Interaction Models Group

Agents Chapter


In everyday life arguments are "reasons to believe and reasons to act". Until recent years, the idea of ``argumentation'' as the process of creating arguments for and against competing claims, was a subject of interest to philosophers and lawyers. In recent years, however, there has been a growth of interest in the subject from formal and technical perspectives in Computer Science (CS) and Artificial Intelligence (AI), and a wide use of argumentation technologies in practical applications. In Computer Science and Artificial Intelligence, argumentation is viewed as a mechanical procedure for interpreting events, organizing and presenting documents and making decisions about actions. From a theoretical perspective, argumentation offers a novel framework casting new light on classical forms of reasoning, such as logical deduction, induction, abduction and plausible reasoning, communication explanations of advice, supporting discussion and negotiation in computer-supported cooperative work, and learning. From a human-computer interaction point of view argumentation is a versatile technique that facilitates natural system behaviour and is more easily understood by human users and operators. Generally speaking, argumentation has the potential to add value to any computer-assisted system that provides information and advice to human users or other agents.


The methodology of meta-argumentation instantiates Dung's abstract argumentation theory with an extended argumentation theory, and it is thus based on a combination of the methodology of instantiating abstract arguments, and the methodology of extending Dung's basic argumentation frameworks with other relations among abstract arguments. The technique of meta-argumentation applies Dung's theory of abstract argumentation to itself, by instantiating Dung's abstract arguments with meta-arguments using a technique called flattening. We apply the meta-argumentation methodology on different challenges in formal argumentation: the support relation among arguments in bipolar argumentation, the graded trust evaluation by merging argumentation frameworks in multi-agent argumentation and coalition formation using argumentation theory.

Logics for Security

Neighborhood Semantics

Minimal modal logics based on neighborhood semantics are intersting for many reasons. First they are more general than standard Kripke-based logics, and second they have been deeply studied in their application to first-order modal logic. We are applying techniques of self-firbing to a first-order intuitionistic neighborhood logic to be used for specifying access control policies.

First Order Theorem Proving for Access Control

Recently, Martin Abadi pre-sented a propositional modal intuitionistic logic for access control by using the well known translation of intuitionistic logic into S4. I prove soundness and completness for a more general access control logic than the one presented in Abadi by relying on intuitionistic semantics (with no translation into S4), with this approach we could easly translate its semantics into first-order formulas and use state of the art theorem provers (i.e. SPASS) to reason in a fast way about access control policies.

Intuitionistic Conditional Logics

The semantics of the propositional fragment of FSL it is very similar to the one of conditional logics. With other researchers of the Logic Programming and Automated Reasoning group in Torino we are working on fomalizing an intuitionistic conditional logic for access control, this work is interesting because it introduces the possiblity to develop a sequent calculus to reason about proof-carrying authorization.

Application of FSL in access control meta-modelling

In collaboration with Prof. Steve Barker of King's College, we are using FSL as a language to describe a general meta-model of access control. The meta-model is able to describe well-known access control infrastractures, privacy and integrity mechanisms. This work studied the practical applicability of FSL for real world scenarios.

Neural Networks and Input/Output Logics

With Artur d'Avila Garcez, we are working on translating Input/Output knowledge bases into neural networks (NN) and on exploiting the learning machinery of NN to update knowledge represented by I/O logical formulas. The neural symbolic paradigm is one of the most successful approach in merging symbolic and numerical paradigms into a common computational architecture. I/O logic has lot of applications in non monotonic reasoning and can be used in dealing with preferences in databases of conflicting policies, having such a translation into NN would open the possibility to model neural controllers for access control.

Multi-Agent Systems

The research on Agent Models is split into three main research lines, not completely separated: normative multiagent systems, cooperative multiagent systems and languages for programming organizations of multiagent systems. This research line is based on a long-term cooperation with the Computer Science and Communications (CSC) of the University of Luxembourg in the person of Prof. Leendert van der Torre.

Normative Multiagent Systems (NorMAS)

Open multiagent systems can be regarded as distributed systems where (possibly) large, varying populations of agents exhibiting different (possibly deviating, or even fraudulent) behaviours interact. As in human societies, it seems necessary the need of regulatory structures establishing what agents are permitted and forbidden to do. Social concepts like norms are important for multiagent systems. In this regard, multiagent system research and sociology share the interest in the relation between micro-level agent behaviour and macro-level system effects, the relation between individual agent behaviour and characteristics at the level of the social system. For what concerns the formalization of the concept of norms for multiagent systems we proposed a definition of obligation based on sanctions: an agent who is subject to an obligation, in order to decide what to do, has to take into account the fact that if he does not fulfil the obligation may be sanctioned by a normative agent. The application of norms to regulate the behaviour of a multiagent system follows two different research lines: the first one is based on the use of dependence networks, a kind of network able to expressively represent the relationship among the agents while the second line consists in the application of methodologies coming from multi-agent systems context - i.e. game theory, normative systems - to the analysis of security problems related to business outsourcing.

Coalition Formation

Agents composing a multiagent system can form groups, called coalitions, with the aim to achieve their goals. Coalitions' formation issue is a classical problem of game theory where a coalition is interpreted as a group of players that can make a binding agreement to act as a cooperative unit. Our research line aims to propose a new kind of approach to coalitions' formation and evolution based on the theory of social power and dependence pioneered by Castelfranchi. The theory of social power and dependence is an attempt to transfer theories developed initially in the field of sociology to the field of multiagent systems and refine them. This theory models the potential interactions among the agents which lead to the achievement of a shared goal (cooperation) or the reciprocal satisfaction of their own goals (social exchange). An agent should be able not only to recognize the potential of a group of agents to help it in the satisfaction of a goal, but also to reason about how it can reward them. This involves the development of a social reasoning mechanism that analyzes the possibility to profit from mutual dependencies (two agents depend on each other for the satisfaction of a shared goal) or reciprocal-dependencies (two agents depend on each other for the satisfaction of two different goals). To reason and regulate coalitions' evolution, we are following the methods deriving from the field of argumentation theory and from normative multiagent systems.

Agent Programming Languages

In the last years, the organization metaphor assumed a relevant importance in informatics. An organization acts as a coordination artifact, in which each participant has to play a well-determined role. One of the perspectives by which roles can be viewed is that of the "affordances", by Gibson. Gibson refers to an ecological perspective, where animals and the environment are complementary, but the same vision can be transferred to objects, and agents. By "environment" we intend a set of objects and by "animal of a given specie" we intend another object of a given class which manipulates them. Objects have affordances when they are considered relative to an object managing them; in this way, we have that the properties characterizing an object in the environment depend on the properties of the interactant. The interaction possibilities of an object in the environment depend on the properties of the object manipulating it. The research on this kind of issues can be done both in Object Oriented programming, in web developing, and in Multi Agent Systems. In Object Oriented programming, the goal to be reach is the introduction of organizations and roles as first class entities into a well-known language (Java). These new features will help the programmers to simplify communication issues between objects, having, in this way, the possibility to concentrate on other problem-depending issues. In web developing, the introduction of organiztions and roles can leade to a new situation in which the interaction between users and web systems will be driven in an automatic way by the role that the user has to play in that particular environment. In Multi Agent Systems, one of the main problems is that no general purpose languages offering primitives for organizations and roles are available. Our aim is to improve a large use and general purpose environment (named JADE), in order to offer what is needed to model organizations, roles, and the interaction between them.