Welcome back, readers, for our fourth blog on the ISO/PAS 8800:2024 standard.
In our previous blog, we looked at the type of “faults” addressed in ISO/PAS 8800 standard and how they are different from ISO 26262 standard.
In this blog, let us have a quick peek into few basic jargons used in this AI safety standard, so that we all can understand and speak the “vocabulary” used in the ISO/PAS 8800 standard going further.
As we know, ISO/PAS 8800 standard addresses the risk of hazard due to unintended and/or inadequate behaviour of AI elements used within safety-related item/system of the vehicle.
To understand what an “AI Element” means, let us start with few other terminologies such as “Item”, “Encompassing System”, "AI System", "AI Component" and “AI Model”.
An Item means a system/combination of systems which are safety-related and implements a function or part of a function at vehicle level. This definition of an item is borrowed from ISO 26262 standard.
An item that has an AI system is considered as Encompassing system. Since the scope of AI safety standard is only AI system, we would be referring to AI system and/or encompassing system in ISO/PAS 8800 standard.
AI system means an item/element that uses one/more AI models. It can consist of one/more AI components.
AI model refers to logical and/or arithmetic operations used to generate an inference/prediction based on input without being fully defined by human knowledge.
Element of an AI system is called AI component. It can be an AI model/a conventional SW/conventional HW component.
AI component or AI system itself is referred as AI element.
Below table summarizes some examples for each of the above terminologies.
Terminology | Examples |
Item | Safety-related systems such as Automated Lane Centering, Autonomous Emergency Braking for vehicles and pedestrians, Steering system, Braking system, Air bag etc |
Encompassing system | Safety-related systems that uses AI based methods such as Machine learning based models, deep neural networks etc., for a safety-related applications such as Traffic Sign Recognition, Object detection and classification etc |
AI System | Camera sensor used for lane marking detection, object classification, collision detection etc., Radar/Lidar sensor used for target vehicle path prediction. Domain controller used for fusing data from multiple sensors such as camera, radar etc., to track and detect vehicles and objects |
AI Model | Computer vision-based model used within Camera sensor for Object detection & classification, traffic sign detection, lane marking and type detection etc |
AI Component/AI Element | It can be AI model or conventional SW used for plausibility check of output of AI model or conventional HW component/part used within AI system such as CPUs, GPUs or FPGAs etc |
Example of an AI System and AI terminologies
Informative