To Örebro University

Simulation has become, in many application areas, a sine qua non. Most recently, COVID-19 has underlined the importance of simulation studies and limitations in current practices and methods. We identify four goals of methodological work for addressing these limitations. The first is to provide better support for capturing, representing, and evaluating the context of simulation studies, including research questions, assumptions, requirements, and activities contributing to a simulation study. In addition, the composition of simulation models and other simulation studies’ products must be supported beyond syntactical coherence, including aspects of semantics and purpose, enabling their effective reuse. A higher degree of automating simulation studies will contribute to more systematic, standardized simulation studies and their efficiency. Finally, it is essential to invest increased effort into effectively communicating results and the processes involved in simulation studies to enable their use in research and decision making. These goals are not pursued independently of each other, but they will benefit from and sometimes even rely on advances in other sub-fields. In this article, we explore the basis and interdependencies evident in current research and practice and delineate future research directions based on these considerations.

This paper studies the role of trade for the joint uptake of AI-enabled automation in manufacturing and engineering. It develops an agent-based model (ABM) where the agents are heterogeneous manufacturers and engineering firms. The ABM features two technology-related business models: engineering as a face-to-face consultancy service and engineering as automated software. The software adoption rate follows an S-shaped curve for manufacturers and a boom and bust cycle for engineers. In the early phase, shortage of engineers constrains AI uptake, while engineers become abundant when AI is fully adopted. Trade affects the cut-off productivity level at which manufacturers switch technology, the shape of the adoption rate curve, and the incentives for engineers to develop software. Bulky transactions and different productivity distributions across countries are drivers of trade in their own right.

Affordances are an important basis for many human-environment interactions such as navigation or geo-design. In this short paper we present an approach to modelling affordances based on treating affordances as emergent phenomena in an agent-based simulation. We use the notion of an affordance schema to represent the setting in which the emergence of an affordance is made possible. We use a case study to show that (unexpected) affordances emerge during the course of the simulation. While the general approach is promising and may be used for other emergent phenomena such as landmarks, we also acknowledge and discuss the problems incurred during the modelling process. The paper closes with a reflection and some ideas for future work.

Agent-based Simulation Modelling focuses on the agents' decision making in their individual context. The decision making details may substantially affect the simulation outcome, and therefore need to be carefully designed.

In this paper we contrast two decision making architectures: a process oriented approach in which agents generate expectations and a reinforcement-learning based architecture inspired by evolutionary game theory. We exemplify those architectures using a technology uptake model in which agents decide about adopting automation software. We find that the end result is the same with both decision making processes, but the path towards full adoption of software differs. Both sets of simulations are robust, explainable and credible. The paper ends with a discussion what is actually gained from replacing behaviour description by learning.

Q-Networks are used in Reinforcement Learning to model the expected return from every action at a given state. When training Q-Networks, external reward signals are propagated to the previously performed actions leading up to each reward. If many actions are required before experiencing a reward, the reward signal is distributed across all those actions, where some actions may have greater impact on the reward than others. As the number of significant actions between rewards increases, the relative importance of each action decreases. If actions have too small importance, their impact might be over-shadowed by noise in a deep neural network model, potentially causing convergence issues. In this work, we empirically test the limits of increasing the number of actions leading up to a reward in a simple grid-world environment. We show in our experiments that even though the training error surpasses the reward signal attributed to each action, the model is still able to learn a smooth enough value representation.

Navigation apps have become more and more popular, as they give information about the current traffic state to drivers who then adapt their route choice. In commuting scenarios, where people repeatedly travel between a particular origin and destination, people tend to learn and adapt to different situations. What if the experience gained from such a learning task is shared via an app? In this paper, we analyse the effects that adaptive driver agents cause on the overall network, when those agents share their aggregated experience about route choice in a reinforcement learning setup. In particular, in this investigation, Q-learning is used and drivers share what they have learnt about the system, not just information about their current travel times. Using a classical commuting scenario, we show that experience sharing can improve convergence times that underlie a typical learning task. Further, we analyse individual learning dynamics to get an impression how aggregate and individual dynamics are related to each other. Based on that interesting pattern of individual learning dynamics can be observed that would otherwise be hidden in an only aggregate analysis. 

New technology is of little use if it is not adopted, and surveys show that less than 10% of firms use Artificial Intelligence. This paper studies the uptake of AI-driven automation and its impact on employment, using a dynamic agent-based model (ABM). It simulates the adoption of automation software as well as job destruction and job creation in its wake. There are two types of agents: manufacturing firms and engineering services firms. The agents choose between two business models: consulting or automated software. From the engineering firms' point of view, the model exhibits static economies of scale in the software model and dynamic (learning by doing) economies of scale in the consultancy model. From the manufacturing firms' point of view, switching to the software model requires restructuring of production and there are network effects in switching. The ABM matches engineering and manufacturing agents and derives employment of engineers and the tasks they perform, i.e. consultancy, software development, software maintenance, or employment in manufacturing. We find that the uptake of software is gradual; slow in the first few years and then accelerates. Software is fully adopted after about 18 years in the base line run. Employment of engineers shifts from consultancy to software development and to new jobs in manufacturing. Spells of unemployment may occur if skilled jobs creation in manufacturing is slow. Finally, the model generates boom and bust cycles in the software sector.

Modelling agent-environment interactions in an agent-based simulation requires careful design choices. Selecting an interaction partner forms an often neglected, but essential element.

In this paper we introduce affordance schemata as an element of agent-based simulation models. We describe how affordances can be generated based on them during a running simulation to capture action potential that an interaction partner offers. We illustrate the introduced concepts with a small proof-of-concept implementation.