Much of Taiwan’s industry is being transformed by the introduction of AI technologies, particularly in the areas of manufacturing and healthcare.
Industry the world over is experiencing some radical changes, propelled by the continued development and application of AI technology across a wide variety of different fields. While in most countries everything from finance and banking to transportation and logistics is being revolutionized by AI, certain areas in Taiwan are undergoing more drastic transformations than others due to the country’s natural advantages in those sectors. These fields include robotics for industrial automation, semiconductor and hardware manufacturing, as well as for healthcare.
In these areas, as well as many others, Taiwan stands to reap the most benefits by integrating innovative software capabilities into its existing hardware infrastructure.
Taiwan’s economic miracle began with its shift to export-oriented manufacturing in the 1970s, and manufacturing is still a core driver of Taiwan’s economy today. Furthermore, an increasing number of Taiwanese manufacturers have begun leaning toward industrial automation, boosting demand for smart machinery – including intelligent robots – in production lines. Both government and industry have seized this opportunity to promote the development of AI-enabled manufacturing capabilities in Taiwan.
The Ministry of Science and Technology (MOST) in 2018 launched the NT$2 billion (US$66.59 million) Robot Makerspace initiative, which established smart robotics hubs in Taichung and Tainan within the Central and Southern Taiwan Science Parks. These centers offer co-working spaces and accelerators for startups to test their solutions.
According to MOST Deputy Minister Hsu Yu-Chin, the central and southern robotics hubs had hosted 71 startups as of the end of last year – 43 of them local or international AI-related startups from the ministry’s Taiwan Tech Arena program. Startups at the two hubs have been responsible for more than 99 new products or technologies, and in 2019 generated over US$400 million in overseas venture capital and international business collaboration opportunities.
Other industrial automation projects are being carried out by the government-backed Industrial Technology Research Institute (ITRI), including a self-taught robot technology that utilizes deep-reinforcement learning algorithms. Vincent Feng, general director of the Computational Intelligence Technology Center at ITRI, notes that the robotics currently used in manufacturing are semi-automatic and do not have AI capabilities. However, since production in Industry 4.0 can vary from time to time, Feng says, robots should be able to learn to recognize the shape of different objects and materials.
AI is not solely the province of startups and large tech multinationals. A growing number of Taiwan’s more established technology companies are beginning to explore AI-enabled robotics solutions for their client base as well.
Founded in 1973, Solomon Technology Corp. has expanded its scope of business over the years, going from the initial distribution of power generators and electrical components to the addition of LCDs, semiconductors, and batteries in the 1980s and 1990s. In 2009 the company became a distributor for Rockwell Automation, the world’s largest industrial automation company. It was at that time, says Solomon Chairman Johnny Chen, that the company began to get interested in what it could do with robots.
“I was going to a lot of trade shows, and all of the robots I saw were just doing repetitive tasks,” says Chen. “We started looking into vision systems and figured that this would be key to solving a lot of the more complex problems you find in production lines.”
The company began building up an R&D team to develop its own 3D computer vision system – a robot’s “eyes,” as Chen puts it. The machine vision is combined with AI deep-learning technology to give the robots the added benefit of a “brain” – the ability to recognize complex objects and patterns.
According to Chen, Solomon’s decision to delve into AI and machine vision has really begun to pay off. Its systems are now being applied by some of the world’s leading automobile, consumer goods, and e-commerce companies. Solomon’s 3D vision design has received international recognition as well, winning the prestigious Vision System Design Innovators Award in Chicago last year.
Taiwan’s push to introduce AI into manufacturing processes is not restricted to robotics. Among other ways in which high-tech operations can benefit from the integration of hardware and software capabilities is predictive maintenance. ITRI’s Feng says that knowing when factory equipment is likely to break down or needs to be replaced is especially important in the semiconductor and petrochemical industries, where an unplanned equipment failure can stall the pipeline and cause heavy losses.
Another function that is benefiting from AI integration is defect inspection and classification, ensuring the quality of production. This technology can be applied in the production of semiconductors, printed circuit boards (PCBs), D-RAM, and panels, says Feng.
“For example, in the PCB industry, we collected millions of images, then trained the AI model to distinguish between actual defects and false alarms,” Feng explains. These false alarms are common with the current system of automatic optical inspection and require human inspection for verification. ITRI’s AI inspection model, on the other hand, is much more accurate.
One inspection-equipment vendor for which ITRI developed an algorithm was able to raise the price of its product tenfold and has provided the AI-enabled equipment to major PCB manufacturers, Feng says.
Although the benefits of AI to manufacturing are clear, there is still some hesitancy among Taiwan’s manufacturers about adopting the technology wholeheartedly in their operations.
Richard DeVries, managing director of Geber Brand Consulting, says that “Taiwan is well-positioned to take advantage of AI,” but in his experience working with local B2B manufacturers, companies are not always ready to embrace it. He says that this reluctance is generally due to silo issues within an organization, leading to a lack of communication between departments and a persisting top-down hierarchy that’s common in Taiwanese companies. There may be impetus to incorporate AI from lower-level employees, but the top executives – the decision-makers – might be unaware of how it could benefit the company.
“We get around this by starting off with a general introduction of how AI in all its forms can apply to B2B manufacturers,” DeVries wrote in an emailed response to Taiwan Business TOPICS. “You need to show them via statistics, trends, and case studies that this is a huge and growing trend. Using fear of being left out, along with the opportunities AI can bring, often gets buy-in from the top level.”
Making healthcare smarter
AI for the medical field currently revolves mainly around disease diagnosis and drug screening. It relies on vast amounts of data to learn from and improve its accuracy, something which Taiwan is well-suited for. Since the mid-1990s, the health data of almost every Taiwanese citizen has been collected under the National Health Insurance program from around 30,000 hospitals and clinics and stored in the NHI central database.
Having access to such a massive amount of data is a real boon to developing AI and smart health systems. Of most immediate relevance to AI researchers is the cache of medical image data contained within the NHI’s database. This data can be used to train algorithms to detect certain conditions in patients, which doctors and technicians then confirm, increasing the efficiency and reliability of diagnoses.
Taiwan’s medical image data is currently being used by ITRI to assist Taiwan’s ophthalmologists in determining whether patients are suffering from diabetic retinopathy, a condition that can cause blindness if left untreated. The algorithm ITRI devised was fed around 10,000 images of retinopathic patients’ eyes and then installed in a smart funduscope – a medical device used to examine the interior structure of the eye.
ITRI’s Feng emphasizes the importance of this technology, given the high prevalence of diabetes among Taiwanese – 2.45 million people or 10.83% of the population are diabetic. However, most ophthalmologists are based in Taiwan’s big cities. Patients in more remote locations that have eye conditions like retinopathy are thus often referred to bigger hospitals in the city, a real inconvenience to diabetics. The AI-enabled funduscope thus allows doctors to better serve rural patients.
The Healthcare Lab at privately funded research organization Taiwan AI Labs is also using the NHI’s medical image data for its Malaria Diagnostics Project, which uses deep learning to more quickly diagnose the disease by locating the parasite that causes it in images of blood samples. The data is also used for the Lab’s brain cancer detection software Deepmets, which it launched in partnership with Taipei Veterans General Hospital in 2018.
That same year, AI Labs teamed up with Microsoft to launch a precision medicine platform called TaiGenomics. The platform’s algorithm processes and analyzes medical and genetic data which is stored in Microsoft’s Azure cloud computing platform to help doctors make better diagnoses more efficiently.
An ongoing issue with the NHI data is that while it is available for academic and research purposes, it cannot be accessed by private industry because of Taiwan’s stringent data protection laws.
The restrictions on data use indicate the government’s desire to align with data protection standards from places like Europe and the U.S. and to be a “good global citizen,” says Stephen Su, vice president and general director of ITRI’s Industry, Science and Technology International Strategy Center (ISTI). But the government has recognized the need to make the data more accessible, says Su, who also heads ITRI’s Office of AI Application Strategy.
Still, given the competition Taiwan faces from China and other regional players in the AI realm, the situation now is “like running a race that’s started before the gun even sounded,” Su says. “We need to have technology and controls in place so that data can be used in a more transparent way and that its value can be maximized as well.”
Joe Yeh, founder of the medical image AI startup aetherAI, says that his company has had to work around the data use restrictions by obtaining institutional review board (IRB) approval to collect data directly from local hospitals. This method is legal but time-consuming, he says
Yeh and the aetherAI team now use the data they collect from individual hospitals in Taiwan to develop AI for digital pathology. While this field has been around for nearly a decade, having the AI diagnostic systems in place to manage the image data and help create reliable prognoses takes a lot of the burden off of pathologists.
“AI can be very sensitive, it doesn’t get tired, and it trains on a vast amount of data,” Yeh says. He notes that “in pathology, you can now have double readings, where AI screens once and the pathologist screens a second time.” There are indications that Taiwan’s data regime may become more flexible in the future. According to a March article in local publication Digitimes, NHI Administration Director-General Lee Po-chang has said that the administration is beginning to discuss with life insurance companies the possibility of using NHI data to better tailor plans for their clients.