Autoformalization has emerged as a term referring to the automation of formalization in the context of the formalization of mathematics using interactive theorem provers (proof assistants). Its rapid development has been driven by progress in deep learning, especially large language models (LLMs). More recently, usage of the term has expanded beyond mathematics to describe tasks that involve translating natural language input into verifiable logical representations. At the same time, a growing body of research explores using LLMs to translate informal language into formal representations for reasoning, planning, and knowledge representation, but without explicitly referring to this process as autoformalization. As a result, despite addressing similar tasks, the largely independent development of these research areas has limited opportunities for shared methodologies, benchmarks, and theoretical frameworks that could accelerate progress. Our goal is to review - explicit or implicit - instances of what can be considered autoformalization and to propose a unified framework, encouraging cross-pollination between different fields to advance the development of next generation AI systems.

Multi-agent simulations are a versatile tool for exploring interactions among natural and artificial agents, but their development typically demands domain expertise and manual effort. This work introduces the Generative Agents for Multi-Agent Autoformalization (GAMA) framework, which automates the formalization of interaction scenarios in simulations using agents augmented with large language models (LLMs). To demonstrate the application of GAMA, we use natural language descriptions of game-theoretic scenarios representing social interactions, and we autoformalize them into executable logic programs defining game rules, with syntactic correctness enforced through a solver-based validation. To ensure runtime validity, an iterative, tournament-based procedure tests the generated rules and strategies, followed by exact semantic validation when ground truth outcomes are available. In experiments with 110 natural language descriptions across five 2 × 2 simultaneous-move games, GAMA achieves 100% syntactic and 76.5% semantic correctness with Claude 3.5 Sonnet, and 99.82% syntactic and 77% semantic correctness with GPT-4o. The framework also shows high semantic accuracy in autoformalizing agents’ strategies.

The rapid rise of large language models (LLMs) has shifted artificial intelligence (AI) research toward agentic systems, motivating the use of weaker and more flexible notions of agency. However, this shift raises key questions about the extent to which LLM-based agents replicate human strategic reasoning, particularly in game-theoretic settings. In this context, we examine the role of agentic sophistication in shaping artificial reasoners’ performance by evaluating three agent designs: a simple game-theoretic model, an unstructured LLM-as-agent model, and an LLM integrated into a traditional agentic framework. Using guessing games as a testbed, we benchmarked these agents against human participants across general reasoning patterns and individual role-based objectives. Furthermore, we introduced obfuscated game scenarios to assess agents’ ability to generalise beyond training distributions. Our analysis, covering over 2000 reasoning samples across 25 agent configurations, shows that human-inspired cognitive structures can enhance LLM agents’ alignment with human strategic behaviour. Still, the relationship between agentic design complexity and human-likeness is non-linear, highlighting a critical dependence on underlying LLM capabilities and suggesting limits to simple architectural augmentation.

In the legal field, accurately interpreting and applying legal definitions is crucial yet challenging due to inherent ambiguities. This paper introduces DeCoRA, a novel framework that enhances legal argumentation by incorporating context-based reasoning to address these ambiguities, with a focus on the judge as the central decision maker. Unlike black-box models, such as generative models or outcome prediction systems, which often produce outputs without fully explaining the reasoning behind their conclusions, DeCoRA emphasizes transparency by modeling the judicial decision-making process in a structured and interpretable manner. Our key contributions include: (1) a tree-based knowledge base that organizes legal definitions, highlighting their relationships and effects; (2) a context-aware definition framework enabling judges to interpret definitions considering legal and contextual relevance; and (3) an effective method for handling complex legal scenarios with conflicting or overlapping definitions.

Large language models (LLMs) are increasingly explored as general-purpose reasoners, particularly in agentic contexts. However, their outputs remain prone to mathematical and logical errors. This is especially challenging in open-ended tasks, where unstructured outputs lack explicit ground truth and may contain subtle inconsistencies. To address this issue, we propose Logic-Enhanced Language Model Agents (LELMA), a framework that integrates LLMs with formal logic to enable validation and refinement of natural language reasoning. LELMA comprises three components: an LLM-Reasoner, an LLM-Translator, and a Solver, and employs autoformalization to translate reasoning into logic representations, which are then used to assess logical validity. Using game-theoretic scenarios such as the Prisoner's Dilemma as testbeds, we highlight the limitations of both less capable (Gemini 1.0 Pro) and advanced (GPT4o) models in generating logically sound reasoning. LELMA achieves high accuracy in error detection and improves reasoning correctness via self-refinement, particularly in GPT-4o. The study also highlights challenges in autoformalization accuracy and in evaluation of inherently ambiguous open-ended reasoning tasks.

Negotiating in uncertain environments, where user preferences are only partially known, poses a challenge for traditional negotiation models that rely on rigid, pre-defined strategies. These models struggle to adapt to changing conditions or transfer knowledge across different negotiation contexts, making them ineffective in dynamic environments. To address this research gap, we propose a novel negotiation model that uses deep reinforcement learning (DRL) to enable agents learn adaptable, generalizable strategies through the notion of “strategy templates”. These templates include (a) choice parameters to select tactics, (b) time parameters to control when tactics are activated, and (c) attribute-value parameters to guide acceptance and inform bidding decisions. As a result, we enable negotiation agents dynamically adapt their strategies, through pre-training on teacher strategies and refining them via online learning in diverse environments. Our agents also derive a user model to approximate partially specified user preferences, thus handling preference uncertainty more effectively. We developed a proof-of-concept prototype using an actor-critic architecture based on DRL, supplemented by stochastic search techniques for the estimation of user model and multi-objective optimization for making mutually beneficial offers. Experimental evaluations show that our model outperforms state-of-the-art approaches in terms of both individual and social-welfare utilities, demonstrating its ability to transfer experience across domains and excel in previously unseen scenarios. This work provides a robust framework for dynamic, adaptable strategy formation, bridging the gap in current negotiation models by addressing uncertainty in user preferences and strategy flexibility.

Game-theoretic simulations are a powerful tool for exploring strategic interactions and guiding decision-making in fields ranging from business to economics and politics. However, simulations of such complex systems are often difficult to understand intuitively and debug. In this context, particularly challenging and hard to detect are logical (intentional) errors resulting from faulty logic in an agent's strategy. To address this challenge, we develop a framework for question-based explanations, enhancing the understanding of agents' behaviour. We focus on teleo-reactive agents in game-theoretical simulations, specifically in tournament and evolutionary contexts. Our approach centres on trace-based explanations, utilising behavioural logs to identify the steps leading to particular outcomes. We formally describe explanation templates for “why” and “why not” question types, linking them to agents' goals, beliefs, and condition-action rules. Furthermore, we provide formal definitions of the answers to these questions, linking them to output templates. The methodology is demonstrated through example dialogue scenarios, showing how these explanations can improve debugging efficiency by offering high-level insights into agents' behaviours.

An agent's ability to distinguish between sensory effects that are self-caused, and those that are not, is instrumental in the achievement of its goals. This ability is thought to be central to a variety of functions in biological organisms, from perceptual stabilisation and accurate motor control, to higher level cognitive functions such as planning, mirroring and the sense of agency. Although many of these functions are well studied in AI, this important distinction is rarely made explicit and the focus tends to be on the associational relationship between action and sensory effect or success. Toward the development of more general agents, we develop a framework that enables agents to disentangle self-caused and externally-caused sensory effects. Informed by relevant models and experiments in robotics, and in the biological and cognitive sciences, we demonstrate the general applicability of this framework through an extensive experimental evaluation over three different environments.

We propose a generic knowledge representation framework that supports evolutionary game-theoretic simulations using cognitive agents. The framework allows an experimenter to test a population of such agents via generations to study how specific population behaviours evolve over time. A generation is composed of rounds which can be further divided into encounters according to model-specific requirements. As agents in the population interact, events (caused either by agent actions or by separate environment processes) take place. These events change the environment, changes are then perceived by agents that, in turn, decide to take new actions that affect the environment. This process continues until it is time to evolve a new generation, when strategies of the fittest players are selected for the next generation to start evolving. This evolutionary loop continues until all the terminating conditions of the simulation are met. We use the framework to show how to successfully repeat existing experiments from evolutionary simulations of agent cooperation. Our results validate our framework and pave the way for EVOCOGNISIM, a simulation platform that implements the key aspects of the framework in a systematic manner.

We present LawGiBa, a proof-of-concept demonstration system for legal assistance that combines GPT, legal knowledge bases, and Prolog’s logic programming structure to provide explanations for legal queries. This novel combination effectively and feasibly addresses the hallucination issue of large language models (LLMs) in critical domains, such as law. Through this system, we demonstrate how incorporating a legal knowledge base and logical reasoning can enhance the accuracy and reliability of legal advice provided by AI models like GPT. Though our work is primarily a demonstration, it provides a framework to explore how knowledge bases and logic programming structures can be further integrated with generative AI systems, to achieve improved results across various natural languages and legal systems.

We present World of Bugs (WOB), an open platform that aims to support Automated Bug Detection (ABD) research in video games. We discuss some open problems in ABD and how they relate to the platform’s design, arguing that learning-based solutions are required if further progress is to be made. The platform’s key feature is a growing collection of common video game bugs that may be used for training and evaluating ABD approaches.

We study the problem of how an agent that negotiates over multiple issues with an opponent can make offers given that it has incomplete information about the user it represents and the opponent it plays against. To tackle this problem, we take a multi-objective optimization stance, where the negotiating agent estimates the preferences of both user and opponent to generate bids that are (near) Pareto-optimal. However, since the negotiating agent needs to approximate the actual preferences of two parties, uncertainty is involved. To handle this uncertainty, we propose a fuzzy approach consisting of a two-phase Pareto-bid generation step where Phase-I generates the non-dominated solutions using a fuzzy multi-objective evolutionary algorithm, and Phase II ranks them to find the best bid to offer the opponent using a fuzzy multiple-criteria decision-making method. Rigorous experimentation shows that the hybrid fuzzy approach of generating the (near) Pareto-optimal bids reduces the average distance to the Pareto curve and increases the average joint or social welfare utility of the agents leading to “win-win” situations.

We propose a novel knowledge representation framework called COGNISIM that supports game theoretic simulation experiments using cognitive agents. The framework allows an experimenter to evolve a population of such agents with strategies expressed teleo-reactively as logic programs. When agents encounter each other, events take place in the environment, caused either by agent actions or by environment processes. Such events change the environment’s internal state, and these changes are then observed by agents that, in turn, decide to take new actions that affect the environment. This loop continues until the terminating conditions of the simulation are met. Using this framework, we show how to repeat experiments from the literature based on Axelrod’s tournament. We also evaluate our platform’s performance in efficiently supporting large simulations in game theoretic settings.

Abstract We present a novel negotiation model that allows an agent to learn how to negotiate during concurrent bilateral negotiations in unknown and dynamic e-markets. The agent uses an actor-critic architecture with model-free reinforcement learning to learn a strategy expressed as a deep neural network. We pre-train the strategy by supervision from synthetic market data, thereby decreasing the exploration time required for learning during negotiation. As a result, we can build automated agents for concurrent negotiations that can adapt to different e-market settings without the need to be pre-programmed. Our experimental evaluation shows that our deep reinforcement learning based agents outperform two existing well-known negotiation strategies in one-to-many concurrent bilateral negotiations for a range of e-market settings.

With the aim of designing automated tools that assist in the video game quality assurance process, we frame the problem of identifying bugs in video games as an anomaly detection (AD) problem. We develop State-State Siamese Networks (S3N) as an efficient deep metric learning approach to AD in this context and explore how it may be used as part of an automated testing tool. Finally, we show by empirical evaluation on a series of Atari games, that S3N is able to learn a meaningful embedding, and consequently is able to identify various common types of video game bugs.

We present a novel negotiation model that allows an agent to learn how to negotiate during concurrent bilateral negotiations in unknown and dynamic e-markets. The agent uses an actor-critic architecture with model-free reinforcement learning to learn a strategy expressed as a deep neural network. We pre-train the strategy by supervision from synthetic market data, thereby decreasing the exploration time required for learning during negotiation. As a result, we can build automated agents for concurrent negotiations that can adapt to different e-market settings without the need to be pre-programmed. Our experimental evaluation shows that our deep reinforcement learning based agents outperform two existing well-known negotiation strategies in one-to-many concurrent bilateral negotiations for a range of e-market settings.

MagnetDroid is a novel artificial intelligence framework that integrates a security ontology, a multi-agent organisation, and a logical reasoning procedure to help build a bridge between the worlds of Android application analysis and law, with respect to privacy. Our contribution helps identify violations of the law by Android applications, as well as predict legal consequences. The resulting implementation of MagnetDroid can be useful to privacy-concerned users in order to acknowledge problems with the privacy of the applications they use, to application developers/publishers to help them identify which problems to fix, and to lawyers in order to provide an additional level of interpretation for any court when considering the privacy of Android applications.

We study a novel location-aware strategy for distributed systems where cooperating agents perform the load-balancing. The strategy allows agents to identify opportunities within a current unbalanced allocation, which in turn triggers concurrent and one-to-many negotiations amongst agents to locally reallocate some tasks. The tasks are reallocated according to the proximity of the resources and they are performed in accordance with the capabilities of the nodes in which agents are situated. This dynamic and on-going negotiation process takes place concurrently with the task execution and so the task allocation process is adaptive to disruptions (task consumption, slowing down nodes). We evaluate the strategy in a multi-agent deployment of the MapReduce design pattern for processing large datasets. Empirical results demonstrate that our strategy significantly improves the overall runtime of the data processing.

We develop a novel strategy that supports software agents to make decisions on how to negotiate for a resource in open and dynamic e-markets. Although existing negotiation strategies offer a number of sophisticated features, including modelling an opponent and negotiating with many opponents simultaneously, they abstract away from the dynamicity of the market and the model that the agent holds for itself in terms of ongoing negotiations, thus ignoring information that increases an agent’s utility. Our proposed strategy COncurrent Negotiating AgeNts ( Conan ) considers a weighted combination of modelling the market environment and the progress of concurrent negotiations in which the agent partakes. We conduct extensive experiments to evaluate the strategy’s performance in various settings where different opponents from the literature provide a competitive market. Our experiments provide statistically significant results showing how Conan outperforms the state-of-the-art in terms of the utility gained during negotiations.

Abstract We present a knowledge representation framework on the basis of the Event Calculus that allows an agent to recognize complex activities from low‐level observations received by multiple sensors, reason about the life cycle of such activities, and take action to support their successful completion. Activities are multivalue fluents that change according to events that occur in the environment. The parameters of an activity consist of a unique label, a set of participants involved in the performing of the activity, and a unique goal associated with the activity revealing the activity's desired outcome. Our contribution is the identification of an activity life cycle describing how activities can be started, interrupted, suspended, resumed, or completed over time, as well as how these can be represented. The framework also specifies activity goals, their associated life cycle, and their relation with the activity life cycle. We provide the complete implementation of the framework, which includes an activity generator that automatically creates synthetic sensor data in the form of event streams that represent the everyday lifestyle of a type 1 diabetic patient. Moreover, we test the framework by generating very large activity streams that we use to evaluate the performance of the recognition capability and study its relative merits.

We present Hydra: a multi-agent hybrid diagnosis and monitoring architecture that is aimed at helping diabetic patients manage their illness. It makes use of model-based diagnosis techniques, where the model can be developed by two different approaches combined in a novel way. In the first approach, we build the model based on the medical guidelines provided for diabetes. A computational logic agent monitors the patient and provides feedback based on the model whenever the current observations regarding the patient are sufficient to draw a conclusion. In the second approach, we assume a function for the model, and learn its parameters through data. The model is consistently updated via incoming observations about the patients, and allows prediction of possible future values. We describe the components of such an architecture, and how it can integrated into the existing COMMODITY12personal health system. We implement a prototype of Hydra, and present its workings on a case study on hypoglycemia monitoring. We report our prediction results for this scenario.

Today's self-managing systems would ideally be able to adapt themselves (their internal structure or behavior), as well as to autonomously participate in larger, self-organizing systems. Analogously, the enterprises or other socio-economic systems autonomously manage themselves - they make decisions on how to adapt their structure and behavior, and how to organize with other entities in the environment. To connect internal self-adaptive with external self-organizational behavior, an enterprise is “aware” of itself and of its environment, and acts according to this awareness. This position paper proposes to address the challenges of a complex distributed self-managing system by making entities in such a system able to adapt themselves similarly to how companies manage themselves in socio-economic systems. To enable the knowledge transfer between these two fields, the paper proposes to utilize symbolic models which will be used by self-managing systems for knowledge representation and reasoning. This will make such systems in a way also self-aware and enable both self-adaptive and self-organizing capabilities. The paper discusses research directions to make this approach possible.

Very commonly, multi-agent systems built for ubiquitous computing and ambient intelligence applications require from their members to perform collaborative tasks or to attempt to communicate regarding potentially unknown objects in their environment. The specific class of this type of systems that constitutes the domain for the proposed research defines multi-agent systems utilizing agents that encompass high-level symbolic name worlds that are linked to their environment via low level sensory input. A successful outcome when attempting collaborative tasks in such systems is always dependent on the ability of the agents to refer to the correct objects in their communication. In my research I will pursue an approach towards confronting the above problem and propose a solution. This solution will be applied in this category of multi-agent systems enabling them to deal with the imperative to verify that all agents refer to the correct object in order for the outcome of a collaborative task upon it or the communication regarding it to be successful.

This paper presents the computational logic foundations of a model of agency called the KGP (Knowledge, Goals and Plan model. This model allows the specification of heterogeneous agents that can interact with each other, and can exhibit both proactive and reactive behaviour allowing them to function in dynamic environments by adjusting their goals and plans when changes happen in such environments. KGP provides a highly modular agent architecture that integrates a collection of reasoning and physical capabilities, synthesised within transitions that update the agent's state in response to reasoning, sensing and acting. Transitions are orchestrated by cycle theories that specify the order in which transitions are executed while taking into account the dynamic context and agent preferences, as well as selection operators for providing inputs to transitions.

This is the second part of a pair of special issues dedicated to the area of Intelligence and Interaction in Community-based Systems. The first of these special issues(Stathis and Purcell, 2002) addressed the range of spontaneous interactions that may form part of a set of concentric patterns of personal, familial, social and civic circles. The focus then was on the needs of physically co-located communities, those lay people who may live next door, frequent the local pub, send their children to the same school, or bump into each other intermittently at meetings of local societies or professional associations. This second part of the special issues still keeps the notion of local community in the physical neighbourhood but here the attention is diverted to the kind of relationships that the members of such a community may be establishing with other communities, whether local or global, real or virtual, more simple or more complex. Such relationships can vary enormously in the manner, in which, interactions can be structured and organised. It is this potential development that orients this part of the special issue towards those more social activities that are, by their nature, quite structured and which involve citizen interactions in more formal settings, such as the buying and selling of goods, learning in a local setting, sharing knowledge at work, and accessing local cultural resources in order to support the needs of a modern community's lifestyle.

We investigate the application of multi-agent systems to develop intelligent information interfaces for connected communities, a class of computer applications aimed at enhancing the way people interact and socialise in geographically co-located communities such as neighbourhoods. In this context, we study the problem of providing effective information management in support of social interaction when a diverse range of computing devices is employed. The novelty of our approach is based on combining innovative interactive devices with a framework based on agent roles in order to support the effective flow of community-related content for the people of a given locality. In particular, we have integrated existing techniques for information retrieval and filtering with measures of content popularity, to ensure that documents in the community system are optimally available. After reporting on the potential presence of the system in the community, we report on the development of a framework for multi-agent systems in which agents provide a number of services aimed at facilitating personalised and location-dependent information access to members of the community. We also present a summary of the results of an expert evaluation of the information flow resulting from the communication between agents, and a user-evaluation of the information dissemination facilities provided by the system. © 2002 Elsevier Science B.V. All rights reserved.

In a world of virtual and real communities, this special issue of ‘Interacting with Computers’ focuses on the real. The issue addresses the communication needs of physical co-located communities, those lay people who may live next-door, frequent the local pub, send their children to the same school, or bump into each other intermittently at meetings of local societies or professional associations. This is then the world of informal and spontaneous interactions located in a set of concentric patterns of personal, familial, social and civic circles. The context represents a major challenge for the profession of human–computer interaction to construct information technologies that support the interactions of ordinary people in these social settings. With pervasive internetworking, computers have become an extremely effective and economic means by which people communicate. A new generation of applications further challenges computer software to serve as the intermediary between people, to both facilitate and sustain their social relationships. Indeed, a group of such applications under the umbrella of social computing (Schuler, 1994), focuses on the relationships between people when workplace tasks are defined via software in organisations, when people learn and teach with computers, when governments devise and implement policies over networks and when people interact socially in the context of modern community lifestyles.

We present an abstract framework for designing and developing globalised interactive systems from simple components viewed as games [1] (K. Stathis, M.J. Sergot, Games as a Metaphor for Interactive Systems, in: M.A. Sasse, R.J. Cunningham, R.L. Winder (Eds.), People and Computers XI (Proceedings of HCI′96), August 1996, London, UK, BCS Conference Series, Springer-Varlag, pp. 19–33). We identify a set of concepts required to specify and implement interactions, in such a way whereby instantiating the specifications and implementations of games we obtain components that correspond to localised instances of an interactive system. Localisation is also obtained by either customising the specification or the implementation of the interactive system, or both. The framework also caters for complex interactive systems which are interpreted as compound games built-up from sub-games. In this case, coordination of sub-games is the main issue that we must address at the global level. This is resolved by specifying and implementing sub-games as active components of the more complex games, and, as a result, we localise the coordination of components in the interactive system. The framework lends itself towards a methodology that is suitable for globalising the development of interactive systems.