Geonho Roh, Zhiyuan Jin, Praveer Tiwari
On February 28 2026, two Tomahawk missiles struck an elementary school in Minab in the Hormozgan province of southern Iran, resulting in more than 150 casualties, including at least 123 children, as reported by CNN. According to a Bloomberg investigation, a key detail about this airstrike is that Palantir’s AI systems were involved in the command’s decision-making process. In the first 24 hours after the war began, the U.S. military command had to strike nearly 1,000 targets in Iran. To manage the increasing volume of information and intense time pressure, commanders relied on AI target-recommendation algorithms, the report states. During this process, additional human target-validation procedures were reduced or skipped. As a result, outdated information about civilian facilities was not adequately filtered out during the targeting process. According to Bloomberg, this allowed outdated location data to be used in identifying attack coordinates, contributing to the deadly strike that claimed innocent lives. We want to raise the concern: Is AI-optimized warfare trading off more and more human lives?
As students of computer science and mathematics with one of us having military experience, we are interested in the use of AI to “optimize” combat decision-making. From a military perspective, we understand that the time spent confirming a target can decide whether friendly forces live or die. From the perspectives of computer science and mathematics, we also know the significance of optimizing a system for a single objective—speed. Because constraints on time, including human verification, may that look like unnecessary delays, they are usually the first things they system tries to remove.
From this perspective then, by adopting Palantir’s AI systems, the U.S. military effectively achieved the key strategic goal of extremely shortening decision-making time. The successful strikes early in the war against senior officials, including the commander-in-chief of Iran’s Islamic Revolutionary Guard Corp were possible because the existing, slow human command decision-making process was largely replaced by AI algorithms, which determines targets even before the opponent gets any time to prepare. AI systems may offer strategic advantages in modern warfare. They can process targets dozens of times faster than the opponent and respond to attacks with less time and personnel.
The increasing use of AI-assisted military command and control is not limited to the United States. South Korea has been pursuing AI integration into its Army Tactical Command Information System (ATCIS), while the Israeli military has used AI systems such as Lavender and Gospel to support target identification and prioritization. NATO and its member states have also been exploring ways to integrate data from multiple battlefield sensors into faster decision-making processes.
In this increasing integration of AI in military systems globally, the speed of military decision-making is valued. In May 2025, the director of the U.S. National Geospatial-Intelligence Agency reported that during an exercise, Palantir’s Maven Smart System reduced targeting workflow timelines from hours to minutes. This suggests that AI-assisted systems can accelerate parts of the targeting process, although faster information processing does not necessarily ensure that every strike is approved or carried out faster.
What has to be kept in mind, however, is that behind the military achievements brought about by maximizing speed, are mathematical and systemic limitations. When a system is optimized using “decision-making speed” as its sole objective function, existing multi-layered human verification procedures (Human-in-the-Loop) are treated as the factor that is causing the risk of delay by hindering speed, making it a factor that can potentially be reduced or bypassed. We can imagine a situation where commanders become overconfident in the validity scores and recommendations provided by algorithms (a phenomenon known as automation bias). What if an algorithm calculates target coordinates quickly, but becuause it uses underlying input data that contains outdated information about civilian facilities, the system ends up producing unreliable recommendations (Bartenstein and Karra)?
Data integrity is, thus, very important; though, on an actual battlefield, we cannot always guarantee the reliability of data, and in many cases, the data may even be incorrect. The 2021 Kabul drone strike disaster during the war in Afghanistan is another example of the dangers of targeting decisions based on mistaken intelligence. In that incident, a U.S. drone strike killed ten civilians, and the Department of Defense later acknowledged that it was a tragic mistake. Such cases illustrate why faster targeting must be accompanied by careful verification. Otherwise, the capability to strike in the shortest possible time may also become the capability to cause civilian casualties in the shortest possible time.
In the current state of military systems, it is not likely that AI will be abandoned in the search for increasing speed in the decision making process, much less likely is the return to older systems of manual verification, given how AI has now intertwined with human life and decision-making processes. Faster responses increase the possibility of survival for soldiers in the same way that a few minutes can have serious consequences for civilian lives. The importance of timely decision-making is also reflected in the U.S. Marine Corps’ doctrine on command and control.
We have so far shown how military systems seek out AI for faster decision-making. Returning to the Minab school incident, what do we make of this search for speed when civilian lives are at stake? This is not merely hypothetical. History shows how the demand for operational speed has pushed the boundaries of what is considered acceptable. A 2017 Guardian report examined a rise in reported civilian deaths from U.S.-led airstrikes, relating the rise to changes in procedure to achieve quicker decision-making. What is different today is the role of AI in the link between the speed of military decision making and the consequences for civilians. How should the risk for civilian lives draw the line in the increasing integration AI in to military decision-making?
Course Instructor: Dr. Ainehi Edoro
Associate Professor, Dept. of English
email: aedoro@wisc.edu
Editor: Shrinjita Biswas
PhD Candidate, Interdisciplinary Theatre Studies
Works Cited:
Abraham, Yuval. “‘The Machine Did It Coldly’: Israel Used AI to Identify 37,000 Hamas Targets.” The Guardian, 3 Apr. 2024.
Bartenstein, Ben, and Krishna Karra. “Inside the US ‘Kill Chain’ That Destroyed an Iranian School.” Bloomberg, 18 Sept. 2026.
Borger, Julian. “Civilian Deaths from US-Led Strikes on Isis Surge under Trump Administration.” The Guardian, 6 June 2017.
CNN. “Photos Appear to Show US Tomahawk Missile Fragments at Site of Deadly Iran School Strike.” 10 Mar. 2026.
Defense Acquisition Program Administration (DAPA). South Korea’s ATCIS Modernization.
Lopez, C. Todd. “DoD: August 29 Strike in Kabul, ‘Tragic Mistake,’ Kills 10 Civilians.” U.S. Department of Defense, 17 Sept. 2021.
National Geospatial-Intelligence Agency. “VADM Whitworth Delivers Keynote at 2025 GEOINT Symposium.” 21 May 2025.
NATO Communications and Information Agency. “NATO Acquires AI-Enabled Warfighting System.” 14 Apr. 2025.
U.S. Marine Corps. Command and Control Theory. Marine Corps Doctrinal Publication 6.
Image: Max Ernst. “Garden Airplane Trap.” 1935.
Image source: The Space Between: Literature and Culture 1914-1945 (Vol 14, 2018).
