Once part-sci-fi, part-theory, Artificial Intelligence (AI) has emerged as a new frontier for rapid fire innovation and collaboration. Breaking from traditional norms in technology innovation, AI companies are increasingly working to make their tech transparent through the AI Open Source Software (AOSS) framework, or the practice of granting the public access to one’s own AI software and code. While open sourcing has gained significantly more traction in Western markets, China has become a petri dish for AOSS culture over the latter half of the past decade, and the movement is quickly gaining pace.
As the AOSS movement has received increasing attention, it has also drawn wider support. AI firms benefit from having a developed and widely-used open source platform because it serves as a free testing zone. Looking at the bigger picture, technology companies stand to benefit by leveraging this space to improve its own technology, gain market dominance and thus attract top talent. This cyclical slew of benefits has caught the eye of the Chinese government, earning support at the top-level. Much like with other industries or initiatives, once AOSS received government-backing, more and more AI firms are quickly following suit and driving China’s AI sector forward.
Why Open Source?
While it has been argued that open sourcing code can potentially decrease revenues in the short-term, many Chinese companies are playing the long game and have chosen to embrace it because it can bring significant benefits, including: raising publicity, gaining independence, and increasing innovation.
Open source software can attract a large base of developers in ways that paid or locked code cannot, which in turn can be used as a resource. For a company, having free access to a talent pool can be a huge incentive to become a leader in AOSS. An example of this is Google’s TensorFlow. TensorFlow’s industry dominance contributes to Google’s status as an AI powerhouse, which in turn attracts top talent to Mountain View. Chinese companies have inevitably taken note and also seek to recruit more AI experts onto their teams this way, increasing the incentive to adopt AOSS usage.
Apart from publicity in terms of hiring, a high-profile leader can also set the tone for AI industrial standards. Standards in AI are particularly affected by AOSS, as it can establish de facto standards prior to the standard being officially set. Google benefited from this when open sourcing Blaze, Google’s internally used system for building software. As external developers learn how to use the system, they reinforce and spread these standards while also setting a precedent for other tech firms to conform and accept them across the industry. China has taken particular interest in this and has outlined it as a tool in “China Standards 2035,” a plan for China to lead industrial standard-setting that is expected to be released later this year.
Just like a country relying on a specific source for oil, being overly reliant on another nation’s AOSS is also risky. GitHub, the main platform for sharing open source code, is owned by Microsoft and subject to U.S. trade controls. As such, the threat of code sanctions within GitHub is a real possibility. In June of 2019, GitHub announced that developers in Syria, Iran, Crimea, Cuba and North Korea would have restricted access to certain services. China reacted to the potential strategic risk of over-reliance on GitHub through the forming of GitCafe, a Chinese platform with similar services to GitHub. While a good first step towards self-sufficiency, GitCafe still remains underdeveloped, and Chinese open source technology is not yet a sufficient replacement for GitHub.
Finally, there is a clear macroeconomic benefit for a country to use AOSS. When tech giants establish open source softwares, smaller players get easy and free access to powerful resources. Thus far, China has already reaped considerable benefits from existing AOSS. Chinese developers are the third largest group contributing to the Cloud Native Foundation projects, and the second largest group working on GitHub-hosted repositories. As a country that so heavily emphasizes advancements in AI, China is the perfect candidate for AOSS because it saves companies time, labor, and resources.
Industry at a Glance
China’s budding AI market took root in 2011, and has since launched to a 2019 valuation of CN¥71 billion (US$10.14 billion), with 45% average annual growth from 2015 through 2019. AI has been driven by researchers and private companies in China from the onset, and largely remains so today. However, in 2017, the State Council released the New Generation of Artificial Intelligence Development Plan. In this report, the State Council established AI as a strategic interest and outlined its development strategy. Additionally, Beijing identified AOSS as foundational for facilitating AI development in areas such as theoretical research, big data, and cloud computing. In a 2018 white paper, contributors from Peking University, Baidu, and Huawei stated: “AOSS is an important foundation for China’s AI development…it is an area that must be fought for in seizing global dominance in AI.”
One of the private companies that became an early adopter was Baidu, which set precedent in 2016 by releasing PaddlePaddle, a powerful open source AI platform. Since then, many companies have followed suit. Some of the new major AOSS platforms include Angle from Tencent in 2017, Alibaba Cloud’s Machine Learning Platform in November of 2019, and Huawei’s Mindspore in March of 2020.
PaddlePaddle Makes a Splash
TensorFlow was the first ground-breaking platform of its kind in the US, and is now arguably the dominant AOSS. Likewise, PaddlePaddle has evolved from being the first major Chinese AOSS to becoming integrated with the technology of more than 84,000 enterprises. Fundamentally, PaddlePaddle is designed for industrial applications of image recognition and natural language processing. What distinguishes PaddlePaddle is its accurate and efficient image recognition tools, adaptability, and applicability to the particular needs of Chinese markets.
PaddlePaddle is now used globally by 1.9 million developers on over 230,000 models in a variety of sectors, including autonomous vehicles, healthcare, and forest management. Going forward, Baidu has announced PaddlePaddle’s collaboration with hardware companies such as Intel, Huawei, NVIDIA, Arm China, MediaTek, Cambricon, Inspur, and Sugon. The platform offers development frameworks for other cutting edge software, such as quantum-computer algorithms. Although the name of the major Chinese AOSS is still foreign to some Western data scientists, the development and application of PaddlePaddle puts Baidu at the forefront in the world of AOSS.
Following in its footsteps, Jinlu Technology
There is an abundance of companies that have piggybacked on the rise of PaddlePaddle’s code. For example, Jinlu Technology is a firm that innovated ways of sorting garbage with AI. The specific technology used is known as “computer vision,” which was originally trained on Western data but could not be used to perform similar tasks in China. This meant a new solution was needed.
Instead of forming a team of experts to build a new model from scratch, Jinlu Technology simply adapted code from PaddlePaddle. CEO of Jinlu Technology stated, “Using traditional computer-vision models in China would be useless,” and continued on how AI can reduce labor by 96%, improve accuracy, and optimize sorting. In this case, having readily available access to AOSS with strong computer vision that was developed by a Chinese firm proved to be extremely useful.
Specialized AOSS: Linkingmed
Beyond direct application of the code, PaddlePaddle has spurred new developments in AOSS. Linkingmed is a Beijing-based firm specializing in medical data analysis. In March of 2020, Linkingmed created and released an open-source AI framework for CT scans using PaddlePaddle’s code as a base. Its framework aimed to improve the ease and accuracy of diagnosing pneumonia, one of the severe side effects of COVID-19. Linkingmed’s software can identify pneumonia in less than a minute, with 92% detection accuracy, while human analysis of CT scans have similar accuracy but take at least several minutes. The development of this AI was critical during the pandemic to provide China’s overwhelmed healthcare system with efficient and effective diagnostic tools. Baidu itself has open-sourced more AI platforms in light of the pandemic. Particularly, the company open-sourced software for an AI to detect whether an individual is wearing a mask. This development is the first of its kind in the industry and holds an accuracy rate of 97.27%. Business and government agencies were able to use this to supplement or automate mask checking entirely.
Future of AOSS Growth in China
When discussing AI, the term “big data” is often thrown around. With a population of 1.4 billion, and about 800 million internet users, China’s data is by far the “biggest.” In 2018, the total volume of new data created was 7.6ZB and is projected to increase to 48.6ZB in 2025, compared to the US as 6.9ZB and 30.6ZB, respectively.
AI is best when trained on real data, and the more of it the better. In this sense, there is a natural synergy between China’s massive amounts of data and the development of AI. Already this can be seen by how China boasts excellent AI in facial/image recognition and natural language processing, which are often used in both the public and private sector. One example of a company that is taking full advantage of the resources available is Megvii, a Beijing based AI founded in 2011.
Megvii: From Startup to Blacklist
Megvii focuses on computational photography, facial recognition, and object recognition applications. The startup, backed by Alibaba, is one of the major players in the growth of Chinese AI. In March 2020, the company released parts of its platform, MegEngine, for developers at large. Of note is that the company’s image-recognition algorithms defeated Google, Microsoft, and Facebook in the 2017 COCO Image Recognition Competition, and the company received an the World’s Leading Scientific and Technological Achievements Award at the 2019 World Internet Conference in Wuzhen, China. Megvii has released Face++, which is now the world’s largest open source image-recognition platform. Based on its achievements and world-wide recognition, Megvii is often considered one of the unicorns of Chinese tech startups. The story of Megvii can fit into the narrative of the rise of great Chinese tech, and it also aligns with the growing sentiment that China is increasing its capacity for state surveillance.
Megvii’s development has many success stories, but the company has faced its fair share of challenges in going public. In October 2019, the US Department of Commerce added Megvii to the Entity List, meaning American companies must receive special permission to export technology or software to Megvii. This was due to the belief that Megvii was using its technology as a tool for state-surveillance by storing data on individuals, and facilitating unethical usage of the obtained information. Megvii was one of 8 tech companies placed on the Entity List, and one of top 3 leading AI firms in China.
A Plan Going Forward
In 2020, AOSS culture is alive and well in the Chinese market, following the 3 principles established in the 2018 White Paper on the Development of China’s Artificial Intelligence Open Source Software (Original, English).
First is “construct an open ecology,” which translates to gathering necessary talent, promoting a culture of open sourcing, and focusing on basic AI technologies and segments where Chinese AI already excels. Second is to “build up an external environment,” which recommends establishing the necessary institutions to support AOSS growth. This includes items such as financing, putting IP protections and proper security measures in place, and a more substantial government role to guide AOSS development. Third is to “drive forward industry applications,” which emphasizes AOSS innovation, integration with existing technologies, and application in various industries. To learn more about the driving forces behind technological innovation, consider this article that examines the US-China decoupling.
Using these three pillars as the roadmap to AOSS, China is working towards fostering AOSS to play a bigger role in the years to come. Tech giants have set a precedent for the burgeoning AOSS market, and smaller developers are utilizing the new resources to create innovative products and further the environment of collaboration. However, if China wants to attain its position as an AOSS leader, the country will need to double its usage and participation in existing AOSS programs, increase its research in areas where China has an advantage, and continue to strengthen its domestic platforms. When these elements come together, China will be able to better serve its AI-using business communities, get closer to reaching its larger economic and military goals, and gain an edge when competing on the global stage.