Hybrid attacks coordinate the exploitation of vulnerabilities across domains to undermine trust in authorities and cause social unrest. Whilst such attacks have primarily been seen in active conflict zones, there is growing concern about the potential harm that can be caused by hybrid attacks more generally and a desire to discover how better to identify and react to them. In addressing such threats, it is important to be able to identify and understand an adversary’s behaviour. Game theory is the approach predominantly used in security and defence literature for this purpose. However, the underlying rationality assumption, the equilibrium concept of game theory, as well as the need to make simplifying assumptions can limit its use in the study of emerging threats. To study hybrid threats, we present a novel agent-based model in which, for the first time, agents use reinforcement learning to inform their decisions. This model allows us to investigate the behavioural strategies of threat agents with hybrid attack capabilities as well as their broader impact on the behaviours and opinions of other agents.

This model has been developed together with the publication ‘Modelling Value Change - An Exploratory Approach’

Value change and moral change have increasingly become topics of interest in the philosophical literature. Several theoretical accounts have been proposed. Such accounts are usually based on certain theoretical and conceptual assumptions and their strengths and weaknesses are often hard to determine and compare, also because they are based on limited empirical evidence.

We propose that a step forward can be made with the help of agent-based modelling (ABM). ABM can be used to investigate whether a simulation model based on a specific account of value change can reproduce relevant phenomena. To illustrate this approach, we built a model based on the pragmatist account of value change proposed in van de Poel and Kudina (2022). We show that this model can reproduce four relevant phenomena, namely 1) the inevitability and stability of values, 2) how different societies may react differently to external shocks, 3) moral revolutions, and 4) lock-in.

A flexible framework for Agent-Based Models (ABM), the ‘epiworldR’ package provides methods for prototyping disease outbreaks and transmission models using a ‘C++’ backend, making it very fast. It supports multiple epidemiological models, including the Susceptible-Infected-Susceptible (SIS), Susceptible-Infected-Removed (SIR), Susceptible-Exposed-Infected-Removed (SEIR), and others, involving arbitrary mitigation policies and multiple-disease models. Users can specify infectiousness/susceptibility rates as a function of agents’ features, providing great complexity for the model dynamics. Furthermore, ‘epiworldR’ is ideal for simulation studies featuring large populations.

This model was designed to study resilience in organizations. Inspired by ethnographic work, it follows the simple goal to understand whether team structure affects the way in which tasks are performed. In so doing, it compares the ‘hybrid’ data-inspired structure with three more traditional structures (i.e. hierarchy, flexible/relaxed hierarchy, and anarchy/disorganization).

Implementation of Milbrath’s (1965) model of political participation. Individual participation is determined by stimuli from the political environment, interpersonal interaction, as well as individual characteristics.

An agent-based framework to simulate the diffusion process of a piece of misinformation according to the SBFC model in which the fake news and its debunking compete in a social network. Considering new classes of agents, this model is closer to reality and proposed different strategies how to mitigate and control misinformation.

A simple model is constructed using C# in order to to capture key features of market dynamics, while also producing reasonable results for the individual insurers. A replication of Taylor’s model is also constructed in order to compare results with the new premium setting mechanism. To enable the comparison of the two premium mechanisms, the rest of the model set-up is maintained as in the Taylor model. As in the Taylor example, homogeneous customers represented as a total market exposure which is allocated amongst the insurers.

In each time period, the model undergoes the following steps:
1. Insurers set competitive premiums per exposure unit
2. Losses are generated based on each insurer’s share of the market exposure
3. Accounting results are calculated for each insurer
…

This is a simulation of an insurance market where the premium moves according to the balance between supply and demand. In this model, insurers set their supply with the aim of maximising their expected utility gain while operating under imperfect information about both customer demand and underlying risk distributions.

There are seven types of insurer strategies. One type follows a rational strategy within the bounds of imperfect information. The other six types also seek to maximise their utility gain, but base their market expectations on a chartist strategy. Under this strategy, market premium is extrapolated from trends based on past insurance prices. This is subdivided according to whether the insurer is trend following or a contrarian (counter-trend), and further depending on whether the trend is estimated from short-term, medium-term, or long-term data.

Customers are modelled as a whole and allocated between insurers according to available supply. Customer demand is calculated according to a logit choice model based on the expected utility gain of purchasing insurance for an average customer versus the expected utility gain of non-purchase.

This is an agent-based model of a simple insurance market with two types of agents: customers and insurers. Insurers set premium quotes for each customer according to an estimation of their underlying risk based on past claims data. Customers either renew existing contracts or else select the cheapest quote from a subset of insurers. Insurers then estimate their resulting capital requirement based on a 99.5% VaR of their aggregate loss distributions. These estimates demonstrate an under-estimation bias due to the winner’s curse effect.

Confirmation Bias is usually seen as a flaw of the human mind. However, in some tasks, it may also increase performance. Here, agents are confronted with a number of binary Signals (A, or B). They have a base detection rate, e.g. 50%, and after they detected one signal, they get biased towards this type of signal. This means, that they observe that kind of signal a bit better, and the other signal a bit worse. This is moderated by a variable called “bias_effect”, e.g. 10%. So an agent who detects A first, gets biased towards A and then improves its chance to detect A-signals by 10%. Thus, this agent detects A-Signals with the probability of 50%+10% = 60% and detects B-Signals with the probability of 50%-10% = 40%.
Given such a framework, agents that have the ability to be biased have better results in most of the scenarios.