2015 autonomous-weapons open letter

The 2015 autonomous-weapons open letter called for a ban on “offensive autonomous weapons beyond meaningful human control.” It addressed weapons that select and attack targets without human intervention, making the delegation of lethal decisions its central concern. The appeal was a proposal for a restriction, rather than an enacted ban.^[1]

Announcement and support

The original hosting page dates the announcement to July 28, 2015 at IJCAI, an international artificial-intelligence conference. In a report published July 29, Max Tegmark names Stuart Russell and Toby Walsh as the announcers at a Buenos Aires press conference and says the Future of Life Institute (FLI) helped organize the letter. FLI hosted and managed signatures; signing does not establish authorship or institutional endorsement.^[1][2]

Tegmark reported more than 1,700 AI/robotics researchers and more than 11,000 other signatories in that July 29 account. These are the organizer’s reported figures at that stage, rather than audited totals at the moment of announcement. Today’s running signature total and the page’s February 9, 2016 publication field do not describe the 2015 announcement.^[2][1]

What the letter sought to prevent

The text contrasts an armed quadcopter searching for people meeting predefined criteria with remotely piloted drones and cruise missiles with all targeting decisions made by people. Its distinction concerns who selects and engages targets.^[1]

The signatories warned of cheap, proliferating weapons and assassination or repression. They acknowledged a tradeoff between reducing soldiers’ casualties and lowering the threshold for war, while allowing protective military AI uses. These were advocacy arguments, not deployment findings.^[1]

What meaningful control requires

In 2018, Filippo Santoni de Sio and Jeroen van den Hoven explicitly discussed the letter in their philosophical account of meaningful human control. They identified a gap between accepting the principle and knowing what technical designs and human responsibilities could satisfy it.^[3]

Their proposed conditions are tracking relevant facts and human moral reasons, and tracing system outcomes to a person with appropriate understanding and recognition of responsibility. Merely placing a person in an approval loop can fail if that person lacks information, practical options or understanding of the system. This develops the control question raised by the appeal; it does not show that the letter’s signatories endorsed the later theory.^[3]

The 2015 beneficial-AI open letter addressed a separate research agenda. The weapons appeal focused instead on limits to a particular use of AI. Its connection to Meaningful Human Control is the question of what people must be able to understand, govern and answer for when a system can act autonomously.