The ground-breaking monumental piece of Wealth of Nation published in 1776 by Adam Smith paved the way for economics to distance itself from the moral science and established itself as an independent discipline. The gradual introduction of mathematics and statistics into this discipline further alienated it from other social sciences—often perceived as soft sciences for the wrong reason—and aligned itself with other commonly perceived hard sciences—also for the wrong reason, though—such as physics and biology.

The development of information science, computer technology and artificial intelligence since the mid-twentieth century has created further niche that economics can benefit from. The significance of such niche can be perceived in two ways. The first is the advancement in research methodology which establishes simulation as a third approach—other than the currently mainstream approaches of theory and history (i.e., empirical studies)—in the exploration of scientific knowledge. The second is the advancement in tackling the decades-old, if not centuries-old, problem of the interface between micro phenomenon and macro phenomenon—in economics, such an micro-macro interface issue can be best described as the microfoundation of macroeconomics. When humanity is perceived as a macro phenomenon waiting for exploration and explanation, human behaviors can then be programmed in ways—while relying on biological sciences and human experiments such as neuroeconomics and experimental economics for guidance—to the very bottom of each and every individual (i.e., agent) with minimum consideration of the human factors that administer our interactions with others. Simulated interactions among the agents can then be observed and compared with real life observations for further investigation. Moreover, on-line experiments conducted on real agents (i.e., human beings) can even be joined with the participation of software agents so that the “intelligence” of such software agents as compared with real agents can be determined and later be feedback into the development of future software agents.

Basically there is no limit on how this new approach can be applied and what area it can be applied to. When applied to social phenomena happening everyday surrounding everyone, it leads us to a new and different strand of social sciences called computational social sciences. The purpose of the following proposed NCCU Summer School on Computational Social Sciences, jointly proposed by faculty members from three different departments under the coordination of the AI-ECON Research Center of the Department of Economics, is exactly meant to propose and promote the development of computational social sciences in Taiwan. It is hoped that our efforts in introducing this area of research which is relatively new in Taiwan would broaden our understanding of it and invite more interested researchers to put efforts in it.

II. The Structure of the Summer School

A. Artificial Societies:

Herbert Simon, the winner of the Nobel Prize Award in Economics (1978), argued that social sciences are the ‘hard’ science in his 1987 publication “Giving the soft science a hard sell”. The social sciences are so hard because we often need to make some serious assumptions to simplify the complexity of our questions. So in the economic filed, we need to use the representative agent to conduct with the economic issues and then get the conclusions. But we could not deal with the problems of why we could use” representative agent” to consider our question? Do all persons really have the homogeneous utility function or preference? Do all people behave as the same routines?

Traditional thought in the social sciences (especially in the economics filed) always use the “top-down” framework, like above thought methods, to tell us how to solve or observe the things in the realistic economic environments. But all the situations in these sequences of ideas, we always just get so relative static and drab conclusions, not so realistic, and could not let us know more about what operating process does the real-world and how to conduct with them?

Accompanying with the developing of the computer technologies, now we could use the Agent-based architecture to uncover many black-box that we could not deal with before in the social science. Agent-based computing allows us to use the “bottom-up” framework and simulation to simulate what our real-world potentially to be and change. In the Agent-based architecture, we could use the computer to model the agent and their behavioral rule, and then start to study what the human social phenomena, including trade, migration, group formation, combat, interaction with an environment, transmission of culture, propagation of disease, population dynamics and so on. By the Agent-based architecture, it really enriches our abilities to deal with the questions with all kinds of possibility deeply.

In the 1996, Epstein and Axtell design the model which named “Sugarscape” to simulate the relatively simple CA or other “emergent” simulations which deal with one or two major social dynamics as a very general level like trading behaviors, food acquisition, or residential segregation. And through the variable network relationship between different agents, the model generates numerous data and finds the different situations with the artificial-life research. By those findings, they bring up some important conclusions in their book (1996,”growing artificial societies”). Beside this, recently there are so many applications to use the same thought and methods to observe the spread of the HIV pandemic or civilization collapse and so on.

By so many Agent-based applications with the topic of “artificial society”, it may be useful to let us enter into a brief thinking that what the function of the “artificial society?” and if we could let the society science become more not so hard by “artificial society?” or not?

B. Agent-Based Computational Finance:

The agent-based computational finance is a growing research field, and one of the popular fields to which agent-based modeling are applied. Agent-based modeling is to analyze evolving systems of autonomous interacting and heterogeneous agents. Starting from initial conditions, the computational model evolves over time as each of its agents repeatedly changes her behavior through interactions and learning. Simulations of the model produce the dynamics of the market as well as agents. As a result, we can possibly analyze how agents’ behavior is related to the market dynamics. The analysis of agent-based models is a bottom-up approach to the study of the markets to be analyzed.

Applications of such agent-based modeling to financial markets are particularly appealing for some reasons. First, such an applied field, often called Agent-based computational finance, provides a mechanics to understand empirical features in financial markets. It is well known that the financial time series have many interesting puzzles that are not fully understood. For example, the financial time series contains volatility clustering, fat tails, ARCH dependence of the returns, long-memory of trading volume, and so on. Those empirical features do not match traditional macroeconomics approaches. Agent-based computational finance has potential for solving those puzzles as shown in many papers (e.g., LeBaron et al. in Journal of Economic Dynamics and Control 1999). The models formulated in agent-based computational finance, which include agents’ heterogeneity and learning and adaptation mechanism, are well suited for explaining puzzles in financial markets.
The second reason is that financial markets provide plentiful datasets. For example, prices and volume series for more than 30 years at different frequency are normally available to researchers. In addition, we can possibly get experimental data so that we can use it for calibrating agent behavior. Those datasets make it possible for researchers to test and calibrate agent-based models on financial markets. For this reason, financial markets also fit for applying agent-based modeling.

C. Agent-Based Social Networks:

All human actions that involve the participation or interaction of two or more people are necessarily network phenomena in nature. Current researches suggest that there are some interesting issues or special characteristics that can be found in the social networks in our daily lives such as the small world phenomenon, the centrality issue, the effect of local network structure, and the dependency issue of a network-formation mechanism, etc. By looking at these phenomena, we are able to see how people interact with each other, how the chains of relations are formed and how other important human activities are coordinated and achieved.

The biggest problem in social network researches is the difficulty in getting data related to the “network relations” because this will often require the revelation of a lot of particular knowledge or even inside information pertaining to the network in concern. This problem becomes even worse when the network gets too large. To circumvent such difficulty, one can either look into the internet or adopt an agent-based approach in the research. The internet itself is a big social network and there is no limit on how big it can grow. Modern web technology in the so-called Web 2.0 or even Web 3.0 era enables us to trace the network relations from websites. Citations and cross-references so frequently appearing in web logs and databases provide the best example of how we can reestablish a social network from surfing the internet via. the application of software agents.

The concept of social network, its meaning when applied to the internet, and the software agent technology are then the three focal subjects that we would like to include in the summer school.