My Key Take Aways from the 2024 EASA AI Days
Key take aways of the EASA AI Days (2-3 July 2024) High Level Conference.
The 2nd and 3rd of July 2024 have been packed with valuable information, updates, and reflections on the future of AI in aviation from EASA, agencies, and industry professionals during the EASA Artificial Intelligence Days — High-Level Conference 2024.
A significant representation of the industry was present, including Airbus, Boeing, Honeywell, and Thales, among others.
Despite the main focus being on supervised learning approaches, there were references to unsupervised learning such as Generative AI and Reinforcement Learning, which will be further addressed in the next updates of the EASA Concept Paper.
Let me summarise today my key takeaways along with my personal insights from these two inspiring and promising days:
Pioneering efforts from EASA
A Collaborative Approach with Stakeholders
The Interface with the AI Act
What about Generative AI?
AI Ethics
Conclusions
Let’s dive in! 🤿
Pioneering efforts from EASA
EASA has been a pioneer, starting work on AI back in 2019 to initiate exploration and develop the first documented guidance through the “EASA Concept Paper: Guidance for Level 1 & 2 Machine Learning Applications”, paving the way for AI introduction in aviation.
The EASA Concept Paper and Roadmap are crucial documents, setting the foundation for a unified understanding in the aviation community and anticipating requirements for early applicants.
So far, EASA has hit the milestones of the EASA Roadmap 2.0, now entering the consolidation phase (2024-2027).
They have recently published the ToR for the Rulemaking RMT.0742 and continue to explore:
AI Assurance Technical Scope (Building Block #2), which includes: Learning Assurance; Development and Post-ops Explainability; Data Recording.
Human Factors for AI (Building Block #3), which includes: Operational Explainability; Human-AI Teaming; Modality of Interaction.
Ethics-Based Assessment (as part of Building Block #3).
Advanced Automation (Level 3 AI).
Generative AI and Tools for Operational Use within Organisations.
The following figure is taken from the EASA Concept Paper illustrating the four Trustworthy AI Building Blocks.
A Collaborative Approach with Stakeholders
EASA is actively engaging with the industry, academia, and other regulatory agencies to develop use cases, conduct research, and issue relevant material, notably through the Machine Learning Application Approval (MLEAP) initiative. This initiative underscores EASA’s commitment to integrating AI technologies safely and effectively into aviation.
The development of the anticipated Means of Compliance (MOC) will be based on the outcomes of various research and innovation projects, especially the Horizon Europe ‘Machine Learning Application Approval’ (MLEAP) initiative. These efforts are complemented by discussions within certification projects and advancements in industrial standards. Key contributions are expected from joint working groups such as EUROCAE/SAE WG-114/G-34 and EUROCAE/RTCA WG-72/SC-216.
Overview of the MLEAP Project: Led by Airbus Protect, the MLEAP research project has included consortium members Numalis and LNE. Airbus successfully presented the results of the research project, which primarily aimed to identify concrete means of compliance with key learning assurance objectives, including the following expected benefits through three complex aviation use cases:
Data completeness and representativeness
Generalisation capabilities of deployed models
Algorithm and model robustness
Pathfinder for future approvals
By presenting these results, the aim is to support the establishment of clear guidelines and frameworks to ensure that machine learning applications in aviation meet stringent safety and performance standards. The Final MLEAP Report is available at the provided link.
Monitoring of AI Standardisation Efforts: EASA is also closely monitoring the progress of other AI-focused working groups, particularly ISO/IEC SC42 and CEN CENELEC JTC21. ISO/IEC JTC 1/SC 42, a subcommittee within the joint technical committee formed by ISO and IEC, plays a pivotal role in the standardisation of artificial intelligence across various dimensions. This ongoing surveillance ensures that EASA's guidelines and standards remain aligned with global AI standardisation efforts.
Impact on the FAA Roadmap: The insights and developments from these efforts are contributing to the Federal Aviation Administration (FAA) roadmap, which is expected to be released in 2024. This alignment between EASA and FAA highlights the importance of international cooperation in the standardisation and regulatory oversight of AI in aviation.
Learning from Other Sectors: Recognising the novelty of AI technologies, EASA is drawing lessons from other sectors. Despite some mismatches, EASA has successfully mapped AI classification typologies from other domains, ensuring a coherent and comprehensive approach to AI integration in aviation.
The interface with the AI Act
As introduced in my article “How the EU AI Act shapes the future of Aerospace”, the EU AI Act represents a landmark regulatory framework aimed at ensuring the safe and ethical development and deployment of artificial intelligence across Europe.
Given the increasing integration of AI technologies in various sectors, the aerospace industry is significantly impacted by this legislation.
This act follows a risk-based approach and includes a High-Risk AI System list in Annex I, which encompasses aviation legislation. This imposes obligations on providers and deployers, with requirements coming into force 24 months post-publication.
Aviation falls within the category of “high-risk systems embedded in products”, as it is defined as AI systems that can have a significant impact on safety or rights, including critical infrastructure, education, employment, and law enforcement applications. Hence, these systems must meet strict requirements for data governance, transparency, robustness, and human oversight.
What about Generative AI?
EASA will consider Generative AI tools as "Operational Tools," to be approved as part of organisational approval (e.g., Part 145, Part M, DOA, POA, etc.).
EASA is evaluating the applicability of the Concept Paper, considering public generative tools, prototyping in a sandbox, and developing end-to-end applications.
Generative AI is in the top three of the EASA AI Programme priorities, with a target completion by the end of 2025.
AI Ethics
There has been extensive debate on AI ethics, covering risks like de-skilling, impacts on emotional intelligence, and new implications of responsibility and accountability.
EASA´s role in addressing ethical implications of AI in aviation is key as they should be implicitly be part of the considerations in aviation. These implications are complex and continuous collaboration is required with other stakeholders by continue promoting the investigation of ethics-by-design processes and extending the risk-based approach into the ethics domain.
Conclusions
From my point of view, these two days highlight the progress EASA, along with the aviation industry, is making towards defining the future of AI in aviation.
What am I really excited about? Because here’s what’s on the near horizon:
🌟 Phase 2 of the Roadmap: EASA is entering Phase 2 of the Roadmap: Framework Consolidation (2024 - 2027).
🌟 New Rulemaking Task: The EASA Terms of Reference for rulemaking task RMT.0742 Issue 1 have been issued, applicable across the Basic Regulation, in response to the EU AI Act, and based on the EASA Concept Paper ‘Guidance for Level 1 & 2 Machine Learning Applications’ Issue 02 (target end 2027).
🌟 Focus of the TOR: The TOR for RMT.0742 focus on Level 1 AI (Cognitive Human Assistance) and Level 2 (Human-AI Teaming), leaving Level 3 AI (Advanced Automation) for future development through the EASA AI Concept Paper Issue 03.
🌟 First Approvals: The first approvals of Level 1 AI/ML are expected by the end of 2025.
🌟 Generative AI Tools: EASA will consider Generative AI tools as "Operational Tools," to be approved as part of organisational approval (e.g., Part 145, Part M, DOA, POA, etc.). EASA is evaluating the applicability of the Concept Paper, considering public generative tools, prototyping in a sandbox, and developing end-to-end applications (target end 2025).
What have I missed?
From my side, I’m particularly interested in AI as operational tools within organisations. Hence, I noted a nearly absent reference to organisational requirements, as shown in the figure below from the EASA Roadmap.
However, it´s completely understandable as the say, "We must learn to walk before we can run” hence now the big efforts are put on the development of the AI trustworthiness guidance which will be key to point out the direction of the future regulatory framework including the Deployment, operational Phase and Organisational Requirements.
That's all for today.
See you next week 👋
Disclaimer: The information provided in this newsletter and related resources is intended for informational and educational purposes only. It reflects both researched facts and my personal views. It does not constitute professional advice. Any actions taken based on the content of this newsletter are at the reader's discretion.