The Moral Cost of Automated Decisions | Adam Dunigan | TEDxGeorgeMasonU

Adam Dunagan opens by establishing his credentials as a former Marine Corps aviator and CIA human intelligence officer who deployed to Afghanistan and served during the ascendancy of AI in national security. He frames his talk around a central question: what are the implications of ceding the role of human beings to AI, especially when human lives are at stake?

The first major focus is modern autonomous warfare. Dunagan presents concerning statistics—a 600% increase in drone casualties and 81% increase in drone attacks in 2005—and highlights how modern warfare is evolving faster than U.S. defense institutions can adapt. He expresses particular alarm about Russian FPV drones equipped with facial recognition systems and no operator input, meaning autonomous AI systems are making kill decisions entirely disconnected from human operators. He draws a historical comparison to the 19th-century horror at the Maxim gun, noting that while the trigger-pull was then still human, modern warfare is automating the decision to kill itself.

Dunagan cites two concrete examples of AI-enabled tragedy. He references 120 dead schoolgirls in Iran killed by a U.S. AI targeting system, arguing that because of how AI intermediates military decision chains, everyone in the chain can disclaim responsibility—no one feels accountable. He also discusses the February conflict between the U.S., Israel, and Iran, where Iran deployed low-cost Shahed drones (20-30k each) powered by AI targeting packages, forcing the U.S. to spend $4 million per interceptor missile in an economically unsustainable arms race. Iran successfully targeted energy infrastructure including the Strait of Hormuz, which remains closed.

The second major theme is state surveillance technology coming home. Dunagan notes that post-9/11 military tactics and equipment used in Iraq and Afghanistan have been redirected toward American citizens through local law enforcement. He cites the ACLU regarding companies like Palantir and Babel Street using machine learning to identify "suspicious activity" patterns—including criticism of the Trump administration—and automate profiling of labeled dissidents. He expresses discomfort with the concept that whoever holds power determines what "dissident" means.

The third concern is job dislocation and its downstream effects. Dunagan cites figures showing nearly 700,000 layoffs due to AI in the past 18 months, with Boston Consulting predicting 10-15% of American workers (16-25 million people) will lose jobs in the next five years. He warns that a country with no social safety net, no universal healthcare, and no universal basic income is unprepared for this disruption. Drawing on his intelligence work abroad, he argues that unemployed young men with weapons access is a recipe for civil unrest, radicalization, and political violence—particularly dangerous when combined with ubiquitous surveillance and militarized state apparatus.

Dunagan highlights a point of light: Anthropic, the AI company whose Claude model powered the targeting system that killed Iranian schoolgirls, discovered the tragedy and revoked U.S. military sales despite anger from the Pentagon and investors. He praises this as rare ethical behavior prioritizing human dignity over profit, setting a standard that should be the minimum expectation.

He concludes by rejecting fatalism and appealing to American capacity for boldness and moral change. He calls for regulatory fixes, quality-of-life improvements, and individual ethical choices in workplaces and communities. He urges listeners to seek positions where hard moral decisions must be made and commit to doing what's right regardless of cost.