SWITCHED ON
The daily technology series nobody asked for but everyone needed
Kill Decision: The Autonomous Weapons Reckoning
Should a machine be permitted to decide who dies? The debate is happening right now. Most people have no idea.
The laws of war encode a principle: that lethal force is a human act, carried out by human beings who bear moral responsibility for it. Autonomous weapons don't just challenge that principle logistically. They abolish it structurally.
In Episode 7 we established that the algorithm running your social media feed is optimising for your engagement rather than your enlightenment — with consequences for your politics, your mental health, and your grip on shared reality. If that left you feeling mildly surveilled and vaguely manipulated, good. That's accurate. Hold that feeling, because today we're going somewhere that makes algorithmic manipulation look like a minor administrative inconvenience.
Today we're talking about autonomous weapons. Machines that can identify, target, and kill without a human being making the final decision. AI systems that can prosecute lethal force independently, at machine speed, across battlefields or city streets, based on criteria set by their programmers and validated by their training data.
This is not science fiction. This technology exists, is being developed at pace by multiple major powers simultaneously, has already been used in active conflict, and is the subject of an international debate that is simultaneously one of the most important conversations humanity is having and one of the least visible to the general public. The question at the centre of that debate is one of the most serious questions that can be asked: should a machine be permitted to decide who dies?
01 — What We're Actually Talking About
There is a spectrum of autonomy in weapons systems, and where you sit on that spectrum matters enormously for the ethical and legal analysis. At one end: human-in-the-loop systems — weapons that require explicit human authorisation for each individual engagement. A soldier pulling a trigger. A human being, accountable and present, making each lethal decision.
In the middle: human-on-the-loop systems — automated weapons that can engage targets independently but with a human overseer who can intervene. The Phalanx close-in weapon system, which automatically shoots down incoming missiles, works like this. The speed of modern threats means the human "on the loop" often has neither the time nor the information to meaningfully override the system. The nominal human control is real but practically limited.
At the other end: fully autonomous systems — weapons that can identify, select, and engage targets without any human involvement at any stage. No human authorises the individual engagement. No human can intervene in real time. The machine decides.
It is the last category that is most contested, most dangerous, and most actively being pursued by the countries with the largest defence budgets. Because speed, scale, and cost advantages of fully autonomous systems are, from a purely military standpoint, compelling. The military logic is, from within its own frame of reference, coherent. The ethical logic is considerably more troubled.
02 — The Accountability Vacuum
Here is the core problem with autonomous lethal systems, stated as plainly as possible: if a machine kills someone who should not have been killed, who is responsible? Under existing frameworks of international humanitarian law, individual human beings bear legal and moral responsibility for decisions to use lethal force. The framework depends, at every level, on there being a human being who made a decision and can be held accountable for it.
Autonomous weapons rupture that framework. If an autonomous system misidentifies a civilian as a combatant and kills them, who answers for that? The programmer? The military commander who deployed the system? The manufacturer? The government that procured it? The answer, under current law, is genuinely unclear — which is another way of saying that nobody answers for it.
This accountability vacuum is not a technical problem. It is a fundamental structural consequence of removing human decision-making from the kill chain. And it cannot be resolved by better programming or more rigorous testing.
03 — What's Already Happening in the Field
The debate about autonomous weapons is sometimes framed as if it concerns a future technology that hasn't yet arrived. This framing is wrong, and it is convenient for the parties developing these systems. In 2020, a United Nations panel of experts report described what appeared to be the first use of a fully autonomous lethal drone in combat — a Kargu-2 loitering munition deployed in Libya, reportedly used to autonomously attack and pursue retreating soldiers without requiring a data link to a human operator. The manufacturer disputes the characterisation. The technology, however, is real and deployed.
Israel's Harpy drone — an autonomous anti-radar weapon that loiters over a target area and independently attacks radar emissions — has been in service since the 1990s. Drone swarm technology is under active development by the United States, China, Russia, and several other military powers. The US Air Force has conducted tests of autonomous dogfighting AI that defeated human pilots in simulated combat.
The technology is not arriving. It has arrived. The question is what, if anything, we are going to do about it.
04 — The International Response: Well-Intentioned, Largely Ineffective
The campaign to ban autonomous weapons has generated significant discussion, multiple rounds of talks at the Convention on Certain Conventional Weapons in Geneva, and a series of resolutions that stop well short of binding prohibition. The reason is straightforward: the countries most capable of developing autonomous weapons are the countries whose buy-in would be necessary for any meaningful international agreement. The United States, China, Russia, Israel, and the UK are all actively developing autonomous weapons capabilities. None of them has agreed to a binding prohibition.
The United States' official position is that it supports "appropriate levels of human judgment" in lethal decisions — a formulation so carefully vague as to accommodate almost any degree of automation. China has stated its support for banning the use of autonomous weapons while actively developing the capability to deploy them. The window for establishing binding international norms before autonomous weapons become a standard feature of modern warfare continues to close.
05 — The Question We Keep Avoiding
There is a version of this debate that stays within the technical and legal frame — accuracy rates, accountability frameworks, international law, treaty negotiations. That conversation is important and is not being had loudly enough. But underneath it is a deeper question that we tend to avoid in polite policy discourse, because it doesn't have a comfortable answer.
What does it mean, morally, for human beings to delegate the decision to kill another human being to a machine? Not to assist that decision. Not to support it. To delegate it entirely — to remove human judgment, human conscience, and human accountability from the most consequential act one person can perform with respect to another.
The laws of war encode a principle: that lethal force is a human act, carried out by human beings who bear moral responsibility for it. Autonomous weapons don't just challenge that principle logistically. They abolish it structurally. The machine doesn't feel the weight of the decision. The machine doesn't carry the burden of having chosen. The machine simply executes. History suggests, fairly consistently, that making it easier and cheaper to kill people produces more killing of people. Autonomous weapons are designed to make it easier. That is not a neutral development.
Tomorrow we're coming back down to earth with something that affects every single person reading this in their daily life: quantum computing and what it means for the encryption protecting your bank account, your messages, and your medical records. The short version is that everything you currently consider secure is going to need to be rebuilt. The longer version is considerably more interesting. See you then.
Switched On is a daily technology series covering AI, social media, data privacy, and the digital forces reshaping modern life — with no corporate spin, no false comfort, and absolutely no mercy for buzzwords.
