We are entering the era of intelligent things where smart devices seep into every aspect of our lives. As AI continues to evolve and IoT devices proliferate, the two transformative technologies converge, forming something called AIoT (the artificial intelligence of things) or some may refer to it as the next generation of IoT.
Sounds a bit far-fetched? Not at all.
Home appliances, factories, healthcare equipment, surveillance cameras, smartphones, drones and autonomous vehicles are all starting to adapt to AIoT technologies.
For example, Sharp recently released an AIoT augmented kitchen that allows customers to talk and consult with it about preferred cooking time and calories setting. It is capable of giving suggestions based on information like previous cooking records, seasons and weather, and educate itself by learning the family’s food preference and eating habits.
According to Gartner’s predictions, 1 million new IoT devices will be sold every hour by 2021, and, as one would expect, they will be augmented with artificial intelligence and become a network of learning and thinking objects, each with a brain of its own. This shift from stand-alone smart devices to an intricate network of collaborative intelligent things is what the future of AI entails.
Big players gearing up for big opportunities
Though AIoT is still at a budding stage, its applications are already boasting enormous business potentials. Many global players including China’s tech giants, Baidu, Alibaba, and Tencent (BAT) are all making big moves to ensure their place in the forthcoming expansion of AIoT technologies.
Last November, Xiaomi and Baidu announced their collaboration to tap into the AIoT market at the AIoT Summit in Beijing, which affirmed their commitments to accelerate deployments in the AIoT sector.
The rise of edge computing
In the age of AI, increasingly more data is being generated and gathered from different sources—whether it be a smartphone, drone, sensor, robot, or autonomous vehicle—and cloud computing is under the strain of meeting massive data computing demands of these devices and applications. This gave rise to a new paradigm called edge computing in which information is processed closer to its source instead of sending it to data centers or clouds. Bringing computing to the edge of the network and allowing more important data to be processed and analyzed in real-time is crucial for the development of many applications including VR, IoT, automated vehicles.
This decentralized and distributed way of data processing is not new, but it’s only recently come into more demand as latency becomes more important.
A Taiwanese startup making headway
Back in 2015 when deep-learning—a core technology behind AI—started to be widely applied in various areas, a group of researchers from Taiwan National University saw a problem behind the excitement. “Coming from an engineering background, the privacy aspect of machine learning is a top concern for us. Deep-learning requires an immense amount of computational power and data. The data, visual and non-visual, are all being uploaded to the cloud,” Tammy Yang, the co-founder of a Taiwan-based deep-learning startup DT42 told TechNode. Indeed, cloud computing entails the transfer of data over a network to the cloud, which can have serious privacy and security ramifications. But because of the difficulty and complexity associated with edge computing, “most AI applications at the time resorted to cloud computing for data processing and analytics, but we knew this needed to be changed.”
Seeing the unfilled gap, DT42 decided to simplify the process of applying and deploying AI technology to local devices and equipment, making it accessible to more businesses. How? You guessed it, through edge computing. Essentially, they are squeezing deep-learning models into edge devices, making AI lighter and cheaper. By doing so, cloud and edge computing can work together to share the load—thereby reducing network latency, lower data management cost, and minimize potential security risks. The startup has already gained recognition from Zeroth.AI, Asia’s first AI accelerator, in 2016.
Asia’s competitive edge
Over the past two years, AIoT has been a hot topic, especially in Taiwan. There’s a reason for it: situated in Asia, the world’s hardware manufacturing hub, gives the island the right environment to push forward AIoT.
In fact, Asia’s prowess in hardware manufacturing is a unique competitive edge in the AI revolution. DT42 chose to build their team in Taiwan because a large part of edge computing, an important thread of future AI, has to do with the integration of AI’s hardware and software.
“Asia—China and Taiwan in particular—are specialized in hardware and software integration,” said Yang. A lot of hardware manufacturing factories are located in China and Taiwan region and “naturally, headquartering in Taiwan gives us an advantage not only because there is a rising demand in the region, but it would also be easier in terms of finding the right talent and the right hardware partners.”
The Taiwan-based startup is entering other markets such as China, Hong Kong and Japan where they see similar opportunities.
First to hit the wave
AIoT applications are starting to emerge in different industry verticals, but some are undergoing the disruption earlier than others. DT42 is seeing demand coming from two industry verticals: surveillance and manufacturing—the two industries that are generating an immense amount of data.
A swarm of new AI startups are focusing on surveillance because AI has great potential in enhancing surveillance systems, face-recognition technology, and data-analyzing capacities through data gathered from security cameras and motion sensors.
The manufacturing industries, where cameras and sensors are widely deployed, are looking to AIoT for factory automation. Most recently, Andrew Ng’s Landing.ai announced a partnership with Foxconn to bring AI to manufacturers.
When exactly is the wave going to hit?
The era of AIoT has been looming on the horizon for some time now, but why hasn’t it arrived already? Yang mentioned three key factors that have stunted the development of AIoT.
First, data collection needs time. Take, for example, IoT technologies and sensors used in the agriculture industry need at least a year to have four seasons of data, and even more years of data is necessary to extract meaningful information. The time for collecting big volumes of data is impossible to cut short. Second, data collected from AIoT is immense. There needs to be a technology that can read, analyze, and understand the data. A lot of applications are beginning to emerge, cloud computing is one of them. But currently, there are limitations to what cloud technologies can accomplish. Third, data-analyzing tools currently available are not advanced enough just yet to push forward AIoT in full thrust.
Though IoT is an important driver for the development of edge computing, it is also true that edge computing is indispensable to AIoT deployment. “From a technical point of view, this will be a huge leap forward for AI development,” Yang said, “We are quite optimistic. In 3 to 5 years AIoT industry applications will be at a mature phase where there will be significantly more applications emerging.”
Embracing the change
Although it might seem that AIoT and edge computing technologies are still in their early days, that is starting to change.
“It was challenging at first especially in communicating the idea of edge AI. The concept and its importance weren’t widely understood at the time.” Recognizing today’s cloud computing technology has its limitations and it simply can’t meet the needs of emerging AI applications, industries are slowly waking up to the emerging tech trend. “Fortunately we are past the hard part. Big tech companies like Nvidia, Intel, Qualcomm have all been promoting the concept. There was an apparent change in attitude toward edge AI near the second half of 2017. It then became much easier in terms of communication with the government, clients, and investors.”
AIoT is important to industry structure of Taiwan, China, and other manufacturing hubs in Asia in that it will create a huge demand in hardware sensor, a processing unit, and more. It will most certainly create new demand and become a force driving both the growth of manufacturing industry as well as a swarm of hardware-based SMEs in Taiwan.