Book details how Project Maven moved from drone video to targeting help

In short: A new book reports that the Pentagon’s Project Maven has grown from sorting drone video to helping the military identify targets much faster, with humans still required to approve lethal actions.

What happened

Project Maven began in 2017 as a U.S. Department of Defense effort to use artificial intelligence to scan drone footage. The original goal was to deal with a basic problem, there was so much video that people could not watch it all. The book says much of that footage from the Global War on Terror went unanalyzed.

According to Katrina Manson’s Project Maven: A Marine Colonel, His Team, and the Dawn of AI Warfare, the program soon pushed beyond surveillance. It aimed to help analysts spot and track people and objects, and then develop potential targets. Manson writes that this sped up work dramatically, in some cases from under 100 targets a day to about 5,000, like going from hand sorting mail to using an automatic sorter.

The book also describes setbacks and partnerships. Google left the project in 2018 after employee protests about possible lethal uses. Work continued with other companies, including Palantir, Microsoft, and AWS.

A central tool is the Maven Smart System, software that combines image recognition (computers spotting things in pictures) with large language models, which are text-focused AI systems that can summarize and organize information. Manson reports it is now used across all U.S. military branches, and in places ranging from submarines to space operations.

Why it matters

The story highlights a hard line governments are trying to draw, speeding up war decisions with AI while keeping a person responsible for life and death choices. Manson argues humans still control mortal decisions, but she also notes concern that some newer systems could move closer to weapons that can find and kill targets on their own.

Source: NYTimes