On Safety Assurance of Symbolic Artificial Intelligence

Artificial Intelligence (AI) has gained popularity in recent years. Symbolic Artificial Intelligence (symbolic AI), a subset of AI, holds significant promise, particularly in avionics software, offering potential enhancements to various aspects of its functionality, such as automatic decision inference or gen-erating optimal decisions. Despite its benefits, the integration of symbolic AI into avionics software introduces distinctive challenges, particularly in ensuring the safety and reliability of software powered by this technology. In this paper, we propose a set of supplementary objectives intended to support the DO-178C standard, specifically addressing features associated with symbolic AI. Additionally, we outline a set of metrics de-signed to support the generation of evidence for the proposed objectives. To illustrate the proposed assurance objectives, we made use of a TPN solution finder, which employs symbolic AI techniques to generate optimal routes for remotely piloted aircraft.

1. Introduction

The development of avionics software products has to conform to the DO-178C [17] standard. DO-178C provides a set of certification objectives aimed at guiding the development of avionics software products. The results of all development processes must be then verified for compliance with the certification objectives. Verifying software products can be a complex task. The verification, according to DO-178C, is performed by reviews, analysis, or tests. However, when using formal methods, DO-178C recommends to make use of the guidelines provided by RTCA DO-333 [16], which is a supplement to DO-178C. DO-333 identifies the modifications and additions to DO-178C objectives, activities, and software life cycle data that should be addressed when formal methods are used as part of the software development process. It includes artifacts that would be expressed using some formal notation and the verification evidence that could be derived from them.