DARPA to add reasoning to autonomous teams
A new program aims to build autonomous systems that can react to changing context and learn from experience.
While teams of autonomous systems offer great potential – think drone swarms -- they currently require a human to direct actions, anticipate problems and make decisions when circumstances change. They're also relatively slow, subject to constraints in communication-restricted environments and rely on predefined functions, static swarm behaviors, or other processes that limit the team’s ability to react to unexpected missions or balance competing objectives.
The Defense Advanced Research Projects Agency wants to lay the foundation for integrating unmanned systems into larger missions by developing reasoning, contextually aware teams of autonomous systems.
The Context Reasoning for Autonomous Teaming (CREATE) program aims to use artificial intelligence to build scalable teams of autonomous systems that can react to changing context and learn from experience to make decisions without centralized control and communications. It is specifically focused on teaming for systems constrained by size, weight and power, such as drones, undersea unmanned vehicles or satellites.
CREATE assumes autonomous agents will tap stores of common knowledge and procedures along with data from own sensors that provide context about their surroundings so they can take informed actions to complete a mission. The program also plans to explore decentralized decision making – where individual machines use estimation and inferencing techniques to determine actions in unexpected circumstances.
To ensure the greatest flexibility, DARPA wants researchers to design autonomous systems that can take a machine-readable mission and, without explicit instructions, come up with creative solutions.
Researchers should also address keeping fully autonomous agents from making bad choices – either by hard-coding in safeguards or developing a system whereby the autonomous agents would understand contextual safeguards.
DARPA's previous efforts to integrate autonomous systems into missions have involved advancing manned-unmanned teaming technologies to enhance capabilities of ground troops.
The Squad X program provides warfighters with lightweight, intuitive technologies that use sensors and off-the-shelf tools to increase squads' situational awareness and allow them to expand their area of influence. The Competency-Aware Machine Learning program aims to give autonomous systems the ability to evaluate their effectiveness and communicate that information to humans. And the Artificial Social Intelligence for Successful Teams program looks to demonstrate the basic machine social skills needed to infer the goals and situational knowledge of human teammates, predict what they will need and offer context-aware actions.
This article first appeared on GCN, a partner site to Defense Systems.
NEXT STORY: The Pentagon is looking for an AI ethicist