Minab, Maven and the Hard Questions Around AI and Sentience

Minab: The Civilian Cost Behind a Targeting Failure

Did you know that an elementary school in Minab was struck on February 28, the first day of the Iran war, and that the reported death toll remains disputed? The Associated Press counted about 120 children among the dead; Iranian state media put the total at 168, while AFP reported at least 155 people killed, most of them children. Reporters have had limited access, so those figures should not be treated as a settled final count. The school stood beside a Revolutionary Guard compound, but the two sites had been visibly separated since 2016. The Congressional Research Service reported that a Tomahawk hit the school and that a preliminary U.S. military investigation found the United States responsible, working from outdated targeting information. Secretary of State Marco Rubio said the United States “would not deliberately target a school.” The Pentagon had still not published its findings months later, leaving families and the public without a complete official account.

The human detail is devastating precisely because it is ordinary: teachers called parents to collect their children before Iran’s government ordered schools closed that morning, and some parents did not arrive in time. That detail changes the scale of the story from a dispute over a target list to a specific failure involving a school, a morning routine, and children with no role in decisions about the war. The article does not establish that an AI system selected the Minab target. It reports that a Tomahawk struck and that an investigation found outdated targeting information; it separately describes the Pentagon’s use of Maven across the wider campaign. Keeping those claims distinct matters. Connecting a particular strike to AI without evidence would obscure rather than clarify accountability. In a smart-city context, the lesson is not that all automated systems cause harm, but that decisions affecting civilian places require traceable information, scrutiny, and a clear route for investigation when something goes wrong.

The reported investigation also highlights how the quality and age of targeting information can matter as much as a weapon’s technical accuracy. A Tomahawk is named in the reporting, but the available account does not identify the exact targeting workflow for the Minab strike, the software used to generate its coordinates, or the decision-makers who approved it. Nor does it publish the full military findings. Those gaps limit what can responsibly be said about the system’s mechanics. What is known is that the school had been visibly separate from the adjacent compound since 2016, while the preliminary U.S. investigation reportedly identified outdated information as a factor. An accurate weapon directed using incorrect or stale information can still hit the wrong place. That is a general operational implication, not a claim about an undisclosed algorithm. The distinction matters for any technology-enabled institution: people need to know what information a system relied on, when it was verified, and who had authority to challenge the result.

Minab’s consequences extend beyond the immediate deaths and the investigation. In September, the UN fact-finding mission on Iran found reasonable grounds to believe that U.S. forces committed war crimes in two February strikes, including Minab. The mission said the school was clearly civilian, that the intelligence behind the strike was not adequately verified, and that the attack went beyond negligence. The U.S. government had not accepted that finding; President Trump said in June that “nobody” purposely attacked the school, while the Pentagon’s findings remained unpublished. These statements and conclusions have different origins and statuses, so they should not be collapsed into a single definitive account. For people building or governing systems that help direct action, the practical question is whether responsibility can be reconstructed after an error: what information was available, how reliable it was, and whether human review could intervene. That question is urgent in any city where software influences decisions about people and places.

Maven, Speed, and the Escalation Trap

The conflict article describes a second technology story: the Pentagon relied on an AI targeting platform called Maven to strike more than 13,000 targets during the first 38 days of the war. Michael Klare, writing in Arms Control Today, said Maven’s speed in selecting and reselecting targets gave U.S. forces a combat advantage by helping them disable Iranian capabilities swiftly and continuously, preventing their reconstitution. This is the operational rationale reported for the platform—not a published technical specification. The account does not disclose Maven’s model architecture, training data, confidence thresholds, sensor inputs, or the precise division of work between software and human decision-makers. It therefore cannot support claims about particular algorithms or prove that Maven was involved in the Minab strike. What it does establish is that a platform was used at considerable scale and that speed was regarded by officials as a central advantage. When a system can help process targets quickly and repeatedly, the questions about review, verification, and responsibility become more consequential, not less.

The reporting places that acceleration alongside a major change in civilian-protection staffing. Politico reported in March that Pentagon personnel focused on reducing civilian casualties had fallen from about 200 to fewer than 40. One former official told the AP that the Minab strike was a natural result of those cuts and of Secretary of Defense Pete Hegseth’s emphasis on lethality. That is an attributed assessment, not an official finding establishing a direct causal chain. Still, the numbers describe a striking imbalance: more than 13,000 targets in the first 38 days, while a civilian-protection workforce reportedly shrank to fewer than one-fifth of its earlier size. The article does not explain how those staff were distributed or what review procedures remained. It does invite scrutiny of whether the speed of target selection was matched by enough capacity to identify civilian risks, check information, and respond to warning signs. A high-throughput process is not automatically a well-governed process.

Robert Pape, a University of Chicago political scientist who founded the Chicago Project on Security and Threats, frames the wider danger as an “escalation trap.” He argues that moral certainty on each side can help normalize increasingly aggressive attacks on civilians, as each side treats the other’s civilians as deserving harm. Pape’s research on suicide terrorism, described in the article through his books Dying to Win and Cutting the Fuse, identifies foreign occupation as a leading driver. He warns that anger left by civilian deaths can have future consequences, saying on PBS’s Amanpour & Company that “even with smart bombs, there’s going to be collateral damage of a serious nature.” His point is not that one incident proves a later attack was caused by it. The article explicitly says the Austin gunman and the Temple Israel attacker do not establish that the war caused their violence, and reported motives remained uncertain. The risk he describes is a potential pathway, not a confirmed explanation for those events.

The implications reach beyond military command systems. Cities depend on decisions about infrastructure, transport, public safety, and emergency response being made with reliable data and clear accountability. The comparison is not that urban services are equivalent to wartime targeting; they are not. It is that speed and automation can make consequential decisions feel operationally routine, even when people may bear the cost of bad information. The article reports no details of Maven’s architecture, and responsible analysis should not fill that gap with speculation. It does provide evidence of scale, tempo, and organizational context: more than 13,000 targets in 38 days, a reported decline in civilian-casualty staff, and an unresolved Minab investigation. Together, those facts make a useful governance test for any high-impact system: can decision-makers show what it used, who checked it, what happened when information was outdated, and how harms are investigated? In the next decade, public trust will depend as much on answers to those questions as on technical performance.

Civilian Harm, Accountability, and Multiple Truths

The article’s central warning is not to settle for a single villain when the evidence describes harms and alleged crimes on more than one side. On September 17, the UN fact-finding mission found reasonable grounds to believe that U.S. forces committed war crimes in two February strikes, including Minab. The same mission found that Iran’s crackdown on protesters since late December amounted to crimes against humanity. Iran’s official toll was 3,038 people dead, a figure the mission believed understated the real number. These are separate findings about separate conduct, and acknowledging both is not the same as claiming the acts are equivalent. The article argues that multiple truths must be held together because selective attention can reinforce the moral certainty that fuels escalation. That approach is particularly relevant to reporting on technology: a discussion of system failure must not erase other victims, and condemnation of one government’s abuses must not be used to dismiss evidence about another’s actions.

The proposed responses focus on public accountability as much as technical capability. The article calls for news programs to ask who is most likely to be hurt next, and what might prevent it, before centering the question of who is right. It also calls for civilian-protection investigators, humanitarian workers, and mediators to be heard alongside political partisans. For AI-assisted targeting, the specific proposal is an independent review and public explanation after credible evidence of grave harm. That would make an investigation more than an internal process whose findings never reach the people affected. The source does not specify a particular review board, legal standard, or technical audit method, so those should not be assumed. Its concrete demand is for independent scrutiny and public explanation. This kind of transparency can help distinguish a software contribution from decisions made by officials, while preserving the need to investigate the entire chain rather than treating a system as a convenient substitute for human accountability.

Memory is another element of accountability. The AP described the ruined Minab school as a memorial, reporting the words: “This will never be a school again.” In March, Win Without War placed children’s backpacks on Capitol Hill for the dead. The article argues that ordinary people can keep such memorials visible to lawmakers who might otherwise look away. It also points to diplomacy: a Pakistan-brokered ceasefire in April, described by Imran Khalid of Foreign Policy in Focus as “a shift in the center of the diplomatic world,” demonstrated that diplomacy could still find a path. Pape has said that an agreement resembling the 2015 nuclear deal, with low enrichment, sanctions relief, and inspections, remained possible. Those examples do not guarantee a settlement, but they make clear that conflict is shaped by political choices as well as military tools. No system that accelerates action can replace negotiation, public scrutiny, or the difficult work of compromise.

The consequences are not confined to Iran. The article reports that the UN human rights chief said civilians were bearing the brunt of renewed fighting in Yemen, with 32 civilians killed since August, including 15 children. In Ukraine, the UN documented 15,280 civilians killed or injured in the first eight months of 2026. It also reports civilian deaths in U.S. strikes in Iran, including four people at a wedding according to Iran’s September count, and missile and drone strikes affecting Gulf cities, including airports in Dubai and Abu Dhabi. Those facts do not prove that every event has the same cause or operational pattern. They do show why civilian harm cannot be treated as a local footnote. The next decade’s cities will be shaped by choices about how authorities use data and automated tools, but also by whether people can challenge decisions and obtain an honest account of harm. Technology may change the pace of action; accountability determines whether the public is left alone with its consequences.

Ted Chiang on AI Fiction, Power, and What LLMs Are Not

A different kind of AI question appears in a conversation with science-fiction writer Ted Chiang: how can authors write about artificial minds while technology companies and media outlets promote claims about current AI? The interview begins with an idea from Ali Alkhatib’s 2024 essay: AI should be understood not merely as a tool with political features, but as a political project that can move authority and autonomy away from individuals toward centralized power. Chiang called that framing useful, especially because AI boosters may describe a system as “just a tool — like a hammer” while also insisting it will be transformative in a way hammers are not. He connects this tension to his 2023 essay, “Will A.I. Become the New McKinsey?”, asking whether AI can do something other than sharpen capitalism’s knife. His observation is that it is easy to see how AI can help capital against labor, and much harder to see how it helps labor against capital. This is an argument about institutions and power, not a claim that a machine has intentions.

Chiang also draws a firm boundary between present-day large language models and fictional conscious beings. He says LLMs are not conscious, have no subjective experience, cannot suffer, and therefore are owed nothing as sentient entities. In his view, current LLMs are not relevant evidence that the kind of artificial life depicted in his novella The Lifecycle of Software Objects has arrived. He says that if software ever did have subjective experience and could suffer, moral obligations toward it would be an important question. But he describes that possibility as theoretical, adding that “We are no closer to conscious software now than we were sixteen years ago.” The interview gives no technical test for consciousness and does not claim that a particular architecture could produce it. Its useful distinction is conceptual: a system that generates language is not thereby a person with experiences. Treating those categories as interchangeable can distort public debate about what existing systems do and what remains speculative.

For writers, that distinction creates both creative freedom and a communication challenge. Chiang says no author can guarantee that readers will understand a story as intended, and notes that technology advocates may interpret speculative work through their own certainties. He offers several ways to avoid having a fictional AI read as a prediction about current chatbots: make the systems clearly different, place the story in a distant future or alternate past, use a secondary world, or explain the fictional AI’s scientific basis carefully. Another option is to write stories that directly challenge industry narratives, such as imagining an AI proposing terrible solutions to climate change. These approaches do not guarantee a particular reading. They give authors ways to signal what a story is and is not claiming. Chiang points to Greg Egan’s 2024 novelette Death and the Gorgon, in which wealthy people freeze their brains hoping for future revival, as an example that engages with present-day technology while running counter to some Silicon Valley narratives.

The relevance for technology firms is not that fiction should be treated as a forecast. It is that the stories surrounding AI can shape how people interpret products, promises, and power. Chiang’s question—whether AI can do more than sharpen the knife of capitalism—asks who benefits and who gains control, rather than focusing only on whether a system can produce impressive output. That scrutiny has a practical counterpart in organizations adopting AI: identify what the tool actually does, which decisions it influences, and who remains responsible for the result. The interview does not report market figures, adoption rates, or a new technical product, so none should be inferred. Its contribution is a careful vocabulary for separating current LLMs from hypothetical conscious software and for looking at the political arrangements around technology. In the coming decade, those distinctions will matter in civic discussion as much as in science fiction: public expectations can be shaped by hype, while governance depends on accurately describing present capabilities and limits.

Where Robles.AI Fits

For a company considering AI, the practical question is specific: which processes involve consequential decisions, what information would an AI system use, and how will people review its outputs? Those questions are especially important when speed could obscure stale information or make responsibility difficult to trace.

We offer two-week fixed-fee AI diagnostics covering up to two processes (Esencial), four processes (Avanzado), or six processes (Integral), as well as a scoped direct AI project delivered from a proposal. We also offer live demos for document question answering with RAG, structured document extraction, speech transcription, forecasting, and other listed capabilities.