Decentralized Artificial Intelligence through Controlled Emergence (DICE)
DICE aims to develop decentralized AI algorithms for scalable, resilient, and human-controlled collectives of diverse agents capable of long-term missions in challenging environments.
Funder: DARPA - Information Processing Technologies Office
Due Dates: June 30, 2026 (Proposal abstract due) | August 18, 2026 (Questions deadline) | August 25, 2026 (Full proposal due)
Funding Amounts: Multiple awards anticipated; 36-month, 3-phase effort; award size not specified—proposers define budgets using detailed cost templates.
Summary: Supports research on decentralized AI collectives, enabling scalable, adaptive, and resilient multi-agent systems for long-horizon missions in contested environments.
Key Information: Abstract submission is strongly encouraged and must use the required template; proposals must be submitted via DARPA BAAT.
Description
This opportunity seeks innovative research to advance the theory and algorithms for decentralized coordination and local inference control in heterogeneous AI collectives. The program aims to enable scalable, adaptive, and resilient groups of AI agents that can autonomously execute complex, long-duration missions in contested or unpredictable environments, while maintaining alignment with human intent and control. Unlike traditional, brittle centralized orchestration or ad hoc agent assemblies, this program (DICE) focuses on self-organizing systems that are robust, predictable, and mission-aligned. The research spans artificial intelligence, control theory, formal methods, and game theory, and is expected to produce new architectures, algorithms, and evaluation platforms for decentralized multi-agent systems.
