Artificial Intelligence (AI) draws a lot of attention and media space. But what is it, really? Elon Musk has recently spoken about AI in the terms of dystopian SciFi movies. Recently I’ve heard Eric Schmidt and Michael Dell talk much more positively about the potential for AI for solving human problems.
Most of the discussion about AI these days has little to do with replicating human brains with silicon circuitry. The core involves machine learning and neural networks. Recently two different organizations have sent me studies about the current state of AI. The first comes from ABI Research. The second from McKinsey Global Institute.
ABI Research AI to the Cloud
Artificial Intelligence (AI) will see a significant shift out of the cloud and on to the edge (aka on-device, gateway, and on-premise server). This will happen initially in terms of inference (machine learning) and then by training. This shift means a huge opportunity for those chipset vendors with power-efficient chipsets and other products that can meet the demand for edge AI computing. Edge AI inference will grow from just 6% in 2017 to 43% in 2023, announced ABI Research, a market-foresight advisory firm providing strategic guidance on the most compelling transformative technologies.
“The shift to the edge for AI processing will be driven by cheaper edge hardware, mission-critical applications, a lack of reliable and cost-effective connectivity options, and a desire to avoid expensive cloud implementation. Consumer electronics, automotive, and machine vision vendors will play an initial critical role in driving the market for edge AI hardware. Scaling said hardware to a point where it becomes cost effective will enable a greater number of verticals to begin moving processing out of the cloud and on to the edge,” says Jack Vernon, Industry Analyst at ABI Research.
ABI Research has identified 11 verticals ripe for the adoption of AI, including automotive, mobile devices, wearables, smart home, robotics, small unmanned aerial vehicles, smart manufacturing, smart retail, smart video, smart building, and oil and gas sectors and split across a further 58 use cases. By 2023 the market will witness 1.2 billion shipments of devices capable of on-device AI inference – up from 79 million in 2017.
Cloud providers will still play a pivotal role, particularly when it comes to AI training. Out of the 3 billion AI device shipments that will take place in 2023, over 2.2 billion will rely on cloud service providers for AI training – this is still a real-term decline in the cloud providers market share for AI training, which currently stands around 99%, but will fall to 76% by 2023. Hardware providers should not be too concerned about this shift away from the cloud, as AI training is likely to be supported by the same hardware, only at the edge, either on-premise servers or gateway systems.
The power-efficient chipset is the main driver of edge AI. Mobile vendor Huawei is already introducing on-device AI training for battery power management in its P20 pro handset, in partnership with Cambricon Technologies. Chip vendors NVIDIA, Intel, and Qualcomm are also making a push to deliver the hardware that will enable automotive OEMs to experiment with on-device AI training to support their efforts in autonomous driving. Training at the edge on-device is beginning to gain momentum in terms of R&D, but it could still take some take some time for it to be a realist approach in most segments.
“The massive growth in devices using AI is positive for all players in the ecosystem concerned, but critically those players enabling AI at the edge are going to see an increase in demand that the industry to date has overlooked. Vendors can no longer go on ignoring the potential of AI at the edge. As the market momentum continues to swing toward ultra-low latency and more robust analytics, end users must start to incorporate edge AI in their roadmap. They need to start thinking about new business models like end-to-end integration or chipset as a service,” Vernon concludes.
These findings are from ABI Research’s Artificial Intelligence and Machine Learning market data. This report is part of the company’s AI and Machine Learning research service, which includes research, data, and Executive Foresights.
McKinsey
AI Could Add $2 Trillion to Manufacturing Value, McKinsey Paper Says
Artificial intelligence for manufacturing is like the old BASF chemical company slogan that it does not make many of the things you buy, it makes them better.
As much as $2 trillion better, according to a McKinsey Global Institute discussion paper covering more than 400 AI use cases. In 69 percent of the use cases, researchers found that adding AI to established analytical techniques could improve performance and generate additional insights and applications.
Nearly a quarter of the use cases directly or indirectly touched manufacturing.
“Manufacturing is the second largest domain when it comes to potential in value creation (right behind Marketing & Sales,” said Mehdi Miremadi, an MGI partner and co-author of the paper. “Application of advanced deep learning models in manufacturing and supply chain have the potential to create $1.2-2 trillion in annual economic value.”
Two deep-learning neural networks offer the greatest promise in applying AI to manufacturing.
Feed Forward — information moves forward from the input layer forward through the “hidden” layers to the output layer. It has been enhanced by advances in computer power, training algorithms and available data.
Convolutional – Connections between the neural layers mimic the visual human cortex that processes images. These are well suited to visual perception tasks.
Of the two, Feed Forward Neural Networks is most applicable with wide applications in predictive maintenance, yield, efficiency and energy. The most value is derived when using AI with existing analytics tools.
“In predictive maintenance, neural networks improve the ability to incorporate and process a broader set of data, including unstructured data such as videos and images,” Miremadi said. Better algorithm precision and accuracy can result in better decisions. And there is the possibility of taking better advantage of live data.”
Small and medium size manufacturers, often left on the sidelines of Industry 4.0 technologies like AI, can do more than they might think, Miremadi said.
The cost of setting up AI applications is declining. Hardware sensors and actuators are much more affordable and reliable. And data systems and deep-learning algorithms are more accessible. This allows a choice of internal development or working with one of the many tech providers in the market.
Trackbacks/Pingbacks