Taiwan Sees Big Potential in Artificial Intelligence

But capturing the possibilities will require a difficult shift in focus from hardware and assembly to software, plus a change in regulatory mindset.

Long before there was Facebook, YouTube, or Instagram, Taiwan had an active social network of its own: the PTT bulletin board system (BBS). National Taiwan University student Ethan Tu founded PTT in 1995 as an open forum for online discussion, and it quickly became Taiwan’s most popular social network, attracting millions of users.

Today, PTT looks dated, like a relic from the era of dial-up internet connections. But 22 years ago, the concept of a social network for Chinese speakers to discuss everything from politics to the weather was fresh. Why didn’t PTT commercialize?

“We didn’t think software could be a business,” Tu told Taiwan Business TOPICS in an interview. “It’s been a problem for Taiwan for years.”

Entrenched hardware makers have held the technology industry captive here as software companies have conquered the world. Firms like Google and China’s Alibaba are now poised to take the lead in the emerging field of artificial intelligence (AI). Taiwan has no software company of that caliber.

But the industry is so new that Taiwan still has ample room to capture market share. “AI at this point is like the internet in 1995,” Tu says.

After working for a decade in the United States, including for several years at Microsoft, Tu returned home in April to serve as the head of the newly established Taiwan AI Labs. “Taiwan has lots of AI talent, but some of the best engineers have either been pulled into hardware or have left the country,” Tu says.

AI Labs intends to provide Taiwan’s software engineers with a proper outlet for their talents at home, Tu says. He notes that Taiwanese software engineers helped create Microsoft’s Cortana Intelligence Suite AI software. And in a ranking last year of the AI capabilities of universities in Asia, NTU placed first.

The lab plans to approach issues in AI development from the bottom up. It will collect user experiences and data, create applications, and publish its research. AI Labs plans to carve out a niche for itself in under-served AI areas, Tu says.

Industry observers expect AI to surge over the next decade. Research firm Research and Markets forecasts global AI revenue to grow from US$643.7 million in 2016 to US$38.8 billion by 2025. Businesses able to apply AI effectively could boost profitability by 38% by 2035, according to a June report by Accenture.

To capture those opportunities, Taiwan should anchor AI development on its existing strengths, such as semiconductor manufacturing, says Stephen Su, general director of the Industrial Economics and Knowledge Center (IEK) at the Industrial Technology Research Institute (ITRI).

In August, the Ministry of Science and Technology (MOST) announced that it would invest NT$4 billion (US$131.7 million) in the IC industry over the next four years, focusing on AI applications. “With many international brands having entered AI development, the industry is expected to boost demand for efficient computing and energy-efficient ICs,” MOST Minister Chen Liang-gee was quoted as saying by the Central News Agency (CNA).

In total, the government has allocated NT$16 billion (US$528.7 million) over the next five years to develop an “AI innovation ecosystem” in Taiwan, according to an August press release from the Executive Yuan. Projects slated for completion by year-end include the launch of an AI innovative research center as well as AI Robot Maker Space facilities at the Central Taiwan Science Park and Southern Taiwan Science Park.

Taiwan is also eyeing the autonomous vehicles segment. In September 2016, ITRI announced it would partner with U.S. chipmaker Nvidia to develop driverless car solutions. ITRI is contributing image sensors and deep-learning technology to the project.  Deep-learning software acts like an electronic version of the neocortex – the part of the brain responsible for cognition, language, sensory perceptions, spatial reasoning, and initiating motor commands. “The software learns, in a very real sense, to recognize patterns in digital representations of sounds, images, and other data,” according to the MIT Technology Review.

ITRI’s autonomous vehicle prototype can already drive fixed routes. The vehicle has been tested at slow speeds on routes taken by shuttles within recreational areas and by patrol cars near high-speed railway and subway stations. ITRI and Nvidia aim to eventually deploy AI-powered self-driving shuttles at Taiwan’s airports and train stations.

Meanwhile, under the Asia Silicon Valley Development Plan (ASVDP), the government will focus heavily on autonomous vehicles next year, including assembling a “national autonomous vehicle development team,” according to a September report in DigiTimes.

Rising stars

Thus far, the most exciting developments in Taiwan’s AI field have not come from large tech firms or the government. Instead, a few startups are leading the way. That’s unusual for this market, where neither traditional industry nor government policy has tended to be friendly to tech upstarts.

It helps if the startup does not compete directly with local heavyweights and focuses primarily on the global market. That’s been a successful strategy for AI platform developer Appier, which serves roughly 1,000 global customers and has offices in 12 countries across Asia. In August, Appier announced it had received a Series C investment of US$33 million from a consortium including Japan’s SoftBank and Line, Korea’s Naver, Singapore’s EDBI, and Hong Kong’s AMTD Group. Appier has now received overall investment of US$82 million.

“Appier is a leading AI startup in Asia,” says Jamie Lin, founder of the Taipei-based accelerator AppWorks. “They build highly accurate AI models that predict consumer behavior, so companies [that use Appier’s services] know which ads to target consumers with across different devices.”

Appier did not respond to a request by Taiwan Business TOPICS  for an interview. But in an August press release, the company says that its first product, the CrossX Programmatic Platform, “brought AI into digital marketing.” Its second product, Aixon, is the data intelligence platform to which Lin referred. Appier says that its revenues have doubled since it completed Series B2 funding last November.

“Appier has enabled a growing number of brands to develop consumer-adaptive marketing solutions for the digitally savvy Asian market,” said Chu Swee Yeok, CEO and President of EDBI, in the press release. “Appier’s development activities also synergize well with Singapore’s AI.SG initiative to power, build and augment the local AI capabilities and ecosystem.”

Meanwhile, Umbo Computer Vision (CV) is applying AI technology in surveillance cameras, a segment where Taiwan has long been competitive. In April 2016, the company raised US$2.8 million in a seed round of funding led by AppWorks. Other investors included the Taipei-based VC firm Mesh Ventures and the Taiwanese manufacturers Wistron and Phison.

Founded in 2014, Umbo CV has shipped its system, which includes surveillance cameras and a cloud-based platform, to clients in the United States, Europe, and Middle East. The system has been in mass production for over a year.

Umbo CV’s technology is a quintessential example of how AI can assist humans, rather than replace them, says AppWorks’ Lin. “If you think about security cameras for a large building, and all the angles they display, it’s difficult for a person to constantly watch all of the scenes. The cameras serve more to record crimes or accidents than prevent them from happening,” he says. “Umbo CV’s AI-powered cameras can prevent something bad from happening by detecting risks and then alerting the security guard on duty.”

A March 2016 report by TechCrunch pointed out that the Taiwanese startup would compete with deep-pocketed hardware manufacturers such as China’s Huawei and the U.S.’s Seagate. Umbo CV’s leadership says the company can distinguish itself by using AI technology to scan images from multiple security cameras for abnormalities instead of relying on geometric models that focus on specific shapes (such as a person with a gun), the report said.

Changing a mindset  

Although Taiwan’s progress in AI is laudable, and businesspeople and policymakers alike have a sense of urgency about embracing this technology, regulatory issues have the potential to stall Taiwan’s AI ascent if government officials do not face them squarely.

For instance, Taiwan’s Personal Information Act defines medical records, health-examination information (a separate category), and genetic information as “sensitive information.” The law stipulates that such data cannot be used for research purposes without receiving consent from the people involved, who must appear in person and give their written agreement.

In an email to Taiwan Business TOPICS, John Eastwood, a partner at the Taipei law firm Eiger, notes that there are exceptions to this rule for research institutions working on medical or public health issues. However, “the disclosure of information cannot be sufficient to identify the specific person, so identifying information would need to be stripped out of any data transmitted for research use,” he says. “The last thing anybody would want is for somebody in a medical-school or conference environment to suddenly shout: ‘Wait a second – that’s my dad you’re talking about!’”

There are ways to protect a patient’s identity and still use the data for medical-research purposes, Eastwood says. For instance, a patient’s age might be relevant but the exact birthdate would not be important.

“From the AI standpoint, you have an opportunity to have greater efficiency and an opportunity to remove human eyes from the review and removal of identifying information. However, you have a risk that the AI will get it wrong and transmit personal identifying information,” he says.

Thus far, hospitals have been hesitant to risk disclosing patients’ personal information. Failing to tap the data for research purposes could have serious implications for Taiwan’s progress in both AI and medicine. For example, under the current law, if a Taiwanese company were to develop AI image-recognition technology able to more accurately detect malignancies than the human eye, researchers would not have easy access to the data of Taiwanese patients to test the technology. They could request that patients appear in person to provide written consent, but it’s possible some would refuse.

Because of genetic similarities among populations in East Asia, the Taiwan data would be widely applicable across the region. If the data could be freed up for research use, Taiwan would therefore have a large potential market for AI healthcare solutions.

“There is a need for deregulation so that we can use the data,” says IEK’s Su. “We need to recognize that if we don’t move on this, others will.” He points out that Alibaba, the Chinese internet giant, is moving aggressively into AI healthcare. Last year the Hangzhou-based company invested US$35 million in a medical-imaging venture, and earlier this year its cloud-computing unit launched a suite of AI solutions for hospital use. Alibaba is also partnering with the Beijing Geonomics Institute in a project that uses deep-learning technology to identify lung cancer.

Another obstacle in the path of AI’s progress in Taiwan is the short shrift traditionally accorded to software development in this market. Jing Bing Zhang, the Singapore-based research director at International Data Corp. (IDC) and a robotics expert, says that it will take a great effort to overcome Taiwan’s weakness in software. While acknowledging that Taiwan has some successful AI startups, he says that on the whole “Taiwan is lagging behind in AI development because of the focus of its companies on hardware and assembly.”

Will Taiwanese manufacturers, known for their focus on short-term profits, be willing to bet on unproven AI solutions? Given the fledgling state of the industry, “the risk is high that many AI companies will die,” Zhang says.

As a result, more government support could make a crucial difference in Taiwan’s AI prospects, say industry observers. “The Taiwan government has become more proactive in its support of AI, but it’s not investing as much as other countries,” observes AI Labs’ Tu. For example, the South Korean government announced nearly US$850 million in funding for AI in March 2016, nearly US$300 million more than Taiwan’s initiative. And in July, The New York Times reported that China is preparing a multi-billion dollar AI investment plan.

Paramount to Taiwan’s prospects in AI will be the willingness of the technology industry to change its traditional way of thinking, Tu says. “We can’t rely on OEMs and ODMs forever – we need to have a software mindset instead.”

That won’t be easy, but Tu has reason for optimism. Soon after launching AI Labs, he received thousands of resumes from interested candidates. Applicants were evenly divided among fresh graduates, returnees from overseas, and former employees of contract electronics manufacturers.

“People are genuinely excited” to work for an AI research organization in Taiwan, says Tu. “They never thought they would have this opportunity here.”