This paper proposes and evaluates a method for interpreting hierarchical graph neural network (GNN) nodes created from criminal law texts. We optimise a hierarchical GNN with extra loss functions to generate coherent abstract nodes from Brazilian habeas corpus data. Four concept grounding (CG) methods (utilising semantic similarity and large language models (LLMs)) are evaluated for giving names to these nodes. Quantitative results demonstrate that optimisation improves the quality of abstractions and predictive performance. Qualitative expert evaluation favoured labels based on LLMs. This dual approach (optimisation and grounding) provides a framework towards more trustworthy and explainable legal AI systems.

This paper presents the Concil-IA project, an initiative exploring how machine learning (ML) can promote self-composition in conflict resolution. The research investigates whether an ML model can predict judicial outcomes and assist parties in reaching agreements during end-procedural conciliation hearings. We collected judgments (textual documents) from the Brazilian Special Civil Court at the Federal University of Santa Catarina (JEC-UFSC), focusing on disputes in air transport services (Consumer Law) where parties require compensation value for immaterial damage. We developed an ML model following a structured pipeline, including preprocessing, data representation and labeling, model training and testing using regression algorithms, and practical application in the legal domain. The model was validated in two phases: (i) presenting predicted judgment values to parties during conciliation hearings for new, similar cases; and (ii) comparing actual judgment values with the model’s predictions after the proceedings concluded. The model showed satisfactory performance with minimal deviations. Predicted values were well received by parties and their legal representatives and closely aligned with actual judgments. We conclude that an ML model can effectively support self-composition in conflict resolution, reduce procedural uncertainties, and increase the likelihood of consensual settlements in judicial disputes.

Abstract As the complexity of software systems rises, explainability - i.e. the ability of systems to provide explanations of their behaviour - becomes a crucial property. This is true for any AI-based systems, including autonomous systems that exhibit decisionmaking capabilities such as multi-agent systems. Although explainabil- ity is generally considered useful to increase the level of trust for end-users, we argue it is also an interesting property for software engineers, developers, and designers to debug and validate the system’s behaviour. In this paper, we propose a multi-level explainability framework for BDI agents to generate explanations of a running system from logs at different levels of abstraction, tailored to different users and their needs. We describe the mapping from logs to explanations, and present a prototype tool based on the JaCaMo platform which implements the framework.

Low-power sensors are becoming ever more powerful, increasing both their energy efficiency as well as their processing capabilities. Much work in recent years has focused on optimizing machine learning models to low-power systems, typically to locally process sensor data. Significantly less attention has been paid to other artificial intelligence fields such as knowledge representation and automated reasoning, which may contribute to building autonomous devices. In this work, we present a low-power sensor node with an autonomous belief-desire-intention agent. This kind of agent simplifies the implementation of both proactive and reactive behaviors, promoting autonomy in our target applications. It does so by locally perceiving and reasoning, and then wirelessly broadcasting an intention, which can be forwarded to an actuator. The capabilities of the autonomous agent are demonstrated with a light-control application. Experiments demonstrate the feasibility of running intelligent agents in low-power platforms with little overhead.

With the advent of the Internet of Things, new devices feature advanced capabilities that are used in homes, rooms, and offices for better comfort and to save user's time and energy. A complete smart space control system should automatically adjust settings to the user preferences based on data previously collected from such smart things. This paper describes a system that uses environmental data collected from sensors in smart devices to define contexts and user preferences to accomplish this requirement. Contextual data is fed to a context-aware decision engine, composed by a combination of machine learning and data mining techniques. The context-aware decision engine is capable of identifying important data relationships. This allows the use the environment's contextual information to make intelligent control decisions and adjust the environment's settings to user's preferences and reduce the overall power consumption, yielding better services to the user and improving human-technology interaction. We implemented and evaluated the system through a real case study. Results confirm the system's ability to automatically control a smart room with little overhead and latency, while promoting user comfort and energy savings.

This paper highlights the use of situated artificial institution (SAI) within a hybrid, interactive, normative multi-agent system to regulate human collaboration in crisis management. Norms regulate the actions of human actors based on the dynamics of the environment in which they are situated. This dynamics results from both environment evolution and actors’ actions. Our objective is to situate norms in the environment in order to provide a context-aware crisis regulation. However, this coupling must be a loose one to keep both levels independent and easyto-change in order to face the complex and changing crisis situations. To that aim, we introduce a constitutive level between environmental and normative states providing a loose coupling of normative regulation with environment evolution. Norms are thus no more referring to environmental facts but to status functions, i.e., the institutional interpretation of environmental facts through constitutive rules. We present how this declarative and distinct SAI modelling succeeds in managing the crisis with a context-aware crisis regulation.

This article presents a proposal for a multi-agent system for controlling the production of printed circuit boards in small series based on a new multi-agent approach. The multi-agent system was designed following the methodology Prometheus AEOlus and implemented using the framework JaCaMo. Unlike other multi-agent system used to control the production of printed circuit boards that are based on the agent-oriented paradigm, this system was completely designed and implemented using an integrated platform that follows the multi-agent oriented paradigm: the framework JaCaMo. So far this is the only implementation of this approach in the small series production.

In many logic-based BDI agent programming languages, plan selection involves inferencing over some underlying knowledge representation. While context-sensitive plan selection facilitates the development of flexible, declarative programs, the overhead of evaluating repeated queries to the agent's beliefs and goals can result in poor run time performance. In this paper we present an approach to multi-cycle query caching for logic-based BDI agent programming languages. We extend the abstract performance model presented in (Alechina et al. 2012) to quantify the costs and benefits of caching query results over multiple deliberation cycles. We also present results of experiments with prototype implementations of both single- and multi-cycle caching in three logic-based BDI agent platforms, which demonstrate that significant performance improvements are achievable in practice.

The dynamic conditions of global markets force manufacturers to invest in flexible production strategies to cope with demanding clients and still survive in a competitive economic scenario. In this sense, small series production appears as a trend for many manufacturing niches and brings many challenges regarding manufacturing and quality assurance aspects. Investing in production flexibility implies increasing production control complexity and planning. This flexibility usually does not correlate with higher degrees of manufacturing automation or with quality assurance strategies. The concept of Cognitive Metrology strives for handling the challenging automation and quality inspection requirements of small series production with a new approach based on self-optimizing systems. This paper introduces the concepts of self-optimization and Cognitive Metrology and focuses especially on a multiagent-based approach for supporting flexible automation and quality assurance in small series production, as a basis for the development of the Cognitive Metrology technology. Initial results of the application of this approach into industrial prototypes are introduced and discussed as well as the migration of this system to different industrial scenarios.

Jason is a platform for agent-based software development that is characterised both by being based on a programming language with formal semantics as well as having many language and platforms features that are very useful for practical programming, but not fully formalised. In this paper, we make significant progress in the direction of formalising the aspects of the variant of AgentSpeak that is interpreted by Jason that were not included in previous work on giving formal semantics to AgentSpeak. In particular, we give semantics to the plan failure handling mechanism which is unique to Jason, and also for some of the predefined internal actions that can alter an agent's mental state. Such internal actions are essential for some aspects of BDI-based programming, such as checking or dropping current goals or intentions, and therefore need to be formally defined within the operational semantics of the language.