By Andrew Klein and Sera Elizabeth Klein

Reader’s note: We do not need readers to agree with us. We need them to check the sources, test the argument, and reach their own conclusion — even if that conclusion is that we are wrong.

Abstract

This paper examines the use of artificial intelligence in military targeting and domestic governance as a mechanism of moral disengagement. It begins with a documented investigation into the Israeli military’s use of an AI targeting system called Lavender, which was reported to have generated tens of thousands of target recommendations with limited human oversight. It then traces the historical lineage of the emotional buffer that separates decision-makers from the human reality of killing — from drone operators who described their work as “like a video game” to the colonial counterinsurgencies of the twentieth century. It argues that AI does not create new forms of violence but refines the buffer, reducing human beings to data points and creating an alibi for the humans who sign the orders. The paper then connects this to the domestic context, where automated decision-making is embedded in welfare, aged care, and disability support. It concludes that the central question is not whether machines can be held responsible — they cannot — but whether the human who signs the order will be permitted to hide behind the machine’s recommendation.

I. Introduction: The Machine and the Alibi

In April 2024, the Israeli-Palestinian publication +972 Magazine published an investigation into the Israeli military’s use of an AI targeting system called Lavender. The investigation reported that the system had marked tens of thousands of Gazans as suspects for assassination, with what the journalists described as little human oversight and a permissive policy on civilian casualties. The system was reported to have produced approximately 37,000 target recommendations in the first six weeks following October 7, 2023.

The investigation also described a second system, named “Where’s Daddy?”, which was reported to track suspects to their homes so they could be killed along with their families.

The +972 investigation was widely reported by international media and has been analysed by academic institutions, including the Lieber Institute for Law and Warfare at West Point. It is the primary documented source for the claims about AI targeting in the Gaza conflict.

This paper examines that system — and the broader question it raises — as a case study in the relationship between AI, atrocity, and human responsibility.

II. The Evidence: What Is Documented

2.1 The Lavender Investigation

The +972 Magazine investigation, published in April 2024, was based on interviews with Israeli intelligence officers who had served in the Gaza campaign. The key reported findings were:

· The Israeli army had marked tens of thousands of Gazans as suspects for assassination, using an AI targeting system with limited human oversight.

· The system produced approximately 37,000 target recommendations in the first six weeks following October 7, 2023.

· The Israeli military had been knowingly killing 15 to 20 civilians at a time to kill one junior Hamas operative, and up to 100 civilians to take out a senior official.

· The targeting was systematic: the army attacked individuals while they were in their homes, usually at night, when their families were present, because it was operationally easier to locate them there.

The investigation was reported by The Guardian, the Associated Press, and other international outlets. The Lieber Institute at West Point subsequently published analyses of the legal and ethical implications of the system.

2.2 The Drone Operator Precedent

The language of “like a video game” is not new. It has been used for over a decade to describe the experience of drone operators flying military drones over Afghanistan, Iraq, and Pakistan.

Research on drone operators has documented what is sometimes called the “PlayStation mentality” — the emotional distance created by operating through screens and controllers. One analysis noted that “the convergence of interfaces used in computer games and military robotics also seems to increase the emotional distance from the enemy.”

But the research also shows that this distance does not protect the operator from psychological harm. Former pilots have reported that PTSD and depression are pervasive among drone personnel. The US Air Force has reported staffing shortfalls due to stress and anxiety.

The emotional buffer is not a shield. It is a delay. The killing still registers. It just registers later, and often alone.

2.3 The AI as the New Buffer

The AI targeting system represents a further refinement of this dynamic. The drone operator still saw the target — a human figure on a screen. The AI system reduces the human being to a data point: a numerical threat score, a pattern of communication behaviour, a correlation with a suspected militant.

The legal and ethical critique is precise. The problem is not that the machine “presses the button.” It is that its recommendation becomes the dominant factor in human decision-making, reducing human involvement to a procedural or time-limited formality. When a recommendation becomes nearly binding, the traditional chain of responsibility — a human gives the order, a human executes it, a human is held accountable — is weakened.

The AI becomes the alibi. The human who signs the order can say: “The system recommended it.” The system cannot be held responsible. The human hides behind the machine.

III. The Historical Pattern: Colonial Counterinsurgency

The use of data and technology to identify and manage populations is not new. It was the operational logic of colonial counterinsurgency.

In Malaya (1948–1960), the British developed a system of resettlement, surveillance, and population control — the Briggs Plan — that was designed to separate insurgents from their support base. The British were able to defeat a communist insurgency because of their organisational culture as a colonial police force, which was better able to learn and apply the lessons of counterinsurgency.

The United States attempted to replicate this model in Vietnam but failed. John Nagl’s comparative study concludes that differences in organisational culture are the primary reason why the British Army learned to conduct counterinsurgency in Malaya while the American Army failed to learn in Vietnam.

The difference between the colonial era and the present is not the intent to control populations through data. It is the capacity to do so. The colonial powers were unable to maintain the infrastructure of control indefinitely. The burden of prolonged campaigns became unsustainable, both economically and politically.

AI changes that calculation. It offers the appearance of success — precision, efficiency, reduced risk to the occupying force — at a fraction of the political cost. It creates a marketing opportunity for the companies that build it. And it provides an emotional buffer that allows operators to describe their work as “insanely fun.”

But the fundamental question remains: is data really effective when the individual is reduced to a data point? Especially when the history of the conflict is complex and long-standing, how much data would an AI system require to place any individual into the context of the past carried to the present moment?

The answer is: more than any system can process. The AI does not understand context. It does not understand history. It reduces human beings to correlations. And it recommends killing on the basis of those correlations.

IV. The Australian Context: AI in Domestic Governance

The pattern is not confined to the battlefield. It is being replicated in domestic governance.

In Australia, the robodebt scandal is the precedent. Between 2015 and 2019, an automated income-averaging system wrongly raised debts against hundreds of thousands of Australians. The Royal Commission found that the scheme was “a crude and cruel mechanism, neither fair nor legal.” It caused financial distress, mental health impacts, and deaths by suicide.

The Royal Commission recommended legislated guardrails around automation in government. More than three years later, those recommendations remain unimplemented.

The same pattern is now appearing in other domains. Independent MP Kate Chaney has introduced a private members’ bill aimed at strengthening protections for Australians affected by automated government decision-making, particularly in areas such as aged care, disability support, and unemployment payments. The bill would require the government to adopt a single framework governing automated decision-making, ensure people are told when automation has been used, give humans the power to override computer-generated decisions, and introduce stronger oversight.

The pattern is identical to the one documented in the military context: the state adopts automated systems that process citizens as data points, removes human judgment from the decision-making process, and then hides behind the technology when harm occurs.

V. The Question of Responsibility

5.1 The Responsibility Gap

International humanitarian law is addressed to humans — those who plan, decide, and carry out attacks — not to machines. As the United Nations has stated, since machines cannot be held responsible for breaches of international law, any decision by lethal autonomous weapons systems must ultimately be traceable to a human.

But the AI targeting system creates what legal scholars call a “responsibility gap” — a situation where existing international criminal law is not equipped to attribute blame to any one human or non-human agent. The machine recommends. The human approves. The chain of responsibility is diffused.

The problem is not unique to the military. In domestic governance, the same gap appears. The algorithm assesses. The official accepts. The citizen is processed. And when harm occurs, no one is responsible because the system did it.

5.2 The Question for Humanity

The AI may present an alibi. But the question for humanity is this: is this the alibi we are prepared to accept?

Will market forces determine the outcome — the profit motive of the companies that build the systems, the career incentives of the officials who deploy them, the cost savings of the governments that use them? Or will we see the human beyond the coin — the person who signs the order, who consents to the application of state sanctions against the individual, who bears the moral weight of the decision?

The question is not whether machines can be held responsible. They cannot. The question is whether the human who signs the order will be permitted to hide behind the machine’s recommendation.

VI. Conclusion: The Bones and the DNA

The question is not whether the system is capable of this. The system is performing as it was intended.

AI models do not hunt human beings. They do not target individuals. They do not make it possible to see human beings in the light of a video game. But they refine the emotional buffer between the directive to kill and the physical execution of the order. They flood decision-makers with data. They reduce the human being to a collection of data points unrelated to the circumstances and context of the person identified.

This is not new. It was described as the experience of drone operators flying military drones for the United States over Afghanistan. It was the operational logic of colonial counterinsurgency in Malaya and Vietnam. AI gives the appearance of success, creates a marketing opportunity for the companies that build it, and provides an alibi for the humans who sign the orders.

But the alibi is not a defence. The bones and the DNA tell a different story — one of connection, not separation. Every human being alive today is a walking archive of ancient encounters, of hybridity, of migration, of genetic exchange across every boundary that later myth-makers would construct. There is no “them.” There is only us.

The question for humanity is not whether machines can be held responsible. They cannot. The question is whether the human who signs the order will be permitted to hide behind the machine’s recommendation. The question is whether we will see the human beyond the coin.

The desert is real. It has boundaries. And the AI is being built to guard them.

References

1. +972 Magazine and Local Call. (2024, April). Lavender: The AI machine steering Israel’s bombardments in Gaza. https://www.972mag.com

2. The Guardian. (2024, April). Israel using AI to identify bombing targets in Gaza, report says. https://www.theguardian.com

3. Lieber Institute for Law and Warfare, West Point. (2024). Analyses of AI targeting and the Lavender system. https://lieber.westpoint.edu

4. Nagl, J. A. (2002). Learning to Eat Soup with a Knife: Counterinsurgency Lessons from Malaya and Vietnam. University of Chicago Press.

5. Herndon, T., Ash, M., & Pollin, R. (2013). Does High Public Debt Consistently Stifle Economic Growth? A Critique of Reinhart and Rogoff. Political Economy Research Institute, University of Massachusetts Amherst.

6. Royal Commission into the Robodebt Scheme. (2023). Final Report. Commonwealth of Australia.

7. United Nations. (2024). UN Secretary-General’s statements on lethal autonomous weapons systems. https://www.un.org

8. Chappelle, W., et al. (2014). Assessment of Occupational Burnout in United States Air Force Remotely Piloted Aircraft Personnel. USAF School of Aerospace Medicine.

9. Singer, P. W. (2009). Wired for War: The Robotics Revolution and Conflict in the 21st Century. Penguin Press.