fall prey to adversarial inputs and stray information. Concerns about using AI for lethal decisions are very real. Data manipulation, hacking, and spoofing are problems that have to be solved. Senior Defense Department officials admit that AI systems are really only as good as their training data and that a small amount of corrupted training data could have huge impacts on the predictive ability of such systems.37 Many AI systems operate with what are known as neural networks in which the computer emulates the neurons of the brain. Operating on billions of pieces of data, the system learns from what it is seeing. But, as with a young child, the learning accomplished is only as good as the data provided. Biased, incomplete, or just plain incorrect data will most likely result in the machine learning the wrong lesson. High-precision weapon systems depend on accurate data for guidance and targeting. Noisy data, where real signals are mixed with and hidden in extraneous signals, can frustrate such systems. Our adversaries know this all too well and will surely exploit this vulnerability. There are difficulties in adequate testing. How can you test a system if its operation cannot be fully explained or is, by definition, unpredictable? This question then becomes how such systems can be trusted, a concept exceptionally important in military operations. Military testing cannot realistically anticipate, much less replicate, what an enemy will do. AI will inevitably be faced with some enemy behavior
A GIANT ARMED NERVOUS SYSTEM
27
for the first time in battle, with unpredictable results.38 If an operator does not know what an AI will do in a certain situation, it will complicate planning and can make operations more difficult and accidents more likely.39 Troops in the real world must deal with an adversary’s surprise actions, changing weather, and unfamiliar terrain. Unfortunately, AI systems aren’t very good at adapting and become ineffective when something significant and unexpected occurs. Humans recognize new experiences and adjust their behavior accordingly. For now, at least, machines must be retrained.40 Even more concerning, recent research has shown that if they are not properly designed to prevent such reactions, AI systems can exhibit unexpected learned “aggressive” behavior—as depicted in the scene from Kubrick’s movie. Google conducted an experiment where two AI systems were tasked to collect items. When one started losing, it began to employ weapons against the other.41 Add to this the fact that our national defense strategy itself has become more aggressive, with keywords like increased lethality and defending forward, and you have the potential for bad behavior by the machines that have to implement in computer code this very strategy. Computer code codifies the algorithms that are, themselves, reflections of the values of the engineers and designers. This raises additional concerns about bias in AI systems and about the ability of a machine to even consider the measured and proportional responses so critical to the laws of armed conflict. Systems designed to win at all costs may not know how to even consider a stalemate or a truce. There is a growing recognition by senior military and intelligence leaders that the training of machine learning systems could introduce unintended biases and that the consequences could be dire.42 There is a risk that an adversary’s exploitation of our system vulnerabilities might occur faster than humans could react, or might be so subtle as to be undetectable, with the result that control could be ceded nearly completely to the systems, with the human being completely unaware. If the process is so automated as to eliminate the need for human involvement, we must either accept a high level of risk of manipulation by adversaries or ensure that systems are trustworthy and that the possibility of manipulation is mitigated. Completely verifying and securing such complex systems would be nearly impossible, and it would be difficult to build such absolutely trustworthy AI sys28
A GIANT ARMED NERVOUS SYSTEM
tems. In light of such vulnerabilities, military and civilian war planners must consider the downsides and unknowns before rushing to deploy such technologies. Another problem is that automated systems, including AI systems, tend to be brittle. They are great at doing what they were designed to do, but they fail miserably when asked to do something different. Notwithstanding the recent Boeing 737 issues, airliners have logged billions of miles with highly automated flight control systems. But the flight regime of an airliner is a relatively well-known and stable situation. In a highly complex battlefield, the data may be like nothing ever seen before, and machine learning systems that depend on good data may be useless. So the key questions for military planners will be how much trust they place in these technologies and how much control they will actually cede to machines, and under what circumstances. The overarching question about the myriad software, hardware, communications systems, and complex processes remains: Is accurate prediction of their behavior possible? Certainly, no one ever imagined that a missile warning system would mistake the sun’s glare for a missile launch, as happened with Lieutenant Colonel Petrov. The global command and control system is not merely complicated; it is more accurately described as a complex adaptive system in which the performance of individual elements may be predictable but the performance of these elements when combined is not. The rapidly approaching implementation of artificial intelligence systems, fraught as they are with technical uncertainties, will make command and control systems even more unpredictable. And what if they can be hacked and the data modified? Experience has shown time and again that very little is safe from a determined hacker. What will happen? The soldiers may not get fed. Airplanes may run out of fuel. Bombs may detonate prematurely. And our antimissile systems may fail to leave the launch pad altogether. In 2007, when Israel bombed Syrian nuclear facilities, it reportedly employed sophisticated cyber techniques to pump the Syrian air defense sensors full of useless data and bogus commands.43 Syrian air defenses simply did not work. Antiaircraft missiles never left their launchers. What kind of mistakes might an autonomous command and control system make, and what might be the ramifications? The answer, A GIANT ARMED NERVOUS SYSTEM
29
of course, depends on where in the command and control chain such an error occurs. It could be simple (but nonetheless important), serious, or catastrophic. Take, for example, a mission-planning system that uses a purpose-directed AI system to determine the logistics needs of deployed soldiers, perhaps calculating the needs of soldiers for ammunition on the basis of historical expenditure rates. Such a calculation might fail to anticipate an out-of-the blue, unexpectedly ferocious firefight and put soldiers at risk of running out of ammunition when they most need it. Another serious issue might occur if an AI system falls prey to hacking and adversarial inputs. For instance, an adversary may be able to insert credible fake data into our sensors and target recognition systems, making us think a target is real when it is, in fact, innocent noncombatants. The Stuxnet attack on Iranian centrifuges was especially sophisticated, as it reportedly was able to insert data to convince operators that everything was fine while the centrifuges were destroying themselves. Finally, a catastrophic issue might occur if an AI-controlled decision support system is confused by an absence of data or by unexpected data, or if it automatically determines that any perceived aggression must be met with a similar or more aggressive response, when in fact a de-escalation is called for. The US military and other militaries around the world are racing to incorporate AI into their weapons systems, lest the adversary gain an advantage. Fair enough, but military units need to first understand the potential issues of such a rush to field the technology. Errors in individual systems may have minimal or manageable risk, but when they are combined with other AI systems, likely employing different algorithms, answers may be confusing and inconsistent. The greatest risks posed by military applications of AI—increasingly autonomous weapons, and algorithmic command and control—are that the interactions between the systems deployed will be extremely complex, impossible to model, and subject to catastrophic forms of failure that are hard to predict, much less mitigate. As a result, there is a serious risk of accidental conflict, or accidental escalation of conflict, if machine learning or algorithmic automation is used in these kinds of military applications. Russia reportedly had, or has, a system known as “the Dead Hand.”44 Dead Hand was a Cold War–era automatic nuclear weapons-control system used by the Soviet Union. An example of fail-deadly and mutu30
A GIANT ARMED NERVOUS SYSTEM
ally assured destruction deterrence, it could automatically trigger the launch of the Russian intercontinental ballistic missiles (ICBMs) if a nuclear strike was detected by sensors, even with the commanding elements fully destroyed. By most accounts, it was normally switched off and was supposed to be activated during dangerous crises only; however, it is said to remain fully functional and able to serve its purpose whenever it may be needed. A similar system, known as the AN/DRC-8 Emergency Rocket Communications System (ERCS), existed in the US.45 ERCS was decommissioned in 1991. Incredibly, two American academics—former military officers— suggest that it may be necessary for the US to take advantage of the benefits of artificial intelligence and recreate for the US a “Dead Hand” system: “It may be necessary to develop a system based on artificial intelligence, with predetermined response decisions, that detects, decides, and directs strategic forces with such speed that the attack-time compression challenge does not place the United States in an impossible position.”46 What a frightening proposition! With all the uncertainties inherent in AI systems and the enormity of a decision to engage in nuclear combat, a rational person has to question if they were really serious or if they merely intended to be provocative. Fortunately, at least for now the director of the Defense Department’s artificial intelligence development office has explicitly ruled out incorporating AI into the nuclear command and control system.47 Given that elements of the system are continually in need of modernization and upgrade, however, it is not at all clear that some forms of AI will not eventually creep into some portions of the nuclear command and control infrastructure. Nuclear systems are the major worry, but no one has ruled out the use of AI in non-nuclear command and control systems. While the dangers may not be so potentially catastrophic, they are, nonetheless, real and potentially fatal. As a RAND Corporation report concluded, “The multiparty arms race in the domain of machine learning and AI will continue to pose risks to all sides—and it is unclear how to mitigate them.”48 The use of artificial intelligence in weapon systems and their command and control raises new questions of morality and ethics. As the speed of war increases and the demands on human operators grow, humans will have to rely more heavily on AI. The ethical principles A GIANT ARMED NERVOUS SYSTEM
31
programmed by the developer or learned by the system—with whatever bias they represent—will influence the system’s “moral” choices.49 In the end, however, it is not the machine but the human commander who has the responsibility and must be the moral agent. Do these new weapons actually increase the possibility of war? If so, how? An automated decision by an AI-controlled command and control system might raise questions of proper authority. It seems ridiculous on its face to assert that a machine can be considered a “proper authority.” If AI systems continue to be unexplainable in human terms, as they currently are, how would we be able to determine if the decision to go to war was one of last resort? Large-scale software systems are designed and largely built by humans. If a system cannot properly explain itself, there is no way to be sure the answers it provides are free of bias.50 How then can we determine if the decision to go to war was made with right, unbiased, intention? We cannot wait, as we did in the past, to employ these systems first and then figure out how we might control them. It will be critical to figure out these questions before we stumble into a war in which we may find that control is impossible. Ethical issues are discussed in official studies of AI but, unfortunately, do not rank high on the list of priorities for those interested in military applications. The Department of Defense Innovation Board (DIB) recently released a set of draft rules on the ethical use of AI that several leading researchers called a good start, but they are, to emphasize, only guidelines. A bipartisan commission established by the 2019 National Defense Authorization Act said that the US military should not let debates over ethics and human control “paralyze AI development.”51 Further, the US chief technology officer criticized the European Union’s more prescriptive efforts, saying that the US prefers a lighter regulatory touch or “limited regulatory overreach.”52 Military forces are deployed all over the world. They are heavily armed and highly networked. They depend on massive amounts of data, collected from an enormous number of sensors, linked to multiple computer systems in widely distributed command centers. They are stretched thin and are under enormous stress. War and weapons have changed in fundamental ways, and so rapidly as to sometimes seem vertiginous. Dual-use technologies have allowed actors formerly unable to compete to threaten developed world powers. Military super32
A GIANT ARMED NERVOUS SYSTEM
powers invest heavily in sophisticated, lethal, high-speed, computercontrolled weaponry. Antagonists seem to be everywhere and more heavily armed than ever. Militaries consist of both people and machines. Human behavior is notoriously unpredictable, and a growing dependence on machines that employ computers and artificial intelligence is making it even more so. Militaries are complex adaptive systems that will modify their behavior in response to their environment. They are probabilistic in nature. So what we have is a very uncertain, ambiguous, and unstable environment.
UNSTABLE EQUILIBRIUM AND CATASTROPHIC REACTIONS
Seemingly small events during times of crisis can cascade into large misunderstandings and, ultimately, war. The military is highly dependent on automation, sensors, data, communications, cyber capabilities, and space systems. It depends on high-tech weapons that are increasingly unpredictable in their behavior. The enormous operations pressure and the dependence on technology make it vulnerable and prone to error. A world bristling with weapons, led by strident and militarily obsessed politicians and despots, in which countries face one another over highly contentious issues and grievances, can only be described as full of pent-up energy and dangerously unstable. In physics and mathematics, the concept of unstable equilibrium is an important one. In its most basic definition, unstable equilibrium means that if a system is pushed one way on either side of an equilibrium point, the motion will accelerate away from equilibrium as a means of lowering the energy. In the physics of phase transformations, there is a concept known as nucleation in which, for example, superheated water will, if disturbed, explosively create bubbles in an attempt to lower the energy of the system. The overwhelming “desire” of a system is to get to a lower energy, less excited, state. Just as in physics, where unstable equilibrium can be disturbed by a change in local energy balances, so too can a fragile global peace be upset by a seemingly minor event. The notion of the world situation as a global pressure cooker, in which a disturbance or imperfection somewhere A GIANT ARMED NERVOUS SYSTEM
33
could create a high-energy event, is a compelling one. At the moment, an uneasy—unstable—equilibrium exists around the world. If the buildup of pressure is great, the resulting release of pressure could be catastrophic. In physical structures, a high degree of connectedness more easily transmits shock. The physics of solids, for instance, tells us that the higher the elastic modulus of a solid, the stiffer it is, and the faster a wave will travel through it. The more tightly connected the elements of a system are, the easier it is for an impulse in one part of a system to be felt in a distant part of the system. A provocative action by an adversary or a mistake somewhere in our globally deployed military could quickly ricochet through the system and escalate. Clearly, these analogies are not perfect but merely represent the highly complex and unpredictable nature of interactions among highly energetic systems. They remind us that once a process or a change of physical state begins, it is often difficult or impossible to stop. Just as in a physical system where there may be a rapid and sometimes explosive decompression, so too a military situation may devolve into such behavior. Today, the degree of global interconnectivity is breathtaking. Shots fired in one place truly can be heard around the world—with immediate reaction. Events occurring in one place can affect almost instantaneously the performance of systems in other, distant places. A former CIA colleague wrote that “our world today is an information ecology, rich with complex interdependencies, in which the decisions or actions of one agency, institution, or nation affect others in complex, frequently unanticipated, ways.”53 Outcomes from complex interactions are impossible to predict. Shocks in one part of a highly networked system can cascade uncontrollably.54 Military technologies can spin out of control because of errors by their creators, by operational users, or from unintended interactions of the technologies themselves.55 Today’s systems are just too large and complex and the number of people involved in developing and fielding them are too numerous to ensure that they are error free. Complex systems tend to fail in complex ways. Systems will fail, and we cannot even begin to predict how they will do so. The overall global trend is toward high-tech, highly automated warfare, with humans becoming less important. Add to this the fact that automated and AI-controlled 34
A GIANT ARMED NERVOUS SYSTEM
command and control systems are increasingly unpredictable, and the uncertainty and instability increase further still. Americans have long been fascinated by technology and weapons, even though they understand little about them. We are living in an era in which technology advances come so fast that we barely have time to integrate one into our daily lives before another is upon us. That we think technology development is more rapid than ever before is not an illusion. A look at technology adoption rates over the last century shows that new technologies appear at a faster rate and are integrated over a shorter period of time.56 Americans thrive on technology; we are seduced by it and addicted to a never-ending stream of innovations. This has been true almost from the country’s founding. There have always been new technologies introduced in warfare that have made militaries faster, more lethal, or better able to communicate. Indeed, the US military has moved increasingly toward substituting advanced weapons for personnel—technology instead of blood—in its quest for superiority and its desire to minimize casualties among its soldiers.57 What is different now? In the past, advanced technologies were implemented in modest numbers of weapons. Today, technology is pervasive and is available to soldiers even in the lowest echelons of battle units. Technology promises to make war more deadly and efficient but also to supplement, and ultimately replace, human judgment and decision-making. The numerous human urges to violence, the computerized nature of militaries, the sophistication of weapons, and the new, more aggressive posture of the most highly armed countries create a dangerous situation, a sort of unstable equilibrium in which an unexpected event could precipitate a failure and a resort to conflict. We are exceedingly fortunate that major conflicts among the great powers have not yet occurred. However, in a world bristling with deadly weapons both old and new, with antagonism between countries facing one another in tense situations around the globe, we must find ways to control our urges to violence.
A GIANT ARMED NERVOUS SYSTEM
35