I. What challenges does applying edge AI face?
However, applying AI to solve these problems requires extending AI enhancements to the network edge—the points of data creation and consumption, such as factory floors, hospitals, or stores. Introducing AI capabilities in these environments presents new challenges compared to running AI in data centers or the cloud, including:
Incorporating AI into existing investments: Many edge environments rely on traditional, fixed-function infrastructure equipped with a variety of proprietary devices and software. Space-constrained hardware needs to support real-world requirements for accuracy and performance.
Training and Fine-tuning Models: Edge AI models are unique and must be dynamically tuned for specific industries or use cases. In these cases, human domain knowledge is often crucial. For example, experienced weld inspection personnel can help AI understand how to detect good or bad welds. Businesses need simple tools to help non-data scientist experts translate their expertise into AI capabilities.
Addressing hardware diversity: Edge-native applications may span numerous nodes, operating systems, connectivity protocols, compute and storage requirements, energy and cost constraints, and compliance issues. Developers need to find ways to handle this complexity and support distributed, heterogeneous computing environments.
Protecting and managing distributed applications: Enterprises face new challenges as they seek to support advanced AI at the edge. Manageability is key to the large-scale deployment of AI, while security is essential at every step of the process.
II. Disadvantages of Edge Artificial Intelligence
1. Edge AI requires continuous training
Edge AI systems can be challenging because, like other AI models, they must be trained regularly and continuously—using only data from edge devices. This typically means creating datasets by transferring data from a large number of edge devices to the cloud, which can be quite complex, depending on available bandwidth and connectivity to the edge devices.
2. Edge AI requires additional security measures.
Security is also an area of concern, albeit in a different way. While edge computing can make systems more secure by keeping processing local, the infrastructure and devices themselves require their own security measures. These may include access control, traffic monitoring, data backup, antivirus and anti-malware measures, and even encryption.
3. Edge AI has always been slow (but this is changing).
The edge AI market isn't growing as fast as some had hoped. May says that since PJC initially invested in Deeplite and the broader edge AI space, the market has been "slower than expected" in terms of widespread adoption. He attributes this to the longer design cycles of devices that actually require edge AI, such as drones, phones, and cars. He says another significant part is the general lack of awareness that some tech companies even have this space.
