GENEVA - Armies of Terminator-like warriors fan out across the battlefield, destroying everything in their path, as swarms of fellow robots rain fire from the skies.
That dark vision could all too easily shift from science fiction to fact unless such weapons are banned before they leap from the drawing board to the arsenal, campaigners warn.
On Tuesday, governments began the first-ever talks exclusively on so-called "lethal autonomous weapons systems" - opponents prefer the label "killer robots".
"All too often international law only responds to atrocities and suffering once it has happened," said Michael Moeller, head of the UN Conference on Disarmament.
"You have the opportunity to take pre-emptive action and ensure that the ultimate decision to end life remains firmly under human control," he told the meeting in Geneva.
That was echoed by the International Committee of the Red Cross, guardian of the Geneva Conventions on warfare.
"There is a sense of deep discomfort with the idea of allowing machines to make life-and-death decisions on the battlefield with little or no human involvement," said Kathleen Lawand, head of its arms unit.
The four-day meeting aims to pave the way for more in-depth talks in November.
"The only answer is a pre-emptive ban," said Human Rights Watch arms expert Steve Goose.
UN-brokered talks have done that before: blinding laser weapons were banned in 1998, before they ever hit the battlefield.
Automated weapons are already deployed worldwide.
The best-known are drones, unmanned aircraft whose human controllers push the trigger from a distant base. Controversy rages, especially over the civilian collateral damage caused when the United States strikes alleged Islamist militants.
Perhaps closest to the Terminator of Arnold Schwarzenegger's action films is a Samsung sentry robot used in South Korea, able to spot unusual activity, quiz intruders and, when authorised by a controller, shoot them.
Other countries in the research vanguard include Britain, Israel, China, Russia and Taiwan.