How Edge AI represents the next major opportunity for enterprise

As artificial intelligence continues to advance and adoption rates grow across different industries, more sophisticated tech will be needed to power it. Does Edge computing fit the bill?

Artificial intelligence is without a doubt one of the hottest areas in the technology world right now. Whether it's writing novels, beating champion chess players or strong-arming cyber security plans, AI has limitless possibilities and is constantly making headlines all across the world.

For many, AI - along with the Internet of Things - marks the next major technological revolution. It's set to change the way we live and work in a plethora of ways, automating mundane processes and making life easier across the board. Capabilities include problem solving, speech recognition and machine learning.

What's more, artificial intelligence represents a major commercial opportunity. According to research from Markets and Markets, the AI sector will reach $190 billion in 2025. Meanwhile, IDC claims that spending on AI systems will top $57.7 billion and that 75 per cent of commercial enterprise apps will be powered by AI by 2021.

However, while artificial intelligence has already demonstrated a wealth of capabilities, it's still in the early days and has a long way to go. One of the next big advancements for this technology will be the rise of Edge AI, a technological approach that sees computational algorithms processed locally on devices as opposed to the cloud. We explore the benefits in practice.

Increasing device power

Today, a lot of artificial intelligence technologies rely on the cloud. But as they become increasingly powerful, new processing methods will be necessary. Paul Neil, VP of product management at XMOS, tells IDG Connect: "The issues associated with managing the control of smart home devices in the cloud, combined with questions surrounding the security and privacy of the data transmitted, suggest that a wholly cloud-based model may not scale as we usher in the era of ambient computing.

"As the complexity of our interactions with AI devices increases, along with the volume of data involved, the risk of the user experience being degraded by network connectivity or latency issues also increases."

Neil says that the ability to reduce the amount of data transmitted, to provide higher-level context to the user experience and to use smart devices when the internet is unavailable are all on offer with the shift to Edge-AI. He continues: "Edge-AI enables user experience designers to personalise the interaction with remote AI entities through the fusion of sensor data, and also removes the latency issues associated with cloud-based control.

"Total cannibalisation of the cloud may not be the ultimate endpoint, but a rebalancing of compute utilisation from core to edge in the IoT is a likely outcome of the deployment of Edge-AI capabilities."

Handling large datasets

Another challenge identified by technologists is handling the large datasets associated with connected devices. Figures from Statistica predict that the IoT ecosystem will grow to 31 billion devices worldwide in 2020, which will result in an explosion of data. But this is where Edge AI can help.

Peter Pugh-Jones, head of Technology at SAS, says: "Simply put, edge AI is artificial intelligence that operates close to the sources of the data that it's analysing. A distributed or phased AI deployment running at the edges of a network is more efficient at producing in the moment, actionable insights on data rather than with a single, centralised engine.

"Why is that? AI is essentially a highly advanced form of data analytics. Businesses can benefit hugely from the actionable insights generated by an effective analytics programme - but the sheer amount of data now being generated by the IoT presents a challenge."

"As the IoT grows, connected devices are producing many terabytes of potentially relevant information. How do you sort through that much information? How do you know where to look when it's constantly being generated in real time?"

He said the  answer is to have dedicated AI functions at the edge of the network, dedicated to a specific device or set of devices. "Each AI function will then detect patterns in the data generated by its assigned device. When it spots something relevant, it can then relay just that information back to a central engine," adds Pugh-Jones. 

Other benefits

Marcin Bednarz, product manager for MaaS, Telco and Edge at Canonical, says edge computing as been built up as the next step towards removing latency and to support emerging technologies like AI in working seamlessly with minimal latency.

"This environment is very valuable, with its requirement to collect and process vast amounts of data in near-real-time, with a very low level of latency. It can lower connectivity costs by sending only the most important information, as opposed to raw streams of sensor data and with the huge stress that these activities will place on networks, bandwidth optimisation tools such as edge can help to dodge critical bottlenecks on the network," says Marcin Bednarz,

"For example, a utility with sensors on field equipment can analyse and filter the data prior to sending it and taxing network and computing resources. In the coming years, this will support businesses in increasing agility, driving innovative new use cases at the edge and reducing the costs linked to creating these new services."

Peter Parkanyi, from Red Sift, agrees that this technology has an array of business benefits. He explains: "Edge AI is more complete for user privacy and enables a company to get around tricky regulations such as GDPR, by simply not storing user data. At the same time, it helps the organisation to reduce storage/computation costs by outsourcing these to users.

"However, because of this disconnect, training the AI becomes more difficult because the company still needs access to a training data set. It also becomes more difficult to measure the output or benefits of the AI to the business.

"For example, what do users search for, what are the most important things to them, how do we improve user experience, how well does the AI actually perform? The privacy-preserving methods for using Edge AI are offset by the increased cost of engineering."

There's no denying that artificial intelligence offers many great opportunities, and it's likely that we'll continue to see this area advance greatly over the next few years. But what's clear is that there are challenges - particularly how this technology is processed and managed - and Edge computing appears to be an effective solution.