Edge AI: The Disruptor We Didn't See Coming

Technology infrastructure is on the move, and it's moving to the edge.

This is where applications can be closer to users and customers in a more decentralized world and is how big organizations deliver consistency of performance across their operations.

Data center providers are investing in this area as they provide computing power and storage for their enterprise customers, and—in perhaps one of the biggest moves of all—AI is also going there.

The combination of edge and AI is emerging as one of the most powerful themes in 2024 as organizations look to optimize their developing AI capabilities and understand they need computing power close to their growing army of IoT sensors and endpoint devices, now key data sources.

Today’s enterprises need real-time applications in areas such as autonomous vehicles and virtual gaming. They now have a suite of technologies and ecosystems to deploy that combine hardware accelerators, 5G, memory and storage and synthetic data.

Gartner is already talking about Edge AI as one of the most exciting technology developments at the intersection of AI, IoT and big data.

It will deliver benefits in terms of reduced latency, reduced costs and complexity, better privacy preservation and a greater ability to scale.

Real-time analytics

In a recent report, “Innovation Insight for Edge AI”, Gartner analysts observe that the trend is being "driven by the need for real-time analytics and stringent data requirements."

“Both large vendors and startups are recalibrating their strategy for Edge AI with two distinct architectural approaches: Cloud-Out and Edge In,” the report says.

“While the vendor ecosystem is nascent and fragmented across these approaches, the innovations across these approaches will define the future evolution of Edge AI.”

“As demand grows to process this data at the point of creation, applications, AI training and inferencing will need to move closer to edge environments near IoT endpoints.”

Although most enterprise data is currently generated inside centralized data centers or cloud regions, Gartner forecasts that this pattern will change dramatically in the future. By 2025, 75% of data will be generated outside these centralized facilities.

“As demand grows to process this data at the point of creation—in order to gain real-time insights—applications, AI training and inferencing will need to move closer to edge environments near IoT endpoints,” the report says.

“We forecast that there will be 11.7 billion IoT devices by the end of 2025, and their capabilities will expand as compute, security and bandwidth technology evolves, particularly with the more pervasive adoption of 5G…This will create a rich foundation for edge analytics capabilities to be widely available.”

While there is broad-based interest across industries, Gartner sees early adoption in the automotive, healthcare, retail, transport and process manufacturing sectors, as well as in the packaging of consumer goods, food and oil and gas.

A manufacturer, for example, can implement Edge AI into its processes for predictive maintenance, quality control, and defect detection.

Through AI coupled with localized data analysis from smart machines and sensors, they can make better use of real-time data to reduce downtime and improve production processes and efficiency.

The explosion in video streaming and IoT devices for image capture, pattern recognition, object counting, and facial authentication is combined with real-time analytics to play into the Edge AI story, where improved capabilities can be achieved at a lower cost.

Quality assurance on manufacturing lines, crop and animal monitoring and smart space management are all use cases that harness Edge AI.

Vulnerabilities

However, there are some risks, as Gartner points out.

“The distributed nature of edge environments makes them more vulnerable to security risks,” the report says.

“AI services in general are vulnerable to newer forms of attack, such as data poisoning and model reengineering, where Edge AI services are also vulnerable.”

To minimize risks, improve data governance and be rigorous in rooting out software design flaws and device misconfigurations.

Then, there is the question of processing power. AI needs power behind it, and the edge often suffers from under investment in this area, which is another reason behind the major investments data center providers are making in new geographies.

Agile software engineering is also critical, and that is why enterprises that choose the Edge AI route need to invest in their DevOps teams and tools.

Of course, there are alternatives, and this is why some organizations are choosing to have their AI deployments on-premises.

Another option is Cloud AI, which is likely to form part of the AI ecosystem with edge AI as the landscape matures, with the cloud used for AI training and the edge specializing more in inferencing.

However, many of the new real-time AI applications won't occur in the cloud. They'll be at the edge, which is why it is so important right now. 

Image credit: iStockphoto/avgust01