Killer Robots

Armin Krishnan’s Killer Robots: Legality and Ethicality of Autonomous Weapons arrived in 2009 from Ashgate and was reissued by Routledge under Taylor & Francis in 2016, in both editions one of the first book-length scholarly treatments of the question that has since come to dominate emerging-tech arms control debates: what happens when the decision to kill is delegated to a machine. Krishnan, a political scientist who has worked on private military contractors and the changing face of warfare, was writing in a moment when the Predator and Reaper were already operational over Pakistan and Afghanistan, the US Army was sinking billions into the Future Combat Systems programme, and the SWORDS armed ground robot had just made its first appearance in Iraq. Robotics and autonomy were no longer hypothetical, but the legal and ethical literature had barely caught up. Other works appeared the same year — P. W. Singer’s journalistic Wired for War, Wendell Wallach and Colin Allen’s philosophical Moral Machines, Ronald Arkin’s engineering manifesto Governing Lethal Behavior in Autonomous Robots — but Krishnan’s was the one that aimed squarely at the international-law machinery and Just War tradition, and asked whether existing rules could be stretched to cover machines that select and engage targets without a human in the loop.

The book’s central claim is that autonomous weapons systems are not merely faster, more accurate cruise missiles. They represent a categorical break with the long arc of weapons development because they shift the locus of the engagement decision from a human operator to an algorithm. Krishnan argues this shift creates serious problems under International Humanitarian Law — particularly with the principles of distinction and proportionality — and under the moral framework of Just War theory, which presupposes a human moral agent making judgements about killing. It also opens what later writers would call an accountability gap: when a robot kills the wrong person, the chain of responsibility between programmer, commander, manufacturer, and operator becomes diffuse to the point of vanishing. Krishnan does not call for an outright ban in the manner of the activist campaigns that emerged later. His position is more measured: the international community should treat these systems as a category requiring new law, weapons reviews under Article 36 of Additional Protocol I should be taken seriously, and certain applications — particularly fully autonomous anti-personnel systems — should probably be prohibited before they are normalised.

Structurally the book moves from definitions through history and then into the legal and ethical core. Early chapters work hard on terminology because in 2009 the field had not settled on what “autonomous” meant. Krishnan distinguishes between remote-controlled, semi-autonomous, supervised-autonomous, and fully autonomous systems, and notes that most contemporary systems sit in the middle of that spectrum. He then traces the prehistory of the autonomous weapon back further than most accounts: not just to the Predator, but to the German V-1 flying bomb and V-2 ballistic missile of the Second World War, to acoustic torpedoes, to the heat-seeking and radar-homing missiles of the Cold War, and to the close-in weapons systems already installed on warships for decades. The Phalanx CIWS, the Dutch Goalkeeper, the Aegis combat system in its automatic modes — these, Krishnan points out, are autonomous in the operationally relevant sense. The point is not that nothing is new but that the line between the familiar and the alarming is harder to draw than the debate often assumes. From there the book surveys the early-twenty-first-century landscape: the Predator and Reaper and their armed strikes, the SWORDS and TALON ground robots, the iRobot PackBot for explosive ordnance disposal, the Israeli Harpy loitering munition designed to circle a region and dive on enemy radar emissions, and the South Korean SGR-A1 sentry robot installed along the demilitarised zone with a contested degree of human control.

Middle chapters take up the legal arguments in detail. Krishnan walks through the four cardinal principles of IHL — distinction, proportionality, military necessity, and humanity — and tests each against the capabilities a fully autonomous weapon would require. Distinction asks whether a system can reliably tell a combatant from a civilian, a wounded soldier from a fighting one, a person hors de combat from one still in the engagement. Proportionality is harder still, because it demands a contextual weighing of anticipated military advantage against expected civilian harm — the kind of judgement that depends on framing, intent, and politics. Krishnan is sceptical that any near-term machine perception system can perform these tasks at anything like the level a human commander, however imperfect, brings to them. He grants Arkin’s argument that machines might one day outperform stressed and tired humans in narrow tasks, but argues this is a long way off in unstructured environments and that the burden of proof should sit with those proposing to field such systems, not with those urging caution. He gives close attention to Article 36 of Additional Protocol I, the obligation on states to review new weapons for compliance with IHL — a provision he treats as the most immediate practical lever, even though many states either ignore it or conduct reviews in secret. He also reviews the historical record of weapons that have been restricted or banned outright — the 1868 St. Petersburg Declaration on explosive bullets, the 1925 Geneva Protocol on chemical weapons, the 1995 CCW Protocol IV on blinding laser weapons, the 1997 Ottawa Treaty on anti-personnel landmines — and asks whether the autonomous weapons question fits any of those templates.

The ethical chapters bring in Just War theory, asking whether autonomous weapons can satisfy jus in bello criteria when no human is making the engagement decision in real time. Krishnan engages with Isaac Asimov’s three laws of robotics only to set them aside as inadequate — they were a literary device, not a workable doctrine — and turns instead to Arkin’s proposal for an ethical governor, a software module that would constrain a robot’s behaviour according to encoded rules of engagement and IHL. Krishnan treats this seriously but is unpersuaded. The encoding problem is harder than its proponents acknowledge, the test environments are nothing like real combat, and the accountability problem remains even if the engineering works: a machine cannot be deterred, prosecuted, or punished. He spends time on the responsibility question — who is liable when an autonomous system commits a war crime — and on the related worry that delegating the decision may lower the political threshold for resort to force, because the costs of war become more abstract when no soldier is sent to bleed for it. The dual-use problem, the proliferation problem, and the prospect of asymmetric use by non-state actors all enter the analysis.

The book also looks ahead — to swarming, to the convergence of robotics with miniaturisation and artificial intelligence, to what later writers would call algorithmic warfare. Krishnan is cautious about prediction but treats the trajectory as a serious matter of policy rather than science fiction. He discusses the US Future Combat Systems programme, by then troubled and on the way to its 2009 cancellation, the Patriot missile fratricide incidents in 2003 when the system shot down a British Tornado and a US Navy F/A-18 in semi-autonomous mode, and the 1988 destruction of Iran Air Flight 655 by the USS Vincennes’ Aegis system — cases in which automation contributed to errors with serious consequences. These are not, in his telling, arguments against any automation. They are arguments for taking the human role seriously and not letting the seductive promise of clean, mechanised war crowd out the institutional and legal commitments that have been built up over more than a century. Closing chapters turn to arms control prospects, where Krishnan canvasses the possibilities — a treaty under the Convention on Certain Conventional Weapons, a stand-alone instrument, a code of conduct, national legislation — and assesses each. He is realistic about the obstacles. Major military powers, the United States and China above all, are unlikely to forgo a technology they see as decisive, and verification of autonomy is harder than verification of, say, chemical weapons stockpiles because the relevant element is software. But he sees the early years of a technology as the window in which constraints have the best chance of taking hold. The chemical weapons taboo, the landmine ban, the cluster munitions convention — none of these emerged after the weapons in question had become entrenched military assets. They emerged when the conversation was still in its formative phase.

Since publication the field has moved in the direction Krishnan anticipated. Human Rights Watch published its Losing Humanity report in 2012, the Campaign to Stop Killer Robots launched in 2013, and the UN’s Convention on Certain Conventional Weapons opened informal meetings of experts on lethal autonomous weapons systems the following year, eventually formalising the process as a Group of Governmental Experts. The US Department of Defense issued Directive 3000.09 in 2012, requiring senior-level review for autonomy in weapon systems and articulating an “appropriate levels of human judgement” standard. Christof Heyns’ 2013 report as UN Special Rapporteur on extrajudicial killings called for a moratorium on the development of lethal autonomous robots and drew on the kind of legal analysis Krishnan had laid out. Later books built on the foundation: Paul Scharre’s Army of None in 2018 became the field’s most-cited synthesis, more accessible and more current but covering much of the same conceptual ground. Krishnan’s reception in this conversation has been respectful — he is cited as one of the first to take the question seriously — though some critics have argued he overstated the technical readiness of fully autonomous systems in 2009 and understated the speed at which machine learning would change the engineering picture. Others, particularly in the Arkin camp, have pushed back on the broad scepticism toward machine judgement, contending that the human alternative is not as reliable as the critique implies.

For a reader building a library on this topic the book sits at the early-formative end of the shelf and is best read alongside Singer’s Wired for War to get the journalistic and the legal-ethical takes side by side. Wallach and Allen’s Moral Machines complements it on the philosophical side, Arkin’s Governing Lethal Behavior provides the engineer’s counter-position, and Scharre’s Army of None brings the story forward to the late 2010s. What it does not cover, because the technology had not yet arrived in its current form, is the deep-learning revolution in computer vision and the rise of large-scale autonomy in cyber and electronic warfare; readers wanting that should pair it with more recent work. What it does cover, and what has aged well, is the legal architecture — the IHL principles, the Article 36 review obligation, the Just War framework — because that architecture has not changed and remains the reference frame against which any new weapon will be judged. The specific systems mentioned have dated, as they must; the Future Combat Systems programme was cancelled the same year the book first appeared, and the Predator has been retired from US service. But the questions Krishnan put on the table are the questions still being argued at the CCW in Geneva more than a decade later, and his framing of them — careful, lawyerly, sceptical of techno-optimism without dismissing the technology — has held up.