Artificial intelligence is no longer a concept of the future in warfare; it is the current reality. Israel’s military employs AI for target identification in Gaza, the United States utilizes similar technology in Iran, and Ukraine has advanced its capabilities with smart drones. This marks a significant shift, transforming the landscape of global conflict.
While comprehensive global policies on military AI, with its potential benefits, inherent risks, and complex ethical considerations, will undoubtedly be a subject of extensive study for decades, Katrina Manson’s “Project Maven” offers a more focused exploration. Rather than a broad policy analysis, Manson draws on interviews with over 200 individuals to chronicle the US military’s journey into AI warfare. This narrative is particularly relevant given the reported 800 AI projects currently underway within the Pentagon.
“Project Maven” was initiated in 2017 with the primary objective of developing a tool capable of analyzing vast amounts of drone footage to extract actionable intelligence. The sheer volume of data collected by drones had surpassed human interpretive capacities. Manson details the project’s challenging early stages. For instance, within eight months of its launch, the system was deployed with soldiers in Somalia. During this initial deployment, its algorithms mistakenly identified schoolbuses in cloud formations and misclassified trees as people.
The book follows a project leader whose experiences as an intelligence officer in Afghanistan underscore the limitations of previous methods. He recounts the difficulties of planning missions and directing troops using little more than a basic laptop and the Microsoft Office suite, grappling with fundamental questions about enemy locations, safe zones, and defining mission success.
Human involvement in conflict introduces inherent inefficiencies. Soldiers fatigue, make errors, and are susceptible to the unpredictable nature of warfare. The builders of the otherwise secretive Project Maven, speaking to Manson, believed that AI could effectively “clear the fog of war.” Their ambition extended beyond mere intelligence gathering; they envisioned AI systems capable of autonomously selecting targets, engaging them, and executing lethal operations. This pursuit aimed to circumvent the sometimes slow process of human decision-making, potentially allowing for rapid neutralization of adversaries through autonomous “killer robots.”
“We kill the wrong people all the time,” one insider shared with Manson. “A machine can’t be worse than a human.” The development team worked to refine Maven into a suite of tools, actively seeking to persuade frontline personnel to adopt them. While these efforts led to improved outcomes, instances of error persisted.
Since its initial development, the United States and other NATO member states have integrated Maven into ongoing conflicts. Manson reports that approximately 32 companies are involved in its development, and 25,000 US military personnel are regular users. The book also describes its application in contexts beyond active combat, such as border surveillance and interdiction operations against drug traffickers in the Caribbean. This raises a critical question: can a nation equipped with such advanced surveillance and interdiction tools resist deploying them against its own citizens?
A particularly concerning aspect highlighted by Manson is the ongoing effort to remove human oversight entirely from the decision-making loop. Systems like the “Goalkeeper” flying drones and “Whiplash” naval drones are apparently designed to independently identify and engage targets. Historically, humanity has a consistent pattern of developing and subsequently utilizing any weaponry it invents.
The narrative inevitably brings to mind the incident with Stanislav Petrov, the Soviet lieutenant colonel who, in 1983, trusted his own judgment to dismiss reports of a US missile launch as a false alarm, thereby averting a potential global nuclear war. The question arises: would an AI system make such a call under similar circumstances?
Despite the significant insights into Project Maven itself, the book ultimately sheds more light on the intricacies of Pentagon bureaucracy and Silicon Valley’s readiness to engage in projects, regardless of their ethical implications, provided the financial incentives are sufficient. Manson’s access to interviewees is described as exceptional. However, the inherent nature of military secrecy means that the full extent of the technology developed by the US government, and the precise details of its implementation and timeline, are likely to remain unknown for years.
Warfare has always been a profoundly grim undertaking. However, modern conflicts, where individuals remotely observe targets via drone feeds and make life-or-death decisions, have introduced a level of detachment. Entrusting these critical decisions to AI risks lowering the threshold for engaging in conflict and diminishes the perceived consequences of such actions.
It is imperative to ensure that the power conferred by AI-driven weaponry is approached with the seriousness it warrants. Yet, Manson presents a disquieting account suggesting the opposite is occurring. One interviewee, reportedly seeking to join Project Maven, is said to have stated their motivation as wishing to “reduce the non-American population” and subsequently secured a position within the project.
Further Reading on AI and Warfare
The Making of the Atomic Bomb by Richard Rhodes
This work offers valuable parallels for understanding the potential trajectory of military AI. Much like the Manhattan Project, it suggests the risk of permanently escalating global tensions and raising the stakes of international conflict.
Should We Ban Killer Robots? by Deane Baker
Penned by an ethics professor, this book delves into the complex discussions surrounding artificial intelligence in warfare. It examines the challenging issues of reliability, control, and accountability that arise when governmental responsibilities traditionally held by soldiers are delegated to computer systems.