The Federal Aviation Administration (FAA) has issued guidance on the certification of AI in aviation software, highlighting significant challenges due to the nature of machine learning models. Current aviation software assurance regimes, like RTCA DO-178C, rely on a traceable chain of evidence from requirements to code, which learned models lack. Key objectives such as bidirectional requirements traceability, requirements-based test coverage, structural coverage analysis, and ensuring no unintended functions are difficult or impossible to meet with current ML architectures. While the FAA can issue special conditions for AI components on a project-by-project basis, there is no established certification basis or supplement for machine learning, leading to models being frozen in their current state for certification. AI
IMPACT Highlights significant regulatory hurdles for integrating AI into safety-critical aviation systems, potentially slowing adoption.
RANK_REASON The item details regulatory guidance and challenges for a specific technology within an industry. [lever_c_demoted from research: ic=1 ai=0.4]
- 14 C.F.R. Part 21
- Advisory Circular 20-115D
- Federal Aviation Administration
- machine learning
- RTCA DO-178C
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
