Industry, government, and academia are all working together to create a vibrant AI ecosystem in Taiwan, while recognizing that some major challenges remain.

Ask anyone in Taiwan’s business or technological circles these days what they think will be the next big thing for Taiwan industry, and the answer is almost certain to be “artificial intelligence.”

There is good reason to believe that Taiwan can make the most of this new and dynamic technology, which international consulting firm McKinsey estimates will add US$13 trillion in global economic output by 2030. Most importantly, the country possesses a large pool of highly capable but relatively inexpensive engineering talent. It is also home to probably the strongest and most comprehensive ICT ecosystem in the world, especially for semiconductor manufacturing and IC design.

In addition, Taiwan’s open-source data policies offer a treasure trove of useful data for AI algorithms to learn from. Furthermore, Taiwan’s government and industries are beginning to explore how its traditional focus on tech hardware and manufacturing can be leveraged to take full advantage of what AI applications have to offer.

But some thorny challenges to Taiwan’s aims of becoming an AI innovation hub still remain. For one, while Taiwan’s high-quality workforce is praised by local and international businesses alike, the supply of domestic talent is insufficient to meet the rising demand, especially for those with expertise in such a new technology as AI.

In the past, Taiwan’s dominant tech hardware sector tended to pull most fresh young talent into that sphere. Positions in long-established local companies like Taiwan Semiconductor Manufacturing (TSMC) and Mediatek were traditionally more lucrative than those in the software engineering field.

Ethan Tu, founder of Taiwan AI Labs, a privately funded research organization based in Taipei, says that he encountered a lot of negativity when he first relocated to Taiwan from the U.S. in 2017.

“People said ‘oh, a software company will never survive because Taiwan’s market is too small’ and ‘Taiwan doesn’t have enough AI talent,’” says Tu, who is also the founder of the popular Taiwanese bulletin board system PTT. “At the beginning, only [Minister without Portfolio] Audrey Tang and [Minister of Science and Technology] Chen Liang-gee supported me,” he recalls.

However, Tu and others observe that the mindset has gradually begun changing as more and more companies look to AI as the future of tech on the island. For cultivating the talent pool, the main issue now is the pattern of brain drain that Taiwan is stuck in. Engineers continue to leave the country in search of better-paid options – mostly in China.

The issue of talent insufficiency is not lost on Taiwan’s leadership. Last May, Premier Su Tseng-chang announced plans to train 10,000 new AI specialists each year. Education about the technology will begin as early as elementary school, the premier said.

There are also ethical questions surrounding the adoption of AI. Some of these are universal, such as how to avoid bias in AI models or deal with job loss and inequality associated with AI-driven automation. Others are particular to Taiwan. How, for example, can the National Health Insurance Administration’s cache of healthcare data be properly utilized for AI research or commercial purposes while still preserving privacy and adhering to Taiwan’s data protection regulations?

Challenges aside, the enthusiasm for realizing Taiwan’s potential in AI is palpable. Enormous resources have been poured into the development of AI and related technologies since 2016 as part of the Tsai administration’s push to reform Taiwan’s economy for the digital age. Multiple government ministries have introduced sweeping, multi-million-dollar AI-focused schemes, which involve the participation of industry, academia, and both publicly and privately funded research institutions.

The largest investment by far has come from the Ministry of Science and Technology (MOST), who saw the growing global trend in AI. Its five-year AI Taiwan@MOST program, launched in 2017, has put up around NT$16 billion (US$500 million) and encompasses five different national projects.

MOST’s projects are mainly centered around supporting academic research of AI technologies, as well as encouraging academia-industry collaboration on a range of areas deemed important to Taiwan’s AI development.

One such project is a shared-use AI and Big Data cloud-computing platform, powered by the Taiwania 2 supercomputer at the National Center for High-Performance Computing (NCHC). Launched in May 2019 and expanded for commercial use a few months later, the platform – called the Taiwan Computing Cloud (TWCC) – has been accessed by several startups to develop their AI-focused products and solutions.

Other projects under the MOST aegis have included a NT$4 billion (US$133 million) initiative to help Taiwan’s robust semiconductor industry accelerate its development of AI edge technology, as well as four AI research innovation centers spread among various universities around the island. These facilities, which focus respectively on AI core technologies, AI for medical applications, AI for manufacturing, and AI services, are intended for both research and development purposes, as well as talent cultivation. The centers have so far produced hundreds of academic papers and successfully completed 58 cases of technology transfer.

According to MOST Deputy Minister Hsu Yu-chin, funding for academic research is essential to creating the tech landscape needed for the intelligence era. After all, he says, all new technologies begin with the initial scientific research, later moving on to industrial production.

Yet some in Taiwan’s nascent startup scene have pointed to a lack of flexibility in funding under the program. Joe Yeh, founder and CEO of the medical AI startup aetherAI, says that his company has published three peer-reviewed journal articles – something companies are not normally obligated to do. Nevertheless, because aetherAI was not established by an academic at one of Taiwan’s universities or research institutions, it is not eligible to receive MOST funding.

“In Taiwan, funding for scientific research is for academic institutions only,” says Yeh. “That is a different model from the U.S., and I think that for AI it’s a big misstep. A lot of companies have the required technologies to push forward AI research, but we have no way of getting funding from MOST.”

For his part, Deputy Minister Hsu sees the AI Taiwan@MOST program as a good first step, one that has been followed by additional funding from a few other ministries.

Those subsequent efforts were carried out under the Executive Yuan’s “Taiwan AI Action Plan” introduced in January 2018. The Plan, which is set to last until 2021, contains a broader set of objectives than AI Taiwan@MOST.

One of the main goals of the Plan is to enlarge Taiwan’s existing pool of AI engineering talent. In addition, it also sets out to expand and capitalize on Taiwan’s world-leading position in the semiconductor industry, remold Taiwan into an industrial hub, liberalize regulations to enable the development of innovative technologies, and use AI to transform industry in Taiwan.

The Executive Yuan has also called on industry – particularly large multinational tech firms – to help lead the charge. Giants such as Microsoft, Amazon, and Google have set up large R&D centers on the island over the past few years. These facilities work closely with government and local universities to train and employ Taiwan’s top-tier engineering talent.

In 2018, Microsoft established its AI R&D Center in Taipei. The NT$1 billion (US$33 million) project was launched in partnership with the Ministry of Economic Affairs under its Global R&D Innovation Partner Program. The Center employs a team of 100 local engineers, who are assigned to work on a combination of six independent projects covering the subfields of computer vision, user intention, and vertical industries.

Michael Chang, the Center’s director, says the decision to locate the facility in Taipei was easy to make. Taiwan boasts some of the highest-ranked science and technology universities in the world, producing some 10,000 computer science graduates and 25,000 electrical engineers every year. In fact, 30% of the Center’s employees were recruited directly from Taiwan’s universities.

The island’s compact size, population density, and excellent ICT infrastructure were also major factors for Microsoft. “Taiwan has a really unique position,” says Chang. “The supply chain is so complete and the turnaround time [on production] is so short,” he says. “This is important from an innovation perspective.”

Startup island

With the help of generous government support and some newly introduced regulatory frameworks, a flourishing startup space centered around AI products and solutions has begun forming in Taiwan in recent years. Several of the startups that have emerged from this scene have found success developing innovative vertical applications in a range of different areas. This growing community is complemented by a system of supporting institutions from both the public and private sectors.

The best-funded of Taiwan’s AI startups is Appier, which has raised over US$160 million since its founding in 2012. The company uses artificial intelligence to assist its clients with their digital marketing strategies. Appier’s success in cornering this market in the Asia-Pacific has earned it a place among Forbes’ top 50 AI companies worldwide.

Before joining the company in 2018, Min Sun, Appier’s U.S.-educated Chief AI Scientist, spent most of his career on the academic side of AI development, focusing his research on natural language processing and computer vision.

Sun says that Appier founder Yu Chih-han’s goals of building one of the best AI-based R&D teams in Taiwan, serving customers with innovative AI solutions, and seizing business opportunities in the Asia-Pacific aligned with his own objectives at that time. Joining Appier also allowed Sun to take advantage of his background in multimedia analysis using deep-learning techniques.

Since coming on board, he and the rest of his technology team have led the company’s efforts in pushing for AI-driven business transformation for its more than 1,000 clients. The team uses supervised machine learning models, which allows it to account for consumption and engagement habits across different countries and cultures.

Considering that digital marketing is experiencing one of the most rapid digital transformations of any field, it seems a likely choice for an AI-related startup to lock in on. Indeed, in addition to Appier, around seven other startups in Taiwan are now competing in this area.

One of these startups is iKala, originally an online karaoke and live broadcasting platform that years ago decided to first switch to cloud streaming for enterprises. Sega Cheng, iKala’s founder and CEO, says that after the company made this pivot, “that’s when we started doing serious business.”

Cheng, who was trained in AI programming at Stanford University, says that later applying the technology to the company’s digital marketing solutions made sense. The decision was a wise one. iKala now partners with some of the world’s biggest tech companies, including Facebook and Google. It has raised around US$13 million so far and is preparing for Series B funding in the fall.

Other startups are looking elsewhere for developing AI applications, such as creating solutions for use in the health and medical, security and surveillance, retail, and finance sectors.

Niche opportunities exist as well. Charles Chin, CEO of WritePath, an AI-enabled translation service for the financial industry, says that he and his co-founders initially sought to partner with academia to develop the text-based AI solution. “However, when we brought the idea to them, they told us ‘you guys are very brave,’” Chin says. “It’s a part of AI technology, yes, but it is not the major focus.”

But Chin had observed the growing demand among Taiwanese companies to translate financial statements from Chinese to English in order to attract foreign investment. Besides, the company’s client list was expanding and the particular market they found themselves leaning toward seemed to lend itself well to incorporating AI.

Translating financial documents, says Chin, requires a deep knowledge of the special terminology used in those documents in two different languages. In addition, financial documents contain repeated terms and phrases for AI to learn from.

WritePath’s algorithms are designed to work in collaboration with human translators to create the final product for its clients. These algorithms are fed all the data from previous cases the company has completed, as well as databases and public information that Chin and his team gleaned from around the world. The AI performs the first translation, targeting repeated words, glossary terms, and previously used phrases, and the human translator polishes it.

Appier, Writepath, and iKala all emphasize that their AI solutions are “human-centered.” Asked what this means, Cheng of iKala stresses the involvement of people in AI development and applications, as well as the creation of ethical, humane AI.

He cites the example of one of iKala’s products, an innovative picture-as-a-service (PicaaS) technology, which he calls a “deepfake for good.” The software automatically edits product pictures – removing background, promotional text, and unnecessary overlays – allowing companies to widely circulate clean photos of their products.

However, Cheng says that after initially launching the platform, the company received some complaints that it could potentially be used to infringe on image owners’ intellectual property rights. Cheng and his team then went back and re-trained the software to recognize and reject copyrighted images.

“That’s the kind of responsible AI we want to be working on,” says Cheng. “Putting humans in the equation – not just stealing and not just replacing people.”

Getting businesses on board

One of the biggest challenges to incorporating AI and other software-oriented solutions further into Taiwan’s industries is the slow pace of digital transformation on the island. Cheng of iKala notes that companies need to have a strong digital foundation before they can begin adopting AI technologies.

But many companies in Taiwan currently lack the necessary comprehensive understanding of how, step by step, to digitally transform a business. These companies, Cheng says, need a clear “digital roadmap” that would put all stakeholders on the same page.

Cheng and his team at iKala have seized on the current gap in digital literacy to come up with a framework to help businesses successfully navigate a digital overhaul. Entitled the DAA (Digitalization, Analytics, Application) flywheel, this framework is not only useful for guiding clients through the process, but also helps iKala determine whether AI is even necessary for an individual company’s digital transformation.

“Instead of just boasting about or selling you AI services or solutions, telling you how magical it is, what we do is to first help you define clear business goals,” says Cheng. “In our past experience, there were some circumstances where we held a workshop or looked at a business owner’s goals and concluded that they didn’t need AI at all,” he says.

Michael Chang of Microsoft agrees that digital transformation is important and notes that company culture is the leading factor preventing businesses from taking the leap. He says that the culture at Taiwanese companies has historically been a bit more cautious and conservative to make such a big investment in the company’s future.

“I hope to see that change, so that governments and businesses can soon start to see the value of software,” says Chang. “Software, including AI, is the key to their success.”