A look at autonomous weapons systems being developed by the US and the arguments for and against imbuing autonomous robots with lethal decision-making
Kelsey D. Atherton / New York Times : Tweets: @cnasdc , @nytimesatwar , and @bankillerrobots Tweets: Cnas / @cnasdc : In a conversation with @NYTmag, @paul_scharre describes “a world where we'll have groups of robots operating collaboratively together under one person in supervisory control.” Read more: http://www.nytimes.com/... Nyt At / @nytimesatwar : Advancements in A.I., machine image recognition and robotics have set some of the world's largest militaries on the cusp of autonomous systems, but under what circumstances can and should they delegate the decision to take a human life to machines? http://www.nytimes.com/... @bankillerrobots : Not just governments, @UN, @ICRC & our campaign that are concerned by fully autonomous weapons. @google's ethical principles on artificial intelligence include “an outright refusal to develop” weapons. @DeepMindAI @sundarpichai http://www.nytimes.com/...
Context & Ripple Effects
This 2018 conversation with CNAS's Paul Scharre frames the question that has organized the autonomous-weapons debate ever since: not whether robots fight, but how much lethal authority a human keeps when 'groups of robots operate collaboratively under one person in supervisory control.' It lands between two poles already visible in the coverage — Google's stated outright refusal to develop weapons, and startups like Clarifai signaling they would supply autonomous weapons because internal ethics officers don't suit small companies.
The arc since then runs through [[a:968265|thirty countries backing a total ban while the US calls concerns overblown and Russia says such weapons don't yet exist]], and forward to [[a:1169658|the Pentagon's full embrace of autonomous weapons ahead of its fight over Anthropic's two red lines]]. Shneiderman's argument that full automation absolves humans of ethical responsibility supplies the intellectual case against the supervisory-control model this article popularizes.
First-order effects
- The US military gains a concrete control framework — one supervisor overseeing collaborative robot groups — that makes delegation of lethal decisions administrable rather than absolute, directly shaping what programs like those Scharre examines can propose.
- AI labs split along the weapons line: Google's principles bar weapons work outright, pushing defense demand toward smaller vendors such as Clarifai that are willing to fill it.
Second-order effects
- As big labs refuse and startups volunteer, the Pentagon's supplier base tilts toward state-compatible vendors without internal ethics review structures, making procurement choices a de facto ethics policy.
- The supervisory-control framing gives ban advocates at the UN, ICRC, and Campaign to Stop Killer Robots a specific target — where the human sits in the loop — sharpening the treaty argument beyond a blanket 'killer robots' label.
Third-order effects
- If the pattern holds, lethal autonomy normalizes through a vendor split — refusing giants, willing startups, an embracing Pentagon — until the binding constraint is diplomatic (a ban bloc versus holdouts) rather than technical or corporate.
- The human-in-supervisory-control model risks becoming the industry's standard answer to accountability questions, which critics like Shneiderman argue relocates rather than resolves moral responsibility.
The trend: Militaries are converting lethal decision-making from a human act into a supervisory-control architecture, with corporate refusals and startup willingness deciding who builds it and treaty politics deciding whether it is bounded.