One of the main characteristics of humans is that they adapt their decision-making depending on their knowledge and perceptions of their environment or other people. This makes simulation and analyses of coupled human-environmental system (HES) complicated. Most models assume a static/linear relation between the environment and human decision-makers, often based on utility functions or simple static rules.
In reality, the perception and beliefs of people change continuously as result of feedbacks from their environment and others; opinion of others, discourse, trust and reputation prominently influence the decisions people take. This can be looked at as a (social) learning process.
To be able to create simulation of systems dominated by human-environmental interactions, it is essential to include change in behaviour through time. Agent Based Models (ABM) are able to represent individual behaviour of humans. In most ABM agents are represent with rather static behaviour. To create simulations, better capable of representing, machine-learning approaches seem promising. However currently it is unclear what approaches best suit the needs of agent based modelling of complex HES. In other words, how can we develop an ABM that includes a learning model to agents? Besides evaluating the success of their behaviour, agents also might include social concepts such as trust and reputation in their learning. Various models exists such as: reinforcement learning, Bayesian learning or neural network based approaches. As a case study, a simple agricultural system will be taken.
- Create a review of useful learning approaches for creating adaptive ABM of HES
- Develop an ABM of a simple HES that includes learning
- Evaluate the effects of learning on de outcomes of simulation compared to a simulation without learning
- Basic computer programming knowledge.
- Knowledge of Netlogo or GAMA is a pré.
Theme(s): Human – space interaction