RTX BBN Technologies has been awarded a contract by the Defense Advanced Research Projects Agency (DARPA) to support the Human-AI Communication for Deontic Reasoning DevOps (CODORD) program. The initiative aims to bridge the gap between human regulatory language and artificial intelligence comprehension-helping computers accurately interpret and apply complex rules in a verifiable and explainable manner.
Translating Regulations into AI-Comprehensible Logic
At the core of BBN’s approach is DONATELLO (DeONtic AssistanT Enabled by Large Language mOdels), a human-in-the-loop AI solution that converts dense, exception-filled regulatory texts into executable logic. By doing so, DONATELLO is designed to simplify compliance processes and provide decision-makers with clearer, rule-based reasoning.
Early research from DONATELLO subcontractors highlighted a key challenge: even advanced large language models, such as GPT-4, can generate inaccurate answers and flawed justifications when interpreting regulations riddled with ambiguities and exceptions. DONATELLO seeks to overcome these shortcomings by delivering specific, explainable reasoning that enhances computational understanding and trust in AI-driven regulatory support.
Also Read: Peraton and Okta Collaborate with U.S. DoD to Advance myAuth
“Regulatory texts are complex, full of exceptions and nuances that challenge traditional programming,” said Dr. Brian Ulicny, BBN principal investigator. “We need systems that can explain why specific actions are permitted or prohibited in ambiguous situations.”
Supporting Critical Decision-Making in Defense Environments
In high-stakes operational settings-such as those faced by the U.S. Department of Defense-access to regulatory experts is not always immediate. In these cases, decision-makers require AI systems that can provide clear, actionable insights rooted in established frameworks.
Deontic logic, which defines what is permitted, prohibited, or obligatory, offers a structured way for AI tools to process regulations. By applying this framework, DONATELLO aims to give military and government users confidence in how AI interprets and applies regulatory guidance.
“With DONATELLO, we’re not just interpreting regulations; we’re empowering decision-makers with clarity and precision,” Ulicny added.