Reviewed by
Christopher Watts
Ludwig-Maximilians University Munich
In a contrast to the then fashionable approach of “Expert Systems”, the sociologist, Randall Collins once proposed that a computer with human-like intelligence would need to be capable of engaging in conversations with human beings (Collins, 1992). For this, he argued, it would need to recognise and emulate human emotions. Since the 1990s, many computer scientists have become interested in the tasks of understanding and interacting with human emotions via computer programs, under the title of “Affective Computing” (Calvo et al., 2015; Picard, 2000), or the broader term, “Human-Computer Interaction”. Though the new field made no reference to sociologists of emotions like Collins, it does combine research from many disciplines, especially psychology, linguistics and robotics. Alexander Osherenko’s new book is a contribution to this field, with a particular interest - in an age of globalisation - on simulating emotive interactions between people from different cultures.The field promises much. Behavioural scientists can learn much about humans from the challenges of simulating them. The design of computer programs that interact with humans might be improved if given more appreciation for human emotions, as anyone experiencing automated customer services will testify. Meanwhile, those building social simulations will welcome thoughts on how to represent social agents more realistically, especially when non-technical stakeholders have to comment on these representations. For these reasons, the topic seems very appealing.
Sadly, this reviewer found the book falling far short of its subtitle: “A Practical Guide”. The overall impression is of a dissertation published too early by Springer. With more time, the same author might have produced either a step-by-step how-to guide, or a more analytical and critical survey of the field and its challenges. Chapter 1 defines a few terms, without recognising how contentious the meanings of “globalisation” and “culture” can be in social sciences. Chapter 2’s literature review contains summaries of books and papers from multiple relevant disciplines (sociology, psychology, linguistics, robotics etc.), but reads like a publisher’s catalogue abstracts, with little attempt at overall narrative or themes, or comparing and contrasting. JASSS readers may be puzzled by the chapter’s claim to have covered “social simulation”, since the ten pages of references for this chapter contain just one JASSS paper (on UML). Where are the classic social simulation models?
Chapter 3 suggests example social situations (“meeting someone for the first time”, “negotiating”, “apology”, …), but I found the pseudocode interpretations of these too trivial to be useful. For instance, at the end of “Algorithm 3.10”, the line, “8: calculate_outcome(emotion, culture, personality, context, history, situation)“, is only a placeholder for an algorithm. As a social simulation programmer, I want to know which computation or function to perform: what attribute changes will my agents and their environment undergo as a result of this interaction event?
JASSS readers may be more interested in Chapter 4, which describes two methods for collecting empirical data on human emotions during interactions. This chapter partly reports work already carried out, and partly suggests how data might be collected in future studies. I needed more details here, especially on actual experiences from using the methods. How practical is the example five-emotional-state Hidden Markov Model, given that it requires the populating of a 25-cell transition probability matrix? When might a researcher get sufficient data for one subject person in one type of context, or even for groups of people in a mix of contexts? And this five-state model seems very crude: it would classify both “Shame”, “Anger” and “Fear” by the one state, “High Negative”. When is this adequate to the demands of understanding human interactions? In addition, the approach divorces the emotions from their contexts and triggers. What Collins called “cultural capital”, that is, what was said during conversation and the repertoire for further interactions, is not represented. How large a repertoire would be needed for a particular application? How easy would it be to acquire?
Chapter 5 presents a data framework, and Chapter 6 a prototype architecture for social simulation, with some code in JAVA and JADE. Computer scientists may find these more interesting than this social scientist did. The final chapters, on evaluating the prototypes and conclusions, left me still unsure what applications the techniques might have. More reports from actual projects and specific cases, and less abstraction and “generality” might have made for a more useful book.
So we will have to wait a little longer for a truly practical guide that leads to social simulations. But the field of Affective Computing does have something to offer JASSS readers, and as a starting point this reviewer would direct them to the recent Oxford Handbook (Calvo et al., 2015) instead.
ReferencesCOLLINS, R. (1992). Sociological Insight: An Introduction to Non-Obvious Sociology, 2nd ed. Oxford: Oxford University Press.
PICARD, R. W. (2000). Affective Computing. Cambridge, Mass.; London: MIT Press.
Return to Contents
of this issue