Research

Key topics:

Agent-Based Social Simulation, cognitive modelling and human behaviour, multi-agent systems, complex systems theory, cognition and interaction, context modelling.

Background:

For almost 40 years, Multi-Agent Systems (MAS) have been a central research area in contemporary artificial intelligence. Initially stemming from distributed artificial intelligence, research in MAS has been the subject of wide attention and this work has enabled successful transfers to industry for the benefit of the economy and society in general.

A multi-agent system consists of a number of intelligent agents that make decisions and interact in a shared environment. At the micro level, we are interested in the behaviour of agents over time, while at the macro level, we are interested in global emergent phenomena, such as mutual knowledge, cooperation, social intelligence, etc.

Modelling human behaviour:

The problem with models and simulations of socio-technical systems is their lack of a realistic focus on the human element, particularly human behaviours and human-like reasoning, including the apparent irrationality of human behaviour. I specialise in modelling aspects of human behaviour at the cognitive, individual, and societal levels, using an agent-based approach where my models are used for simulation. More broadly, my work is in the field of Agent-Based Social Simulation. From my background in Artificial Intelligence, I am primarily interested in cognition and interaction, and more specifically in the modelling of cognitive activities of human behaviour, the modelling of cognitive supports in our work environments, and the modelling of the interaction between different groups of people in order to accomplish a task.

The central questions I address in my research are:

  • How can we develop computational models of intelligent agents that are sufficient to represent real human behaviour?
  • How do agents interacting with each other and with their dynamically changing environment create the emergence of phenomena such as social intelligence and mutual awareness?
  • How can we apply our understanding of complex systems to socio-technical systems and how can we build the robustness of a system?
  • How do aspects such as altruism, social attachment, cognitive biases, or even risk aversion affect human behaviour and how can these be modelled by intelligent agents?
  • How does context awareness contribute to intelligent decision-making in complex social systems, and how can this be modelled?

Computer simulators and applications:

Specifically, I address the cognitive mechanisms behind human decision-making and develop formalisms to translate them into intelligent computational models. The challenges in this research are numerous:

  • Choosing the most appropriate reasoning architecture for intelligent agents. Among the numerous representations existing to equip an agent with intelligence are simple ‘if…then’ rules, neural networks, BDI (Belief-Desire-Intention) models, Markov chains, or game theory approaches, etc. The choice is strongly guided by the type of agent being modeled and its required level of intelligence.
  • One of the criticisms of agent-based models concerns their scalability. It is possible to simulate thousands of agents, but as the intelligence level of agents increases, scaling suffers, requiring the use of grid systems, parallel computing, etc.
  • What level of detail to include in a model is a complex choice. Any model is an abstraction of reality. It is not useful to include all the elements and various approaches exist, from very simple models (sometimes called trivial models) to very descriptive models – This is known in the community as the KISS to KIDS spectrum. This challenge consists in evaluating exactly which elements impact decision-making.
  • The results of a simulator are meaningless unless the model has been extensively validated. The question is whether the simulator results are representative of the real-world situation for the research question being addressed. Since we are dealing with complex systems, with their non-determinism, we cannot expect to be able to perfectly reproduce what we observe in the real world.

I have applied my work to many fields, but I am particularly interested in emergency and crisis management, human behaviour and mobility in Smart Cities and autonomous vehicles. My work is highly multidisciplinary, and given its nature, I work intensively with cognitive scientists, social geographers, anthropologists, civil engineers and seismologists, finance experts and emergency managers. In order to model human behaviour, I also draw heavily on social science theories, such as Hutchin’s theory of distributed cognition, Vygotsky’s activity theory, the work of Brown et al. on social intelligence, and Mawson’s theory of social attachment.