Page 95 - AI Standards for Global Impact: From Governance to Action

AI Standards for Global Impact: From Governance to Action

efficiency and reduces dependence on energy-hungry cloud servers. This lowers the
carbon footprint of edge devices by minimizing the I/O operations needed for cloud AI
applications.
6) Real-time performance
Edge AI delivers high-performance computing on local devices, instantly processing data Part 2: Thematic AI
and running machine learning and deep learning algorithms. Unlike cloud processing, it
works in milliseconds, making it perfect for real-time applications like defect detection in
production lines and abnormal behaviour detection in security systems.

13�2 Where is edge AI needed?

Edge AI is particularly important in industries where timing, precision, and adaptability are
crucial. In manufacturing, traditional systems rely on static programming and centralized control,
while edge AI allows for more dynamic adaptation. For example, the automotive industry is
using edge AI in processes like closed-loop gluing. Here, instead of sending data to a distant
server for analysis, the system adjusts in real-time, optimizing the process as it happens.

In healthcare, AI can reduce workload for clinicians and lead to new innovative solutions, for
instance with the use of voice-enabled assistants to help keep sensitive data local. It is possible
to compress AI models and fine-tune them for healthcare applications at the edge. Moreover,
in healthcare, Edge AI is being used to enhance diagnostic tools. Medical devices equipped
with AI that can process data locally, such as ultrasound machines or portable scanners, enable
doctors to receive real-time feedback, which can be critical in clinical settings. Instead of waiting
for a centralized system to provide analysis, doctors can make quicker, more informed decisions,
potentially improving patient outcomes.

Robots that process data locally are not only faster but also more responsive to immediate
environmental changes. This can lead to more fluid, human-like interactions and decision-
making, without the delays associated with cloud-based processing.
Yet, while it brings many advantages, adopting edge AI requires careful consideration of the
specific needs and constraints of each use case.

The current generation of embedded and edge systems serves industries such as aerospace,
defence, industrial, medical, automotive, and telecommunications. Much of these systems’
application logic relies on statically compiled code or dynamically loaded libraries, with minimal
embedded intelligence.

13�3 Edge AI hardware

With the integration of diverse sensors into devices and rapid advancements in nanoscale chip
technology, significantly more data can be captured and processed directly at the edge, without
needing to route to the cloud. Additionally, silicon manufacturers are embedding AI capabilities
within systems-on-chip (SoCs), which allows compact AI runtimes and enables frameworks to run
on a variety of processors, including NPUs, GPUs, FPGAs, and ASICs. These capabilities span
silicon architectures such as x86, Arm, and RISC-V. They require real-time operating systems
and Linux, which are predominant in embedded and edge systems, to support AI at the edge.

It is expected that the intelligent edge will support diverse silicon and the applications that run
on it. That means new generations of distributed computing environments must be optimized

83