by Kay Firth-Butterfield and Beena Ammanath*
Imagine you want your new smart thermostat to quickly turn up the heat so that your house will be warm after your get home from work on an unusually cold day. You connect from your smartphone and ask it to act. You won’t know it, but that action may take several seconds as it moves your request to the cloud and receives instructions back.
Now imagine the self-driving car you’re in suddenly senses a dog running into the road in front of you. The car needs to react in milliseconds to avoid a disaster. That kind of reaction requires edge artificial intelligence (AI) – technology that can make a decision at the closest point of interaction with the user, in this case, the car’s sensors. It’s the definition of a split-second decision.
Data in motion
With today’s Internet of Things (IoT), data is always in motion. It flows from legacy systems to the cloud, all the way to edge devices and beyond an organization’s systems to partners and customers. Answers need to be delivered in real-time and so it’s not always effective to use centralized computing power when data can be processed via edge devices. A self-driving car doesn’t have time to wait for a decision to be made in the cloud when it has mere seconds to react.
Vast amounts of data can be fed into AI algorithms on the edge wherever the device happens to be – and the benefits are plenty. Data in motion can deliver critical patient information to doctors, shorten lines in amusement parks, alert power companies of a potential outage, and make a self-driving car react in time to prevent a tragedy.
Edge AI allows a device to make these decisions on its own, at the device level. It doesn’t necessarily have to be connected to the internet to process the data. Consider a watch that can monitor your sleep patterns, but instead of pushing the data into the cloud for storage and processing, it records the data for processing on the watch itself.
Edge-enabled AI devices also include video games, smart speakers, drones, and robots. Security cameras can also be edge-enabled – a camera on a factory floor that looks for product defects during manufacturing can quickly identify which products to immediately pull. Edge AI can also be used to analyze images for emergency medical care, when speed can save lives. The closer the processing capabilities, the quicker the response time.
Although edge technology will not replace the cloud, user data that belongs only to you – your sleep patterns or gaming data, for example – can be processed in an edge-enabled device. This decentralization of data addresses the issue of privacy, a significant concern in the IoT market. Edge AI can provide convenience without compromising privacy. And, in some cases, it can be much cheaper – one company is currently developing voice-activated home appliances such as washing machines and dishwashers using tiny microprocessors that cost a few dollars apiece.
The trade-off is a less ambitious bot: A voice recognition AI for a coffee maker only needs to recognize about 200 words, all related to the task of brewing coffee only. Think of it this way, says Wired reporter Clive Thompson: “I don’t need light switches that tell bad jokes or achieve self-awareness. They just need to recognize ‘on’ and ‘off’ and maybe ‘dim’. When it comes to gadgets that share my house, I’d actually prefer they be less smart.”
In addition to quicker and less expensive processing, edge AI doesn’t require an ever-expanding internet. With the rapid growth of IoT, there is now a vast amount of data being sensed and produced at the edge – Statista estimates the figure will reach almost 80 zettabytes by 2025.
This is so enormous that it is not technically feasible to use the bandwidth of today’s internet to transfer the entirety of this data from edge devices to cloud servers for storage and processing. Even if the bandwidth was available, there would need to be enough data center resources available to handle all of the data. Less required bandwidth translates into cost savings. Around 10% of enterprise-generated data is created and processed outside a traditional centralized data centre or cloud. Gartner predicts this figure will reach 75% by 2025.
Balancing risk and reward
One of the most vexing problems in the IoT world is the fact that large numbers of people that can’t afford the devices or that live in rural areas where local networks don’t exist may be unable to participate in this remaking of our everyday world. A history of limited network capacity can become a vicious cycle. Edge networks are not simple to build and can be expensive. Developing countries may fall further behind in their ability to process data through edge devices that need updates. The growth of edge computing, then, is another way in which structural inequality could increase, particularly as it relates to the accessibility of life-changing AI and IoT devices.
Another risk with edge AI is that data may be discarded after being processed – by its very nature “at the edge” means it may not make it to the cloud for storage. The device may be directed to discard information to save costs. While there are certainly disadvantages with central processing and storage, the advantage is that the data is there if and when needed.
A huge stream of data about an empty road may not seem important if it’s just you and your autonomous vehicle on it but think again. Much can be learned from data about that empty road, including information on road conditions and how the vehicle, and others like it, behave under those conditions. Finally, a clear business case must be closely scrutinized when it comes to edge computing to ensure the costs of the network balance with the value created.
Still, despite inequalities or lost data, and with the coming advancements in 5G technology and less costly processing chips, it’s easy to see how being “on the edge” could be here to stay – whether it’s your self-driving car or your coffeemaker that gets you ready for your commute.
*Head of Artificial Intelligence& Machine Learning; Member of the Executive Committee, World Economic Forum and Executive Director, Global Deloitte AI Institute& Trustworthy AI/Ethical Tech Leader, Deloitte
**first published in: www.weforum.org
By: N. Peter Kramer
