Feature: The War for AI Talent Rages in China (Part 2)

When it comes to AI, talent means everything. AI talent is not easy to find, not to say there are too few of them compared to the considerable demand.

By

Feature: The War for AI Talent Rages in China (Part 2)

Writer Sun Ran, Yang Lin

Editor’s note: 

Many believe AI-powered automation is likely to replace human workers and spell an end to the human race.

Before resulting in unemployment in large scale or bringing about a catastrophe, development in AI and the competition in the AI industry have brought AI engineers abundant job opportunities and decent salaries.

From big tech names to early-stage startups, companies need to chase after AI talent if they want to make the most out of this technological innovation, regardless of the cost.

This is the Part 2 out of the total three.

Link: Part 1Part 3


Scarcity of talent

Image credit to 123rf.com.cn

The demand for AI talent is so huge, yet the talent pool is so small, making it difficult for even well-financed companies to hire the people they want.

Against this backdrop, helping companies recruit AI talent became a seemingly good business.

In general, a headhunting firm is paid 20% of the annual salary of the candidate hired. According to an estimate by the senior headhunter Chen, whose targets are mid-level and top-tier talent with at least five years of work experience, the average gain for headhunting firms per candidate hired is around ¥80,000 or 90,000 (around $12127 or 13643).

The figure could rise to over ¥100,000 ($15159) or even ¥200,000 ($30318) if the candidate is an AI specialist or algorithm engineer.

However, Chen and his team have decided to put their focus on areas other than AI this year, saying that “it’s not a worthwhile business”.

It’s true that the demand is running high – almost every internet firm that has raised the B round is hungry for talent in algorithm engineering, but the reality is: the talent pool is so small that competition among headhunting firms is extremely intense, with each candidate targeted by a dozen headhunting firms. A company would be seen as doing great if it can headhunt an AI candidate in six months.

Chen’s team didn’t deliver a single AI deal last year. “AI candidates are really a headache for us,” said Chen, a little frustrated: “99% of candidates with expertise in AI or algorithm engineering turned us down directly on the phone, as opposed to an average of 60% for other areas.”

The industry is seeing a serious shortage of AI talent.

Photo by Faustin Tuyambaze on Unsplash.

This shortage is partly caused by universities’ failure to keep up in course design. Most universities in China teach AI and algorithm engineering courses only to postgraduates and students at higher levels. This has greatly limited the size of China’s AI talent pool.

Typically, only 20 to 30 students graduate with master’s degree in AI- or algorithm-related disciplines and around 10 with bachelor’s degree from a university a year.

This is a drop in the ocean in an industry that’s expected to become the bedrock for a host of other fields.

To make matters worse, there are a limited number of universities capable of training qualified AI talent. Based on Roger’s estimate, institutions with the capacity to groom high-quality talent for AI-related positions in China number fewer than 20, prominent ones being Tsinghua University, Peking University, Shanghai Jiao Tong University, Harbin Institute of Technology and Chinese Academy of Science.

Whether a company can successfully poach talent and what talent it can attract have become a crucial factor in the AI race.

Search beneath the surface

Image credit to 123rf.com.cn.

It’s important that a founding team is able to identify and know how to woo top talent.

News aggregator Toutiao and photo-sharing app Kuaishou have arguably done the best among Chinese companies in applying AI technologies to their respective areas.

Zhang Yiming, the founder of Toutiao, studied software engineering at college and, before founding Toutiao, had worked as a technical engineer at Microsoft and the flight and hotel search engine Kuxun.cn.

Su Hua, Kuaishou’s founder, quit a doctoral program at Tsinghua University to join Google China, where he did research on the applications of machine learning in search services.

Top-tier AI specialists are often passive candidates, meaning they are usually referred to jobs by their tutors/colleagues or receive offers directly from employers. Such talent is often hard to identify, requiring companies to search beneath the surface.

“The thing is, people who are busy doing their work are usually not those who get a lot of media exposure,” Min Wanli told Kr-Asia, “so headhunters have to make extra efforts to locate such talent.”

The two most-watched organizations are Microsoft Research Asia and Baidu. They have built a large pool of talent with mastery of cutting-edge technologies. However, since last year, much of their talent has left to join start-ups.

“Toutiao, in particular, keeps a close eye on Microsoft Research Asia,” a headhunter noted.

This February saw Ma Weiying, the former assistant managing director of Microsoft Research Asia, join the over 10 billion yuan-valued unicorn to serve as the latter’s vice president and lead its artificial intelligence laboratory.

Ding Haifeng, who helped build the underlying architecture for Toutiao’s flagship recommendation system and is now a senior data engineer at the company, had previously worked at Baidu’s webpage search unit.

When tech staff job-hop, they often leave in groups. More often than not, when the director of an AI project jumps ship, the engineers working under him or her will follow suit. It’s not uncommon for a company to lose, say, five tech employees at once.

“It’s like in martial art fiction. People naturally want to hang out with top-tier martial artists, except that in the AI landscape, fighting techniques give way to intelligence,” said Wu Jie, who also cited the fact that engineers tend to work together rather than by themselves as a reason.

Link to Part 1 Part 3 of the feature.