China issues its three-year action plan for "Internet Plus" AI

Background

In July 2015 the State Council issued its guidance on "Internet Plus", document 国发〔2015〕40号, which set out 11 key actions for connecting the internet with established industries. Artificial intelligence was one of those 11 actions.^[3] A State Council guidance of this kind sets a general direction; concrete targets, tasks and responsibilities are normally written into follow-up documents issued by the bodies that carry the work out. This plan was one of those follow-ups.

What was issued

On 18 May 2016 the National Development and Reform Commission (NDRC) issued the three-year action plan for "Internet Plus" artificial intelligence, document 发改高技〔2016〕1078号. It was drawn up jointly by the NDRC, the Ministry of Science and Technology, the Ministry of Industry and Information Technology and the Cyberspace Administration of China.^[1] The plan presents itself as a way of implementing the State Council's 2015 "Internet Plus" guidance.^[1] The "three-year" in the title refers to the period the plan covers: it is a schedule of actions to be carried out, not a standing strategy document.

The targets for 2018

The headline date in the plan is 2018, and the document states what should exist by then. First, a basic artificial intelligence resource and innovation platform. Second, an industrial system, a service system and a standards system that are at least initially in place. Third, a market application scale of 100 billion yuan.^[2]

The three targets are of different kinds: one concerns shared infrastructure, one concerns the industrial and regulatory environment, and one concerns how much money AI applications should be generating. Standards are named alongside industry and services, which places measurement and compatibility on the same list as commercial growth.

Data, computing and technology tasks

The plan treats data and computing power as things that have to be built, not merely used. It calls for large training data resources and public service platforms covering text, speech, image, video and map data, and for new computing clusters aimed at very large deep learning workloads.^[2] A training data resource is the labelled material a model learns from; a computing cluster is a group of machines working together on the same job. Both are inputs that a single firm may find expensive to assemble alone, which is why a plan can list them as shared infrastructure.

On the technology side, the plan names computer vision, intelligent speech processing, biometric recognition, natural language understanding, intelligent decision and control, and new forms of human-machine interaction.^[2] These are broad categories rather than single products: computer vision covers machines reading images, speech processing covers machines handling spoken language, and biometric recognition covers identifying people by face, voice or other physical traits.

The plan also lists application projects — smart homes, intelligent vehicles, unmanned systems and intelligent security — together with smart terminals, wearables and intelligent robots.^[2] Keeping technology tasks and products in one document links research subjects to things that can be sold and used.

Its place in policy history

The NDRC later described this document as China's first policy document in the artificial intelligence field.^[3] The same reply notes that the State Council's July 2015 "Internet Plus" guidance had already made artificial intelligence one of 11 key actions.^[3] Read together, the two statements locate the three-year plan fairly precisely: artificial intelligence had appeared as one action item inside a broad internet policy, and this plan took it out and gave it a document of its own, with a dated target, named technology directions and named product categories.