Drafting China's National AI Team for Governance
Companies get special roles, but they're expected to serve as platforms for others
Benjamin Cedric Larsen
Nov. 18, 2019
This article was published as part of the Stanford-New America DigiChina Project's first special report, AI Policy and China: Realities of State-Led Development. Read this article and the others in PDF here.
China’s government has become increasingly active in supporting a national agenda of AI development. In doing so, it is devising new means of guiding development in greater concert with leading private sector enterprises that advance key AI technologies and applications. Mimicking previous development strategies, a few companies have been selected as “National AI Team” members, an endorsement that carries both national and local government support, as well as access to regional projects and related public data resources. In return, the government expects that key standards for AI ecosystem development can be coordinated with greater efficiency among stakeholders, while smaller enterprises are enabled to synchronize with leading AI developments through open innovation platforms.
The notion of “National New Generation Artificial Intelligence Open Innovation Platforms” (AIOIPs) originated in November 2017, when China’s Ministry of Science and Technology (MOST) endorsed four private sector companies to construct platforms for specific purposes. The companies were Baidu (for autonomous driving), Alibaba (smart city), Tencent (medical imaging), and iFlyTek (smart audio, i.e. natural language processing). A fifth AIOIP, SenseTime (smart vision), was added in 2018.
In August 2019 the initiative was expanded to include 15 AIOIPs, and it remains open for further entities to apply. Applicants are assessed by a team of experts organized by MOST, while selection criteria rests on the applicant's technological capabilities and anticipated results of the AIOIP. Each applying entity has to pre-specify a subdomain of AI platform development, which the entity will focus on opening up to a broader array of companies for further interaction and development. While each sub-domain is expected to relate to a distinct area of AI development, it is clear that conceptual overlaps exist between the domain boundaries.
The AIOIP initiative relies on leading enterprises to promote deep integration of AI with the real economy, while companies are expected to deliver on four key tasks across: R&D, ecosystem participation, sharing data and open-source software, and supporting the entrepreneurship of small and medium-sized enterprises (SMEs). MOST, along with provincial-level S&T management departments, actively support the AIOIP initiative, for example by promoting application of the relevant technologies. This implies providing space for testing autonomous vehicles (Baidu), access to city infrastructure for monitoring and upgrade (Alibaba), access to public medical data (Tencent), access to the judicial system (iFlyTek), and access to surveillance systems (SenseTime), among other public areas that are being opened to private-sector development. In return for official support, each AIOIP entity is expected to provide annual reports summarizing the ongoing progress of their open innovation platforms.
Opening Public Data While Supporting the Platform Economy
Through these collaborations, a new model of AI development and an associated governance model are emerging in China, where government-designated platforms and related public-private partnerships emphasize an experimental, gradual, and decentralized approach to selectively opening public domains and associated data repositories.
In the process, leading private sector enterprises are endorsed to apply innovative AI solutions to optimize public institutions and the provision of public goods and services, often implemented on a local and regional basis. Tencent’s Miying platform, for example, partners with Shenzhen Hospital Center, where residents can undergo AI-based remote screening for retina problems related to diabetes, aiming to alleviate strains on the public healthcare system.
In August 2019, the State Council issued a guiding opinion on Promoting the Standardized and Healthy Development of the Platform Economy, which also calls for greater data sharing between government departments and platform companies. The guiding opinion views the platform economy as a new organizational mode of productivity and economic development, while regulatory oversight from the government is to be devised in greater concert with leading platform operators. The platform economy document and the AIOIP initiative should therefore be viewed as congruent, emphasizing a larger structural role to be played by leading AI enterprises and other players operating in the platform economy.
Platform Leaders Devise Standards and Create the Rules for Ecosystem Engagement
The 15 enterprises that have been selected to establish open innovation platforms are, to a certain extent, crafting the structural rules that affect wider industry engagement. As AIOIPs enable start-ups and SMEs to enter and participate in ecosystem development, they do so through open access to data, toolkits, libraries, frameworks, computing resources, and sometimes competitions, which are accessible through Application Programming Interfaces (APIs) on the open innovation platforms. This suggests that the AIOIP initiative is less about granting preferred access to a few select companies, and more about enabling structural mechanisms that afford greater participation and innovation in emerging ecosystems and sectors that increasingly will be powered by AI technologies.
Companies and regulators are jointly charged with creating new standards for data pooling and interoperability between different but increasingly interconnected public and private systems.
For Alibaba’s ET City Brain, which cooperates with several local governments to provide smart city solutions from transportation to energy, water supply, and so on, this implies that Alibaba is responsible for ensuring that ecosystem partners and developers stay compliant with existing legal frameworks on areas of critical public infrastructure and information systems.
As the selected AIOIP companies are all leaders in their respective fields, they are also the de facto architects of system-wide standards and interfaces, which often are shaped in collaboration with research institutes, universities, and policymakers. In the process, companies and regulators are jointly charged creating new standards for data pooling and interoperability between different but increasingly interconnected public and private systems.
iFlytek, for instance, cooperates with the Shanghai courts to develop Project 206, which seeks to upgrade case-handling in the judicial system. iFlytek also collaborates with educational institutions across 10 provinces to provide voice recognition technology for high school oral examination assessments. Several of China’s National AI Team members also provide educational materials and establish AI curricula for use in China’s educational system.
Building China’s Open Source Resources
In terms of data and software sharing, several of the nationally endorsed AI platforms are also behind a culture change toward open source, echoed in the release of deep learning frameworks including Baidu’s PaddlePaddle, Alibaba’s XDL, SenseTime’s SenseParrots, and Huawei’s MindSpore.
Companies, as well as planners, hope that opening AI frameworks to wider use through AIOIPs will unlock greater network effects, benefitting independent developers and SMEs, as well as expanding entrepreneurial activity in the wider ecosystems.
The release of proprietary frameworks by leading AI enterprises, not only marks a new beginning for China’s open-source community but also may decrease China’s reliance on existing frameworks and operating systems, which predominantly originate in the United States and may become less dependable if technological “decoupling” deepens. Huawei’s commitment to rapidly develop its HarmonyOS mobile operating system, as well as the long-standing pivot towards engineering indigenous AI chipsets, should also be viewed in this regard.
Pursuing National Goals Beyond State-Owned Enterprises
The AIOIP initiative extends an opportunity for private firms to rethink the link between public and private sectors. Traditionally, partnerships involving strategic assets and sensitive information have been reserved for China’s state-owned enterprises, but policymakers have realized that rapidly advancing AI capabilities are developed outside the immediate scope of traditional SOEs. This new approach to AI development aims to move in lockstep with the advancements of leading AI companies and platforms, which emerge as regulatory stakeholders in the process. In doing so, the capabilities of private firms are nudged closer to the Chinese government’s long-term visions, such as those articulated in the New Generation AI Development Plan (AIDP).
Parts of the initiative, however, are indicative of a broader international trend in which governments increasingly provide APIs to trusted intermediaries as a way to open public infrastructure and data repositories to private sector development. Framed as Government-as-a-Platform (GaaP), the public sector can act as a catalyzer for AI development through new procurement strategies, in which the public sector is able to incentivize and subsidize the development and innovation of AI systems for the government’s vision of the public good.
In this new approach to AI development, leading companies and platforms, emerge as regulatory stakeholders.
Meanwhile, by helping establish open standards and interoperability between data and AI systems in areas where the government is opening up, policymakers can forestall the risk of single-player market dominance by encouraging broad systemic operability. In China, these tendencies are currently seen, as many of the enterprises behind AIOIPs compete across each other’s designated domains.
From a business perspective, the establishment of AIOIPs mirrors developments of leading AI platforms elsewhere, and would likely have taken place regardless of central planning. In terms of policy, however, the AIOIP initiative embodies a new approach to the development of AI, which is symptomatic of China’s long-term strategic thinking on the role of AI in the broader economy.
In the United States, the Computing Community Consortium’s 20-Rear Roadmap for Artificial Intelligence, released August 2019, perhaps comes closest to the AIOIP vision. The roadmap calls for sustained support from the U.S. federal government and prescribes that the government should focus on establishing an open AI platform, to open resources such as data and machine learning libraries, and to establish national AI competitions, national AI research centers, and AI curricula for the education system.
While China already is implementing its own approach to AI development, the intricate relationship between private sector enterprises and policymakers is likely to come under increased scrutiny as its economic and technological dispute with the United States continues. For China’s National AI Team, the political drive to establish AIOIPs may prove to have adverse consequences for these companies expansion in advanced markets. Nonetheless, the AIOIP initiative exemplifies an innovative approach to supporting the development of AI in a budding platform economy, while it remains to be seen just how effective the approach will be, in terms of spurring wider ecosystem participation and AI innovation.