Conclusions and further considerations
Given the novelty of these systems and the lack of first-hand evidence, more research is needed to understand how different AI applications affect the provision of public services in China.
A cursory overview of the field allows us to make several observations:
- The Chinese government places a heavy emphasis on the development of AI in general, and for the provision of public services in particular. AI is meant to improve the functioning of the existing institutions, not replace or reform them.
- National-level policies tend to focus on technology development, while local governments (for the most part) can choose their own pilots.
- Policies play an important role in agenda setting at national, local and even company levels. Most policies focus on speeding up technology development, data collection and implementing pilots. Issues like risk management, data privacy and accountability are coded in policy, but appear to be secondary to the development imperative.
- Individual projects take the form of public-private partnerships and are led by big technology companies rather than SMEs or startups, at least in the case of flagship initiatives.
- A flagship project may involve many different applications of AI. Some of these applications are straightforward and are in line with global best practices; others raise significant risks to privacy and may have built-in biases.
- Some of the issues related to the development and deployment of AI pertain to values and moral choices and could be less amenable to simple cost-benefit analysis.
Given these observations, we can briefly discuss the implications and possible lessons for the UK and the EU. From a policy perspective, the Chinese smart city policy model has traits of both traditional industrial policy and open methods of co-ordination. The industrial policy component is primarily focused on technology development, while localities are mostly free to choose their own pilot projects in line with a softer approach that resembles the open method of co-ordination, as Jeff Ding explores in his essay - with the notable exception of public security applications of the smart city technologies, which appear to be mandatory rather than optional.
One important lesson of the Chinese policy model is its attempt to reduce policy fragmentation and to nudge (or sometimes force) local governments to develop smart city plans that can be later used to assess progress and induce competition between different regions and localities. While most Chinese policies are rather context-specific and cannot (and should not) be directly transferred to other countries, the overall focus on policy consistency and competition between different localities is an important element that might render further examination.
The second consideration refers to specific technologies and applications. While China has undoubtedly made very significant strides in AI and machine learning technologies over the past few years, most AI technologies that appear in its flagship smart city projects probably couldn’t be described as cutting edge. At the same time, the development of integrated systems with multiple applications of well-known technologies and the ability to scale up rapidly (and cheaply) show that successful development and deployment of machine learning technologies need not be conditional on being at the technological frontier. In fact, a skilful deployment of proven commercial technologies could significantly improve the provision of public service under the right circumstances.
The third consideration refers to the model of interaction between the government and the private sector. The Chinese model features partnerships between large players and local governments, usually with political approvals from the higher level to facilitate interagency co-ordination. When the technologies are sufficiently developed in the domestic market, the government may help domestic firms export its technologies in foreign markets through the Digital Silk Road and other similar initiatives. This approach generally favours large firms and leaves relatively little room for SMEs and startups. The benefits of this approach may include speed of deployment and economies of scale, possibly at the cost of increasing the influence of big tech and suppressing smaller players. This approach is markedly different from the more decentralised practices in Europe, and these differences are likely to persist.
Finally, the issues of data privacy and security and the real (and perceived) differences between Europe and China in this respect have been extensively discussed elsewhere, and are addressed by Danit Gal in this collection. We will only note that learning from actual data (for example, in the case of traffic management or, more recently, epidemic control) may provide unique insights and capabilities that cannot be acquired from synthetic data or human observation. The final decisions about the extent of data collection will be context specific and should involve the relevant stakeholders from the government, industry and civil society.