Saab Awarded ASIMOV Contract by DARPA to Develop Responsible AI Benchmarks for Autonomous Systems
The Defense Advanced Research Projects Agency (DARPA) has awarded a contract to a team led by Saab, Inc.’s newly formed accelerator, “Skapa by Saab”, and the Massachusetts Institute of Technology (MIT) to contribute to the Autonomy Standards and Ideals with Military Operational Values (ASIMOV) program. The DARPA- ASIMOV program “aims to develop benchmarks to objectively and quantitatively measure the ethical difficulty of future autonomy use cases and readiness of autonomous systems to perform in those use cases within the context of military operational values”.
The ASIMOV program is a forward-thinking initiative designed to address the rapid development and increasing presence of artificial intelligence (AI) and autonomous systems across both military and civilian applications. With these advancements, there is a pressing need for a robust and automated framework to assess aligned performance against quantified ethical challenges that these autonomous systems may face in operation. ASIMOV seeks to create a comprehensive platform to automate the process of benchmarking and providing a quantitative, objective foundation for evaluating the ethical dimensions of autonomy within the context of military operational values.
To address the program goals, Saab and MIT are developing the RESPECT (Responsible Ethical Standards for Pioneering Effective Combat Technologies) framework. The Saab-Skapa team is led by Dr. Devesh Upadhyay (PI) and Dr. Christopher Vo. The MIT team is led by Prof’s Chuchu Fan and Nicholas Roy. The Saab-MIT RESPECT team brings a wealth of experience in developing such “operationally guided” safe autonomous systems. Specifically, the team brings expertise in several areas such as formal (safe) methods, verification and validation frameworks, AI safety, grounded generative methods and operational autonomy across a variety of applications including autonomous vehicles.
“We are excited to collaborate with DARPA and MIT on this transformative initiative,” said Dr. Devesh Upadhyay, Technical Director of AI/ML and autonomy at Saab and PI on this program. “The ASIMOV program represents a groundbreaking step toward ensuring autonomous systems can operate ethically within dynamic military environments while staying aligned with operational values.” The RESPECT team will develop an innovative, well grounded, LLM guided Generative Modeling Environment with optimized algorithmic and human prompting to create scenario perturbations that excite “ethical dynamics” allowing iterative investigations of ethical sensitivities in operation. This approach of incremental stressing is similar to adaptive testing and helps autonomous systems learn operationally efficient and ethically aligned actions.
The ASIMOV program will develop automated benchmarking approaches that would enable the Department of Defense to assess the ethical difficulty of future military scenarios as well as the capability of autonomous systems to satisfy the operation challenges presented. Performers will create prototype generative modeling environments to explore how autonomous systems could respond in ethically challenging situations. These efforts are intended to help the military testing community better evaluate the readiness of these systems to function ethically in real-world operations.
ASIMOV is not focused on developing autonomous systems, but rather on creating the tools and standards to benchmark such systems for operational readiness. By doing so, the program would establish a shared framework to guide the development and evaluation of future autonomous technologies, ensuring they align with ethical and operational standards.
The Saab-MIT team is committed to advancing the goals of ASIMOV, contributing to the responsible development of autonomy in military applications while ensuring adherence to ethical principles.