Order this book
University of Groningen
Since 1998, the Multi-Agent-Based Simulation (MABS) Workshop series has been bringing together researchers interested in MAS engineering with researchers focused on finding efficient solutions to model complex social systems in social sciences. The 19th International MABS Workshop took place in Stockholm in July 2018. It included the 17th International Conference on Autonomous Agents and Multiagent Systems (AAMAS), the 35th International Conference on Machine Learning (ICML), the 27th International Joint Conference on Artificial Intelligence (IJCAI), and the 23rd European Conference on Artificial Intelligence (ECAI). The book mirrors the interdisciplinarity and internationality of the 2018 MABS Workshop. It consists of 11 individual papers authored by 38 scholars representing 20 universities around the globe.
The book starts with Bruce Edmonds and Lia ní Aodha’s reflection on a public policy advisor’s dream to be heard. The extent to which that dream is fulfilled, among other factors, depends on the methodology the analyst is using. Scholars representing MABS entertain a privileged position, as present evidence-based trends in policy-making hold this particular method in high regard. The goal of being listened to is often achieved. But what happens if the praised predictions are wrong? Authors, putting things into perspective, contemplate on the role of MABS in the policy-making process.
Paper 2 investigates the systemic effects of cheating in two ideal type cases of English East India Company and Armenian merchants of New-Julfa, long-distance traders active during the 17th and 18th centuries. The authors utilise a popular paradigm in economics, i.e. the principal-agent problem, to set up decision-making for individual merchants (agents) and the company (principal) in an agent-based model. Results of the simulations shed light on combinations of characteristics and strategies, which enhance company performance. Moreover, the reader can find out if there are any strategies, which also make the individual merchants better off.
“Indirect Influence Manipulation with Partial Observability” (paper 3) is a search for an analytical solution to seed selection in studies of diffusion over social networks. The paper considers an abstract case of multiple concepts, inhibiting or boosting each other, spreading through a network. Naturally, successful propagation depends on the characteristics of the node(s) that start to diffuse the concept. However, efficient node selection becomes problematic when information about the entire population is absent, and one must rely on a sample. Target Degree heuristic, which ranks nodes by the number of neighbours with the target concept active, comes to the rescue. As an abstract mathematical solution, it can be successfully implemented for a variety of real-world problems such as diffusion of opinions, behaviours or infections.
Paper 4, authored by Andre Jalbut and Jaime Simão Sichman, is an interesting example of how board games can be used as an inspiration for agent-based studies. The study uses a setup of the Beer Game in an agent-based model designed to study supply chain management. Simulations shed light on the effects of peer recommendations and lying on the performance of regular and greedy players.
The PASHAMAMA model (paper 5) was developed by a team of scholars representing an international research partnership determined to push spatially-explicit agent-based modelling to another level. GAMA modelling platform was explicitly designed to handle large sets of data, including GIS. With this new publication, the reader can judge for him/herself how well can a large-scale agent-based model implemented in GAMA handle qualitative, process-based data. If you fancy programming in Java rather than in GAML, MASON’s development team has some good news, as they plan a series of extensions to the open-source simulation toolkit (paper 6). The goal for the next three years is ambitious and includes among others distributed simulation, distributed GIS, sensitivity analysis and educational aids. Fingers crossed!
The remaining five papers describe ABM implementations. Nuno Trindade Magessi and Luis Antunes (paper 7) compare two team-building styles: the goal-oriented American and the process-oriented German, in terms of the resource demand. In paper 8, Martin Mocko and Jakub Ševcech show how agent-based modelling can be used to simulate probable bank transaction data. Artificially generated dataset and an actual bank transaction dataset are subsequently analysed with the use of Anti Money Laundering classification algorithms. As it turns out, the three fraudulent strategies simulated in an ABM adequately resemble the structure of frauds in the Czech Republic. Following (paper 9), researchers from Laboratoire d’Informatique et de Mathématiques (University of Reunion Island) present temporality model scheduling, which has advantages over classical scheduling approaches (i.e. time-stepped approach, the event-driven approach and the mixed approach), as it adapts to the specificities of the simulated model. Next, Gabriel Santos, Tiago Pinto and Zita Vale propose a set of ontologies that enable interoperability between different types of agent-based simulators representing parts of an energy decision-support system (paper 10). The book concludes with Akira Tsurushima reproducing a social phenomenon of human herding during evacuations with the use of assumptions from the response threshold model used in biology (paper 11).
To sum up, the book covers a wide range of topics, from a meta-level reflection on the role of multi-agent-based modelling in public policy analysis and evaluation, through papers focused on software development and application, to agent-based models applied to optimisation problems or exploration of causal effects. Even though the reader can experience a lack of smooth transition between chapters, individual papers clearly reflect the authors’ passions and therefore most often make for a good read.
Return to Contents of this issue
© Copyright JASSS, 2020