Artificial intelligence

Megvii's IPO raises question: Who is responsible for the (mis)use of AI?

A facial-recognition startup with links to the Chinese government has inflamed the global debate about who should harness the use of artificial technology and to what extent.

When a Human Rights Watch report came out on May 1 linking artificial intelligence group Megvii Technology to police surveillance of a Chinese ethnic minority, investors in the startup grew nervous. To reassure them, executives of the company quickly jumped on a conference call and vowed to root out what was behind the story.

A month later, the New York-based watchdog corrected its report to say Megvii’s facial recognition Face++ code was in the police app but inoperable. How it got there remains a mystery, but by then the public relations damage was done, with some of Megvii’s shareholders, and their shareholders’ shareholders, questioned publicly about their views on the harassment of Uighurs, a Muslim people from the Chinese western region of Xinjiang. 

In another blow, on October 8 the US included Megvii and several of its Chinese peers on its trade black list, barring them from buying components from US companies without US state approval because they had been implicated in abuse of Muslim minority groups. Megvii said it objected to its inclusion on the list and that it had generated around 1% of its revenue from projects in Xinjiang in 2018 and no revenue in the six months ended June 30.

Investors in cutting-edge technology can make fabulous profits but also risk public opprobrium. A poll of investors managing about $12.75 trillion found that 85% of them saw artificial intelligence, or AI, as an investment risk that could potentially provoke a social backlash against job losses, stir up geopolitical tensions and create value that would be hard to measure.

Such concerns are reflected in the market, with the number of investments made into private Chinese AI companies down 21% in 2018 from a year earlier – a weaker trend that has continued into 2019. The looming Hong Kong initial public offering of Megvii – which is backed by Chinese technology heavyweights Lenovo, Foxconn and Alibaba – will likewise test public investors’ appetite for the increasingly high-profile risks associated with AI, as well as whet their appetite for a business whose revenues leapt 356% year-on-year in 2018.

The startup generates roughly a third of its revenues from projects where government departments are the end-users. 

“Megvii’s IPO announcement has raised potential privacy and data protection concerns about facial recognition technologies among investors, which has been a limiting factor on sentiment for this sector,” Miranda Carr, an investment analyst at Chinese broker Haitong International, said in a report.

Among the myriad applications of AI, computer vision is the most widely used in China [see chart on opposite page] and is becoming politically charged as anti-government protestors in Hong Kong mask their faces or point lasers at cameras to avoid identification by the government.

Globally, also, journalists are challenging creepy corporate behaviour, prompting some companies into taking drastic action: Microsoft, for one, deleted a trove of 10 million faces called MSCeleb in June that it had compiled unknown to many of the individuals in its database. Investors, increasingly, are having to ask themselves where the responsibility lies for how computer vision technology is used.

Yin Qi, chief executive of Megvii, has thought a lot about this issue and believes that the responsibility for advancements in technology is shared.

“Everyone needs to think about it and be responsible for it,” Yin told FinanceAsia in an interview on August 20. That includes the government, developers and consumers of the technology, he said. “I believe that technology is always good but as a technology engineer, and as a tech company, we should realize we have the responsibility to consider the ethics.” 

To reassure investors ahead of its IPO that it is doing its bit, Megvii recently set up an AI ethics committee, written an AI code of conduct for its employees earlier this year, and is structuring contracts to guard against the weaponisation and misuse of its technology. 

Socially responsible investors shouldn’t just shun companies caught up in a scandal as “a crisis can crystallise good behaviour,” notes Wendy Cromwell, a director of sustainable investment at investment manager Wellington Management, which has about $1 trillion under management.

German carmaker Volkswagen’s overhaul of its operations after an emissions scandal in 2015 is a case in point, she noted whilst speaking at the Milken Institute’s 2019 Asia Summit.

EYE ON YOU: Megvii's cameras are used to smooth the flow of commuters in Beijing's subway


Founded in 2011, Megvii is trying to proselytise the use of AI by opening up its algorithms to developers, taking the view that if more people look at its code, the more likely its AI will become widespread, bugs can be fixed, and long-term opportunities, as well as the risks, can be worked out.

“I’m a pure tech guy. I’m a technology believer. So, above all, I think technology is the hope to make the world better. All tech is very neutral,” Megvii’s Yin said.

Looking at just economic growth, he’s likely right. According to consultants at McKinsey, AI could potentially deliver an additional economic output of around $13 trillion by 2030, boosting global GDP by about 1.2% a year.

But pundits, generally, see severe economic and social disruption along the way too.
Megvii launched its Face++ open platform in 2012, which is used by 200,000 developers globally, to design and test apps potentially in ways other than the original programmer intended. Access is largely free.

“The purpose of this open-source platform is to promote AI - let developers see what AI can do,” Megvii’s chief financial officer, Haitong Wang, told FinanceAsia in a separate interview on August 1.
Source codes used to be largely proprietary and kept hidden from public view, but now some of the world’s biggest technology groups have created open-source deep-learning platforms including Google’s Tensor Flow and Facebook’s Pytorch.

Image and speech recognition is a more sophisticated form of AI that requires deep learning, data processing through neural networks, and allows algorithms to understand complex data and make determinations.

Use of facial recognition is spreading and, in some cases, leading to controversial new applications. Nasdaq-listed Momo launched its Zao app on August 30, allowing users to swap their faces with those of a celebrity on images they have uploaded. Deepfake app Deepnude, meanwhile, removes the clothes from the images of women.

“We will see cases of the technology being used wrongly, such as mimicking celebrities’ speech, so really it is the responsibility of all parties,” said Yin.

Megvii has shifted its focus away from the online entertainment segment of AI.

There are concerns also that AI could end up being inaccurate and produce biased results. That means that a company like Megvii needs to ensure that the data on which it trains its facial-recognition algorithms is as widely representative as possible.

Analysts at Chinese investment bank CICC said in a report that facial recognition technology in China has little room for improvement citing a face recognition vendor test last year held by China’s National Institute of Standards and Technology, in which the highest accuracy rate reached 99.3% versus 95.5% in 2017.

“It’s totally achievable. It’s just that you need to be more mindful when you develop products. So, when you are training the algorithms you need to feed it different data sets and [the process] needs to be supervised by a human who is held accountable,” said Haitong.


In the summer of last year, the mayor of Beijing called Megvii to solve a problem: abandoned dogs had attacked people and he wanted to hold the owners responsible for their pets’ behaviour. Megvii built an algorithm that could recognise a dog based on the unique pattern of its nose. Photos of dogs’ noses are now stored online to identify strays and return them to their owners.

“After the mayor called us, we got that done in two to three weeks,” said Haitong. Megvii is now commercialising this technology to keep an eye on valuable pooches too.,

The speed of development across the AI industry is breathtaking, boosted by open platforms, and industry regulators are struggling to keep pace. Megvii’s proprietary deep learning framework, called Brain++, is akin to a factory that can swiftly train algorithms to suit customers’ requirements and at scale too.

“We have been working with the authorities for many years, on how to shape the privacy regulations for the AI industry,” said Haitong. “Why this industry is so hard to regulate is because it’s so new and dynamic and also technical.”

It is for politicians to delineate how far companies can commercialise people’s data and balance privacy concerns against stifling innovation at fast-growing companies and using technology to improve people’s lives and boost business activity. Where to draw the line exactly is unclear. The European Union, for instance, has one of the world’s strictest regimes after it promulgated in May 2018 the General Data Protection Regulation, more commonly known as GDPR.

Investors can struggle to assess the policy risks as the laws and regulations that might apply are often complex and constantly evolving as new interpretations come to light and the technology extends its reach into new areas.

The industry is looking to self-regulate to avoid a crackdown by heavily encrypting data that could be used to identify a person if it got into the wrong hands.

Megvii staff work on training algorithms

Another danger for investors is geopolitical as nation-states vie for leadership in technology and resort to tariffs and blacklists to thwart one another. In such circumstances, companies such as Megvii, which has about 250 registered AI-related patents, can become collateral damage.

In 2017, the Chinese government set a goal of becoming a world leader in AI by 2030. It is the second-largest spender on research and development (R&D) after the US, filing the highest number of global patents since 2010. Credit rating agency Moody’s says that China is already more advanced in the field of facial recognition and digital payment systems than many developed economies.

This swift ascendancy has provoked widespread concern among US politicians. The US Commerce Department's inclusion of Megvii and its peers on its Entity List on October 8 is an escalation in the tension between the two super powers.

To be sure, Megvii sees fairly little direct damage from the US-China trade war since its revenues from the US are very small. As a software company, its list of suppliers is also short. Megvii’s algorithms are trained by its own proprietary deep-learning framework Brain++, instead of relying on US third-party platforms.

Even so, as the company’s IPO prospectus notes that if China were to increase tariffs on its US imports, Megvii’s suppliers might not be able to find substitutes of the same quality and price, which would boost its costs.

Given the chill wind from the US, Megvii’s international expansion is focused on the rest of Asia and the Middle East.

Yin said Megvii is expanding where there are booming economies and local industries have pain points that AI could help solve, singling out ecommerce companies in Southeast Asia as potentially fertile ground.

“Last year was actually our first year for the international business. So, our international business is small. But the potential is huge,” Haitong said during a visit to Hong Kong. Megvii has less than a hundred pilot projects outside of China.


Surveillance and security are still by far the biggest applications of facial recognition technology in China. So its Orwellian ‘Big Brother’ associations are somewhat inevitable.

But it doesn’t necessarily have to be sinister. In retirement homes, for example, cameras can keep an eye out for older people’s welfare.

Companies such as Megvii and rivals Sensetime, Hikvision and Yitu are also branching out in new directions, from using the tech for  other applications or looking to combine it with the internet of things (IoT) in homes or intelligent robots in warehouse management, as well as exploring how to use it to help make cities run more efficiently.

Megvii’s cloud software as a service is used in a multitude of ways – from allowing fintech companies to verify borrowers’ identities for online lending to helping ride-hailing firms check their drivers’ identities before they start their shifts. Haitong said this segment is profitable and growing fast.

Megvii has also been trying to offset the contraction in the number of China’s mobile internet users by selling cell phone manufacturers more products. Haitong said she sees the mobile phone segment becoming profitable soon.

But Megvii’s biggest revenue generator is its smart-city business, which started in 2015 and is still growing fast. Revenues from a mix of government security agencies and companies swelled from Rmb18.65 million ($2.6 million) in 2016 to Rmb1,057 million ($148.4 million) in 2018.

“Local governments really pay a lot of attention to AI as AI will be a very important tool to improve quality of life and efficiency in societies,” said Yin. Sovereign analysts at credit rating agency Moodys agree as they think AI has the potential to raise economic, institutional and fiscal strength and reduce a government’s susceptibility to event risks.

In mainland China, Megvii’s video surveillance is used to smooth the transit of passengers on the Beijing subway and to clock staff in and out of buildings, providing employers with an efficient attendance check. Megvii is providing access control for Beijing’s tallest building China Zun, for instance.

And overseas, Megvii works on city IoT projects in Hong Kong, Singapore and Thailand. In Macau, it also provided services to a casino hosting a party.

Another field that AI companies are exploring with facial recognition tech is education. At Beijing’s Tsinghua University, Megvii’s cameras are used to control access to dormitories and speed up online registration for classes. China’s education ministry has, however, reportedly told schools to use facial recognition sparingly to protect children’s data privacy.

Megvii newest business segment, supply-chain IOT, was launched in 2017 and has just a few proof-of-concept projects running around the world. The company made just Rmb46,969,000 ($6.6 million) from this area in the first six months of 2019 but believes it has considerable promise. Megvii is betting that companies in rapidly ageing societies such as Japan will look to deploy robots and sensors to manage warehouses controlled by AI instead of employing an expensive labour force. To that end, it bought the Beijing Ares Robot Technology company in April 2018.

“We believe that the supply-chain [IOT business] will become a great revenue contributor, probably in two to three years,” Haitong said.

Thanks to this innovation across its business, Megvii’s gross profit margins have remained healthy at around 65% in 2018 and the first six months of this year, according to its draft IPO prospectus lodged with the Hong Kong stock exchange.

Its growth and R&D mean Megvii is still in the red, with operating losses of just over Rmb115 million ($16.2 million) in the first six months of 2019.

Investment bank Roth Capital Partners estimates Megvii is on pace to generate over $300 million in revenue this year while losing about $28 million to $32 million. Based on a pre-money valuation of around $4 billion, Megvii is targeting a punchy 2019 revenue multiple of about 14 times, the bank calculates.

“Megvii’s IPO filing in Hong Kong acts as a proof point for the value investors are putting on AI businesses at scale,” Darren Aftahi at Roth Capital Partners, said.

The company hopes to attract more investors that can overlook some of the potential reputational and regulatory bumps in the road and buy into AI’s long-term prospects, much like early investor Lee Kai-Fu, a Taiwanese-born American computer scientist who heads Sinovation Ventures.

“AI will be the main breakthrough of the next 30 years,” said Yin. “AI will have ups and downs, maybe in three years or five years it won’t be a hot topic anymore. So, I need the other investors, who believe in tech in the long term.”

It is an emotive subject, one that is clearly going to become a bigger part of all our lives and make the leaders in this field very powerful and prosperous. Having put people under growing surveillance, it is now Megvii that is being carefully scrutinised by investors far and wide.

This story has been updated to include Megvii's designation on the US Commerce Department’s Entity List

Assistance from Elizabeth Utley



¬ Haymarket Media Limited. All rights reserved.
Share our publication on social media
Share our publication on social media