关键词: 人工智能, 社会影响, 实证研究, 数据采集, 科学测量

Abstract: Artificial intelligence(AI) is a disruptive technology leading the new round of scientific and technological revolution and industrial change. It has great impacts on individual psychological cognitions and behavioral preferences, social values and the social order while improving productivity. The social impact and governance of AI has become an important research topic in the world. The novelty and advance of AI issues make the cases and data available for the observation and analysis of the social impact of AI very limited. Therefore, the empirical study on evidence-based logic and realistic data of the social impact of AI is still relatively scarce. To conduct evidence-based analysis on the social impact of AI, Chinese scholars, political leaders and administrators have collaborated on the Artificial Intelligence Social Experiment (AISE). However, scholars and practitioners still have great perplexities and controversies about what experimental scenes should be selected and what data should be collected for the AISE. In this context, it is significant to further clarify the concept and the application scenario AI, and to summarize the different dimensions and representations of the social impact of AI, so as to accurately grasp and govern the real impact of AI.
By sorting out the application scenarios and impacts involved in 2905 Chinese and English articles from 2001 to 2020, AI application scenarios and their impacts can be divided into three dimensions: The first is the micro-level personal application to enhance the interaction between technology and people. The second is the middle-level application of industry and organization, which lead to the reform of industry and organization. The third is the macro-level application of urban and social governance to promote policy changes and institutional responses.
Based on the three dimensions, the application scenario, impacts and data observation interface of each dimension are further refined and the integrated analysis framework of the social impact of AI is proposed. At the micro level, personal application scenarios mainly include the accurate information push, the biometric identification and the man-machine fusion. Impacts include risk prediction, interest balance, value shaping, public opinion control, etc. Data observation interfaces include the personal interest hotspot, emotional perception, psychological dynamics, usage frequency, behavior traces, etc. At the middle level, application scenarios of industry and organization include the automatic vehicle, the mobile education, the precision medicine, the digital surveillance, etc. Impacts include the re-engineering of business processes,
rules and regulations, power lists, divisions of labor, responsibilities, etc. Data observation interfaces include changes in business processes, organizational networks, precision and efficiency, rules and procedures. At the macro level, application scenarios related to urban and social governance are the City Brain, the "Public Opinion Through Train", the "Visit Once" reform, emergency management system, the Industry Brain, etc. Impacts includes the social risk, the policy feedback, the public participation, the emergency response, the resource allocation, etc. Data observation interfaces include social risk cases, social hot issues, social networks, public opinion about policies, policy text changes, etc.
The analytical framework establishes the data observation, collection and evaluation system of the comprehensive impact of the application of AI on individuals, organizations and the society from the perspective of scientific measurement. It provides a reference for transforming vague notions into scientific variables with clear boundaries for measurement and analysis, so as to ensure the smooth progress of the social experiment work of AI and to achieve expected results.

Key words: artificial intelligence, social impact, empirical study, data collection, scientific measurement