- Streamline a Pathway for Integrating AI Technologies in Aviation
Current FAA regulations require any AI use in aircraft systems to go through the Issue Paper process, regardless of the system’s safety criticality. While this process is useful for novel technologies, it creates unnecessary regulatory burdens for AI applications that have minimal safety risks.
Merlin recommends that the FAA exempt low-risk AI applications from additional compliance requirements and allow them to follow existing certification pathways. Expanding the Non-Required Safety Enhancing Equipment (NORSEE) framework to include AI-driven safety improvements would also help accelerate innovation while maintaining high safety standards.
- Continue Industry Engagement Through the FAA STEP Program
The FAA’s AI Summits and STEP (Safety Through Emerging Programs) Program have been invaluable in fostering collaboration between regulators and industry. These efforts ensure that policy development reflects real-world AI applications and challenges. Merlin encourages the OSTP to support ongoing industry discussions through initiatives like STEP, helping prevent unnecessary complexity in AI regulations and ensuring that existing aviation frameworks can be leveraged effectively.
- Set Clear Regulatory Thresholds for AI Systems Based on Safety Impact
AI applications should be regulated based on their actual safety impact, rather than a one-size-fits-all approach. Merlin recommends that the FAA apply its existing safety continuum framework to AI systems:
- Low safety impact AI applications (e.g., those classified as Item Development Assurance Level [IDAL] D or lower) should follow existing regulations without additional certification burdens.
- Higher safety impact applications should undergo proportionally increased scrutiny, following established certification principles.
By aligning AI certification pathways with existing aviation safety frameworks, regulators can avoid unnecessary barriers while ensuring appropriate oversight.
- Create Aviation Communication Datasets for NLP Training
AI-driven aviation systems, particularly those relying on Natural Language Processing (NLP), require large, high-quality datasets to ensure accuracy and safety in real-world operations. However, such datasets are not readily available to developers.
Merlin urges the OSTP to support FAA-led initiatives to develop publicly available aviation communication datasets, similar to previous FAA collaborations with MIT Lincoln Laboratory on airspace encounter models. This would:
- Accelerate AI advancements in aviation safety and automation.
- Support industry-wide standardization for AI verification and validation.
- Enhance operational efficiency within the National Airspace System (NAS).
