The European Union Aviation Safety Agency (EASA) has published a new report on the use of onboard artificial intelligence in aviation, in collaboration with AI specialist Daedalean.
The Zurich-based Daedalean’s mission is to build a certified airworthy AI autopilot system that can pass human pilot exams.
The new report from the pair is the result of 10 months of work, and aimed to investigate the challenges and concerns of using neural networks in aviation.
This is part of a wider effort on the part of European authorities to adapt to the ever-evolving technological landscape, with a number of similar moves made around AI technologies since the beginning of 2019.
The following are some of the reported obstacles to the use of AI in aviation, as described by EASA:
- Traditional development assurance frameworks are not adapted to machine learning;
- Lack of predictability and explainability of the machine learning application behaviour;
- Lack of guarantee of robustness and of no ’unintended functions’;
- Lack of standardised methods for the evaluation of operational performance of the machine learning applications;
- Issue of bias and variance in machine learning applications;
- Complexity of architectures and algorithms;
- Learning processes are adaptive.
Guillaume Soudain who led the project at EASA said: “Our investigation allowed us to take a decisive step in defining a learning assurance framework, which is one of the fundamental building blocks of the EASA AI Roadmap for the creation of an ‘AI trustworthiness framework’.”