Intelligence notes

◉Dashboard↗

War-Algorithm Accountability

3 min read

War-Algorithm Accountability

Authors: Dustin A. Lewis, Gabriella Blum, Naz K. Modirzadeh
Published: 2016-09-12
arXiv: 1609.04667
Source: arXiv (cs.CY) | Harvard Law School Program on International Law and Armed Conflict

Abstract

In this briefing report, we introduce a new concept (war algorithms) that elevates algorithmically-derived choices and decisions to a, and perhaps the, central concern regarding technical autonomy in war. We define war algorithm as any algorithm that is expressed in computer code, that is effectuated through a constructed system, and that is capable of operating in relation to armed conflict. Through that lens, we link international law and related accountability architectures to relevant technologies. We sketch a three-part (non-exhaustive) approach highlighting traditional and unconventional accountability avenues: state responsibility, command responsibility, and individual criminal accountability. By not limiting our inquiry only to weapon systems, we take an expansive view, showing how war algorithms might already fit within the existing regulatory system established by international law.

Why This Work Matters

This 2016 paper is the conceptual precursor to legal accountability debates that became acute in the Lavender context. The authors’ concept of “war algorithm” — defined to include any algorithm capable of operating in relation to armed conflict, not just fully autonomous systems — is broader than the LAWS (Lethal Autonomous Weapons Systems) frame that dominated contemporaneous discourse and more accurately describes hybrid human-machine systems like Lavender, where formal human authorization was compressed to seconds of review.

The three-part IHL accountability framework anticipates the accountability vacuums that +972 investigative reporting documented: when the “human-in-the-loop” is a rubber stamp, none of the three traditional accountability pathways function as designed.

Core Concepts and Contributions

War algorithm concept: The definitional move — unbounding analysis from “autonomous weapon systems” to any conflict-relevant algorithm — captures the full spectrum from fully autonomous to “human-on-the-loop” to minimally supervised systems in a single analytical frame. Lavender, where human review was 20 seconds and gender confirmation, falls within this broader concept.

Three-part accountability framework: (1) State responsibility — states bear legal responsibility for systems’ outputs under IHL; (2) Command responsibility — commanders who knowingly deploy systems with foreseeable IHL violations bear criminal liability; (3) Individual criminal accountability — operators who authorize strikes with awareness of algorithmic error rates may bear liability. Existing law formally covers these pathways; enforcement mechanisms are systematically inadequate for algorithmic-speed, algorithmic-volume warfare.

Scale as accountability challenge: The report explicitly anticipates that the volume of algorithmic decision-output would overwhelm traditional review mechanisms — borne out by the Lavender model (37,000 targets, 20-second review, 10% accepted false-positive rate).

Connections

Graph View

Backlinks