Introduction to Computational Social Science

ion In computer science, abstractionmeans hiding information. In CSS, abstracting from the world “reality”—whether directly experienced (observing a riot downtown) or indirectly learning about it (reading history)—is a process involving stimulus signals, perceptions, interpretation, and cognition. CSS relies on several sources for abstracting key entities, ideas, and processes from raw stimulus signals from the real world. These sources span a hierarchy in terms of their social scientific status. At the very top of the hierarchy are social theories with demonstrable validity in terms of formal structure (internal validity) and empirical observation (external validity). Not all existing social theories meet these stringent requirements, although an increasing number of them do as research progresses. Examples of social theories that meet 10A little-known fact among many social scientists is that the theory of mechanics in physics is built around the abstraction of singleand two-body problems. Already three-body problems are hugely difficult by comparison; and, most interesting, N -body problems defy mathematical solution in closed form. 11Interestingly, humanistic fields such as music and ballet also use systems of specialized notation, far beyond what is used in traditional social science. In music, Guido d’Arezzo [b. A.D. 991 (or 992), d. 1050] is considered the founder of the modern music staff; in ballet, Rudolf von Laban [b. 1879, d. 1958] invented the symbolic system known as “labanotation” (Morasso and Tagliasco 1986). 50 2 Computation and Social Science internal and external validity standards include Heider’s Theory of Cognitive Balance in psychology, Ricardo’s Theory of Comparative Advantage in economics, and Downs’s Median Voter Theory in political science, among others. Social theories are abstractions that point to relevant social entities, variables, and dynamics that matter in understanding and explaining social phenomena. A second source of abstraction consists of social laws. Examples of social laws include the Weber–Fechner Law in psychometrics, the Pareto Law in economics, and Duverger’s Law in political science. Theories explain; laws describe (Stephen Toulmin 1967).12 Some of the most scientifically usefully social laws can be stated mathematically, as in these examples. Social laws also contain relevant entities, variables, and functional relations for describing social phenomena. A third source of abstraction consists of observations that can range from formal (e.g., ethnography, content analysis, automated information extraction, text mining, amongothers) to informal (historical narratives,media, andother sources about social phenomena). Observations of social phenomena can describe actors, their beliefs, social relations, and other features ranging from individual to collective. Finally, a fourth source of abstraction consists of computational algorithms capable of emulating social phenomena, as in artificial intelligence (AI). Artificial (i.e., not really human) algorithms do not claim to be causal in the same sense as social theories. They “work,” but without causal claims in the same sense as social theories. They are efficient, in the sense that they (sometimes) can closely replicate social phenomena. AI algorithms are typically (and intentionally) efficient and preferably simple; extreme parsimony in this case comes at the expense of realism. Examples of AI algorithms include Heatbugs (Swarm, NetLogo, MASON), Boids (Reynolds 1987), and Conway’s (1970) Game of Life. In spite of their lack of social realism, AI algorithms can be useful sources for abstracting social entities, ideas, or processes because they can highlight features that either elude theories or are hard to observe. An example would be the agglomeration patterns generated in a Heatbugs model, as a function of varying parameters of “social” interaction among the set of agents, or the role of apparent “leadership” in a flock of boids.