In his book on the dynamics that will drive the future of AI, Lee Kai-Fu (李開復) set out the potentially fundamental advantage China has over the U.S. While the U.S. might still have more top-level AI engineers, that will ultimately be outweighed by China’s superior access to the data on which AI systems can be trained to recognize patterns. The reason: As well as a larger population, its top apps — such as WeChat (微信) — offer a wider range of services than the most high-profile U.S. apps — such as Facebook or Twitter — and working with weaker privacy norms and laws are able to gather more and better information on users.
Here’s the interesting thing about that for Taiwan: As well as presenting a challenge to the U.S., Lee’s premise also presents a challenge to other, smaller AI contenders. If he’s correct, it raises the question of how the likes of Taiwan can compete on AI. How can Taiwanese companies, or governmental projects, make any significant impact on the world, when their access to data is limited compared to larger rivals?
At the AI Taiwan event in Taipei yesterday, various answers were on offer.
One is that Taiwan’s government has been supportive of the nascent industry and that matters. “I’ve found government in [the last] three [or] four years really open and very agile in helping startups like us to get marketing exposure, like winning this kind of award,” Sung Yeh (葉松瓚), chief operating officer of DeepWave (迪威智能股份有限公司), a Taiwanese company that creates audio AI applications, told us after collecting an award handed out by Taiwan’s Digital Industry Agency (數位發展部). Yeh said the government has helped his company take their products abroad and provided subsidies — in particular supporting up to 40% on research and development. On top of this, he added that Taiwan’s National Development Fund provides funding for those companies who have already secured angel investors.
This isn’t to say everything is easy, though. Three key elements can generate successful AI, according to Yeh, “algorithms, data and computing power,” and what companies like his lead on right now are very sophisticated and advanced algorithms. However, from here, this still means the next step is getting access to the “larger pool of data, computing power and venture capital” that larger players like China and the U.S. have to offer. In other words, you can only get so far on your own.
Another view comes from Monique Yang (楊明慧) of AetherAI (雲象科技), a medical image AI company. She explains that AetherAI — another prize-winner — has got ahead within its particular market because it has specific knowledge of the existing industry. Its CEO is a doctor who “knows the hospital system and … knows what doctors really need,” according to Yang, who said one piece of feedback the company had received on why it had been chosen for a project with Tubingen University Hospital in Germany was that the company had been publishing papers in medical journals. Another factor in their favor when dealing with clients from other countries has been the “flexibility” they’ve been able to offer. Yang said in the case of the German hospital the company had answered more than 150 questions in the first stage of bidding, where larger companies had simply written “not available” to some.
A number of other pitches for the viability of the AI industry in Taiwan were also interesting to hear. The ready availability of “cheap” engineers. The close proximity to massive producers of high-quality integrated circuits, which provide hardware support for the operation of AI algorithms. The practical advantages of training AI on local data, like language and culture matching the demands from applications, and even having local customer support centers, which provide an advantage when offering AI services to other businesses within Taiwan. And finally there was also the point that some simple commercial applications do not require large datasets to be useful.
Of course, at an event promoting Taiwanese AI, it’s unlikely that anyone is going to come out and say the whole industry is doomed to failure. And not every AI application at this event looked like a world-beater. But some of the above answers, at least, are convincing on the point that there are AI niches to be carved out outside of the largest centers of data and finance. And Taiwan has some things in its favor for trying to do that.
Image: Kris Lih, Domino Theory