Part of the strength of AI lies in acknowledging that it may provide unexpected results sometimes and this is accepted by the creators and users of the applications. This is orthogonal to traditional safety critical systems development which starts fully deterministic control of the system to be able to verify the system behavior in every possible situation.
There are a couple of options in how to manage the integration going forward:
It is possible to build such deterministic behavior into the AI processor, like Nvidia has done.
An other approach would be to add a deterministic layer in-between that would ensure the safety of the vehicle.
AI could be also be completely separated from such safety critical systems
Building such critical software components is in any case a significantly challenging task that requires deep understanding of software engineering. However, separating the AI with a controlled layer could bring benefits and it might be the only viable approach to realize complex AI-based applications in cyber-physical systems.
Applications for consumers appear to be mostly coming from an IT domain or vehicle domain, however, some examples of hybrids do exist. Applications from each direction have specific advantages. Applications are of a completely new kind, which combine real-world objects with cyber domain information management.
IT domain-based services are big in information processing and agility of companies. Introduction of AI & blockchain have widened this gap as it allows more rapid development of applications while being deployed. IT domain companies are pushing towards expanding their domain into the vehicle engineering by pushing software platforms to be incorporated into vehicles. Examples: Android Auto, Apple Carplay, MS Azure Edge. Disadvantages of IT domain based services are centralized characteristics and exposure to complex real-life objects, such as cars. Integration of complex IoT devices is something that is only starting to take off in the industry.
IT域服务在信息处理和企业敏捷性方面具有很大的应用价值。AI和Basklink的引入拓宽了这一差距，因为它允许在部署时更快速地开发应用程序。IT领域的公司正在推动将其领域扩展到车辆工程中，通过将软件平台纳入车辆。例子：Android汽车，Apple Carplay，MS Azure Edge。IT域服务的缺点是集中特性和暴露于复杂的现实生活对象，例如汽车。复杂的物联网设备的集成在此行业中才刚刚起步。
Vehicle domain-based services’ strengths lie in the precise control of vehicles with an ability to access all information related to the vehicle. Development of embedded computing platforms has made it possible to implement more and more sophisticated applications within an embedded domain, that can control vehicles with better precision. Vehicle originating services are pushing towards cloud-based approaches by implementing their own services, which is often slow and lacking behind ’IT native’ companies. Technologies used in cloud oriented applications, such as AI and Block chain are being adopted to vehicle-based systems as well. Disadvantages of vehicle domain-based services are long development timeframes and transaction-type business / development models.