Final Report Summary - NORMS4SRA (Norm-Governed Self-Organised Systems for Sustainable Resource Allocation.)
The main objective of the project was to develop a probabilistic rule-based argumentation framework for investigating norm-governed self-organised systems, with respect to two lines of research:
* the construction of theoretical models about natural agents to forecast how people will behave or would behave under particular circumstances with regard to a particular technical infrastructure and normative system;
* the design of operational models for building norm-governed self-organising systems which have certain required properties.
To achieve this objective, a layered framework has been developed, where each layer addresses some requirements by integrating techniques layer-by-layer.
The first layer and second layer concerns the use of an intuitive rule-based argumentative logic to represent and reason in dialectic manner upon the represented systems. T his argumentative framework shall also facilitate the communication and any update of models of the investigated systems, that is essential to good science.
The third layer integrates probability theory to the argumentative framework in order to capture uncertainty (as understood in probability theory). The fourth layer accounts for learning aspects of the probabilistic argumentation framework, so that reinforced learning agents can be seamlessly investigated, and models can be learned and reproduced from facts. The fifth layer focuses on norms controlling or emerging in societies of learning agents, and the last layer deals with the societies where agents can govern themselves, by transfiguring their own experiences into deontic provisions guiding agents' decision making.
A proof-of-concept multi-agent simulation plate-form based on this layered framework has been developed to investigate populations of norm-governed learning agents, by directly animating formal rule-based specifications, thereby eliminating time-consuming, error-prone and ad-hoc implementation of normative systems. This tool allows fast prototyping while precisely represent and communicate our models of agents and norms, and (defeasibly) reason upon them.
The framework and its multi-agent simulation plate-form offers a complementary alternative to current equation-based and game theoretical models in Law and Economics, allowing more insights in complex normative systems, with thus important impact in policy making. (e.g. helping policy makers design the appropriate rules to achieve some goals),. However, its development belongs to longer term research agenda, which goes beyond the fellowship. In this regard, the framework is still under investigation, within a cooperation with the Law department in the University of Bologna (Italy), the Computer Sciences department at Imperial College (UK), and the National ICT Australia (NICTA).
Though the probabilistic rule-based argumentation framework was meant to be applied in models of resource allocation and in particular smart grids, the Fellow acknowledged a similar framework (learning argumentative agents) applied to smart grids investigated at the Department of Computer Sciences at Imperial College, and, considering its limits in terms of learning and validation, the Fellow decided to move on a more powerful formalisation which led to the investigation of neuro-argumentative systems.
Whilst the probabilistic rule-based argumentation framework was initially meant to investigate norm-governed self-organised systems, and because a formal and abstract perspective was taken, the Fellow has been able to build a probabilistic argumentative framework that is not only applicable to investigate norm-governed self-organised systems, but to the development of neuro- argumentative systems combining the strengths of neural networks and the reasoning and explanatory power of logic-based formalisms. This has materialised by seamlessly combining the probabilistic graphical model of Boltzmann machines with abstract argumentation.
By pioneering this probabilistic graphical model of neuro-argumentative systems, the project paves the way to many innovative researches and applications, leading to wider societal implications, from next generation of legal expert systems (e.g. to advise people from the civil society on legal matters) to learning reasoning engines at the heart of the Internet of Things (e.g. by on the fly learning patterns of events and giving explanations of events).
The work done so far on neuro-argumentative systems has been to lay some of its foundations.
However, the potential for other work is very large, and during this project we have taken some important and significant foundational steps into a new line of research.
* the construction of theoretical models about natural agents to forecast how people will behave or would behave under particular circumstances with regard to a particular technical infrastructure and normative system;
* the design of operational models for building norm-governed self-organising systems which have certain required properties.
To achieve this objective, a layered framework has been developed, where each layer addresses some requirements by integrating techniques layer-by-layer.
The first layer and second layer concerns the use of an intuitive rule-based argumentative logic to represent and reason in dialectic manner upon the represented systems. T his argumentative framework shall also facilitate the communication and any update of models of the investigated systems, that is essential to good science.
The third layer integrates probability theory to the argumentative framework in order to capture uncertainty (as understood in probability theory). The fourth layer accounts for learning aspects of the probabilistic argumentation framework, so that reinforced learning agents can be seamlessly investigated, and models can be learned and reproduced from facts. The fifth layer focuses on norms controlling or emerging in societies of learning agents, and the last layer deals with the societies where agents can govern themselves, by transfiguring their own experiences into deontic provisions guiding agents' decision making.
A proof-of-concept multi-agent simulation plate-form based on this layered framework has been developed to investigate populations of norm-governed learning agents, by directly animating formal rule-based specifications, thereby eliminating time-consuming, error-prone and ad-hoc implementation of normative systems. This tool allows fast prototyping while precisely represent and communicate our models of agents and norms, and (defeasibly) reason upon them.
The framework and its multi-agent simulation plate-form offers a complementary alternative to current equation-based and game theoretical models in Law and Economics, allowing more insights in complex normative systems, with thus important impact in policy making. (e.g. helping policy makers design the appropriate rules to achieve some goals),. However, its development belongs to longer term research agenda, which goes beyond the fellowship. In this regard, the framework is still under investigation, within a cooperation with the Law department in the University of Bologna (Italy), the Computer Sciences department at Imperial College (UK), and the National ICT Australia (NICTA).
Though the probabilistic rule-based argumentation framework was meant to be applied in models of resource allocation and in particular smart grids, the Fellow acknowledged a similar framework (learning argumentative agents) applied to smart grids investigated at the Department of Computer Sciences at Imperial College, and, considering its limits in terms of learning and validation, the Fellow decided to move on a more powerful formalisation which led to the investigation of neuro-argumentative systems.
Whilst the probabilistic rule-based argumentation framework was initially meant to investigate norm-governed self-organised systems, and because a formal and abstract perspective was taken, the Fellow has been able to build a probabilistic argumentative framework that is not only applicable to investigate norm-governed self-organised systems, but to the development of neuro- argumentative systems combining the strengths of neural networks and the reasoning and explanatory power of logic-based formalisms. This has materialised by seamlessly combining the probabilistic graphical model of Boltzmann machines with abstract argumentation.
By pioneering this probabilistic graphical model of neuro-argumentative systems, the project paves the way to many innovative researches and applications, leading to wider societal implications, from next generation of legal expert systems (e.g. to advise people from the civil society on legal matters) to learning reasoning engines at the heart of the Internet of Things (e.g. by on the fly learning patterns of events and giving explanations of events).
The work done so far on neuro-argumentative systems has been to lay some of its foundations.
However, the potential for other work is very large, and during this project we have taken some important and significant foundational steps into a new line of research.
