Applying Theory of Mind to multi-agent systems enables agents to model and reason about other agents’ minds. Recent work shows that this ability could increase the performance of agents, making them more efficient than agents that lack this ability. However, modelling others agents’ minds is a difficult task, given that it involves many factors of uncertainty, e.g., the uncertainty of the communication channel, the uncertainty of reading other agents correctly, and the uncertainty of trust in other agents. In this paper, we explore how agents acquire and update Theory of Mind under conditions of uncertainty. To represent uncertain Theory of Mind, we add probability estimation on a formal semantics model for agent communication based on the BDI architecture and agent communication languages.
This paper proposes the application of computational social choice mechanisms to establish cooperative behavior within traffic scenarios involving autonomous vehicles. The main aim is to understand the suitability of commonly used voting rules as a potential mechanism for collective decision making in platoon applications considering unreliable communications. To realistically assess the system performance, we designed an integrated simulation platform composed of an agent-based platform, a microscopic traffic and a vehicular network models. Results show the viability of these simple voting mechanism to maintain high satisfaction among platoon members, which that can lead to stable formations and consequently better traffic conditions. However, additional mechanisms might need to be considered for larger platoon formations to timely guarantee consensus between voters.
Recent studies have shown that applying Theory of Mind to agent technologies enables agents to model and reason about other agents’ minds, making them more efficient than agents that do not have this ability or agents that have a more limited ability of modelling others’ minds. Apart from the interesting results of combining Theory of Mind and agent technologies, an important premise has not been yet fully investigated in the AI literature: how do agents acquire and update their models of others’ minds? In the context of multi-agent systems, one of the most natural ways in which agents can acquire models of other agents’ mental attitudes is through communication. In this work, we propose an operational semantics for agents to update Theory of Mind through communication. We not only make our formalisation broadly applicable by defining a formal semantics based on components from the BDI architecture, but we also implement our approach in an agent-oriented programming language that is based on that architecture.
Autonomous agents negotiating on our behalf find applications in everyday life in many domains such as high frequency trading, cloud computing and the smart grid among others. The agents negotiate with one another to reach the best agreement for the users they represent. An obstacle in the future of automated negotiators is that the agent may not always have a priori information about the preferences of the user it represents. The purpose of this work is to develop an agent that will be able to negotiate given partial information about the user’s preferences. First, we present a new partial information model that is supplied to the agent, which is based on categorical data in the form of pairwise comparisons of outcomes instead of precise utility information. Using this partial information, we develop an estimation model that uses linear optimization and translates the information into utility estimates. We test our methods in a negotiation scenario based on a smart grid cooperative where agents participate in energy trade-offs. The results show that already with very limited information the model becomes accurate quickly and performs well in an actual negotiation setting. Our work provides valuable insight into how uncertainty affects an agent’s negotiation performance, how much information is needed to be able to formulate an accurate user model, and shows a capability of negotiating effectively with minimal user feedback.
Rule-based systems have been increasing in popularity in recent years. They allow for easier handling of both simple and complicated problems utilising a set of rules created in various ways (e.g., manually, or (semi-) automatically, via, say, machine learning or decision trees) depending on the situation. Despite their usefulness however, there are still improvements to be made. Knowledge representation technologies have been available for a long time and provide the means to represent domains formally and correlate entities in those domains. They also allow for ontological reasoning that can take advantage of such connections between entities. These techniques can be useful when applied on rule-based systems in order to improve the quality of rules and, hence, overall system performance. We describe and implement an approach to refine rules used in Internet-of-Things scenarios using knowledge representation and reasoning. The proposed solution uses ontological reasoning on the preconditions and postconditions of rules as it aims to reduce the total amount of rules in a system and simplify them.
Open fleets offer a dynamic environment where the fleet is continually rebuilt ad-hoc since vehicles can enter or leave the fleet anytime, and the only immutable entity is the item to be delivered. Therefore, we need to be able to define a changeable delivery plan capable of adapting to such a dynamic environment. Hence, we propose Open Fleet Management, a Self-Management platform capable of optimizing plan delivery dynamically. The platform utilizes information about location, routes, delivery in transit and delivery costs to change the shipment plan according to the available carrier. Therefore, if two carriers are doing a shipment service to the same place, the platform will be able to discover such a situation and put them in contact to optimize the efficiency of the shipment.
In this paper, we consider the problem of route guidance for shoppers in crowded hypermarkets equipped with smart space technologies. This is an actual and a highly computationally complex problem in peak hours due to dynamically changing congestion conditions, the size and complexity of hypermarkets, and the presence of a multitude of shoppers with different shopping constraints and preferences. High computational complexity of this problem requires a computationally efficient solution approach. We propose a shopper route guidance architecture in which a hypermarket is modelled as a network of communicating smart building agents, each one monitoring its exclusive physical area. Moreover, each shopper is represented by an agent installed on a shopper’s app that, by interacting with other shoppers and smart building agents, dynamically updates its shopping route. Each shopper agent resolves the pick sequencing problem with congestion, i.e., given a shopper’s list, the shopper’s items’ locations are sequenced in the route proposed to a shopper so that the overall traveling time is minimized considering congestion in real-time. We propose a (low computational complexity) greedy tour algorithm and a distributed TSP mathematical model solved in Cplex for this problem and compare their performance. The results show that the proposed architecture and methods scale well and provide efficient shoppers’ routes.
This work proposes the application of consensus processes to ensure the consistency of the data stored in distributed ledgers. Consensus allows a group of agents to reach agreements about the value of common variables or, in this case, data structures such as Merkle trees or chains of blocks. Nevertheless, the consensus algorithm requires for all the participants to apply the same equation. A malicious agent can interfere in the process just by introducing some deviation from the expected value. In this work, the authors propose a method to detect when the information has been modified and, under certain assumptions, it can recover the original data.
We study the properties of scoring allocation correspondences and rules, due to Baumeister et al. [7], that are based on a scoring vector (e.g., Borda or lexicographic scoring) and an aggregation function (e.g., utilitarian or egalitarian social welfare) and can be used to allocate indivisible goods to agents. Extending their previous results considerably and solving some of their open questions, we show that while necessary duplication monotonicity (a notion inspired by the twin paradox [21] and false-name manipulation [1]) fails for most choices of scoring vector when using leximin social welfare, possible duplication monotonicity holds for a very wide range of scoring allocation rules. We also show that a very large family of scoring allocation rules is monotonic. Finally, we show that a large class of scoring allocation correspondences satisfies possible Pareto-optimality, which extends a result of Brams et al. [12].
The increase of population in big cities has produced several problems related to mobility of humans in the city, such as congestions, CO2 emissions, etc. Lately, governments are trying to mitigate this situation by promoting the use of greener means of transportation such as electrical vehicles or bikes. In this paper, we focus on station-based bike sharing systems (BSS). This type of infrastructure (bikes and parking docks) is shared by many users. However, there are some inefficiencies in their management due to imbalanced situations in which some stations fail to provide the service (bike hires or returns) because they are empty or full. We tackle this problem by suggesting users to take (or return) bikes from stations with the goal of keeping the system as balanced as possible. We evaluate our proposal with Bike3S, a bike sharing system simulator developed for testing these types of strategies.
Online discussion has major potential for large-scale consensus building. However, existing SNSs, microblogs, and chat systems lack facilitation functions for avoiding stagnation and flaming of discussion. To develop a function for detecting the stagnation of discussion, we need to quantify the degree of discussion progress, as just the amount of content is not enough to accurately gauge the discussion progress. Our definition of the degree of discussion progress is based on the Issue-Based Information System (IBIS). Specifically, it is defined as a sum of weights representing the importance of IBIS node types extracted from online discussion. In this paper, we determine the optimal weights of the IBIS node types to maximize the correlation coefficient between calculated progress and the subjective progress of human participants. The optimal weights are determined using a genetic algorithm. Experimental results showed that the maximized correlation coefficient was +0.54. Although the current definition of the discussion progress is simple summation, we plan to further refine it with the hierarchical structure of IBIS in future work.
Nowadays the studies on the formalization, enforcement, and monitoring of policies and norms is crucial in different fields of research and in numerous applications. ODRL 2.2 (Open Digital Right Language) is a W3C standard policy expression language formalized using semantic web technologies. It is used to represent permitted and prohibited actions over a certain asset, and obligations required to be met by parties involved in the exchange of a digital asset. In this paper, we propose to extend the model of permission and obligation proposed by ODRL 2.2 in two directions. Firstly, by inserting in the model the notion of activation event or action and by expressing event and action as complex constructs having types and application-independent properties. Secondly, by considering the temporal aspects of obligations and permissions (expiration dates and deadlines) as part of their application independent model. The operational semantics of the proposed model of obligations and permissions is specified using Discrete State Machines and is computed using a production rule system. The proposed approach has been tested by developing a framework in Java able to get as input a set of policies formalized using Semantic Web languages, and to compute their evolution in time based on the events and actions that happen in the interaction among the parties involved in the policies.
Artificial intelligence has been increasing the autonomy of man-made artefacts such as software agents, self-driving vehicles and military drones. This increase in autonomy together with the ubiquity and impact of such artefacts in our daily lives have raised many concerns in society. Initiatives such as transparent and ethical AI aim to allay fears of a “free for all” future where amoral technology (or technology amorally designed) will replace humans with terrible consequences. We discuss the notion of accountable autonomy, and explore this concept within the context of practical reasoning agents. We survey literature from distinct fields such as management, healthcare, policy-making, and others, and differentiate and relate concepts connected to accountability. We present a list of justified requirements for accountable software agents and discuss research questions stemming from these requirements. We also propose a preliminary formalisation of one core aspect of accountability: responsibility.
Task delegation lies at the heart of the service economy, and is a fundamental aspect of many agent marketplaces. Research in computational trust considers which agent a task should be delegated to for execution given the agent’s past behaviour. However, such work does not consider the effects of the agent delegating the task onwards, forming a chain of delegations before the task is finally executed (as occurs in many human outsourcing scenarios). In this paper we consider such delegation chains, and empirically demonstrate that existing trust based approaches do not handle these situations as well. We then introduce a new algorithm based on quitting games to cater for recursive delegation.
The game of Diplomacy has been used as a test case for complex automated negotiations for a long time, but to date very few successful negotiation algorithms have been implemented for this game. We have therefore decided to include a Diplomacy tournament within the annual Automated Negotiating Agents Competition (ANAC). In this paper we present the setup and the results of the ANAC 2017 Diplomacy Competition and the ANAC 2018 Diplomacy Challenge. We observe that none of the negotiation algorithms submitted to these two editions have been able to significantly improve the performance over a non-negotiating baseline agent. We analyze these algorithms and discuss why it is so hard to write successful negotiation algorithms for Diplomacy. Finally, we provide experimental evidence that, despite these results, coalition formation and coordination do form essential elements of the game.
Cognitive agent abstractions can help to engineer intelligent systems across mobile devices. On smartphones, the data obtained from onboard sensors can give valuable insights into the user’s current situation. Unfortunately, today’s cognitive agent frameworks cannot cope well with the challenging characteristics of sensor data. Sensor data is located on a low abstraction level and the individual data elements are not meaningful when observed in isolation. In contrast, cognitive agents operate on high-level percepts and lack the means to effectively detect complex spatio-temporal patterns in sequences of multiple percepts. In this paper, we present a stream-based perception approach that enables the agents to perceive meaningful situations in low-level sensor data streams. We present a crowdshipping case study where autonomous, self-interested agents collaborate to deliver parcels to their destinations. We show how situations derived from smartphone sensor data can trigger and guide auctions, which the agents use to reach agreements. Experiments with real smartphone data demonstrate the benefits of stream-based agent perception.
Data sharing is becoming an integral part of many aspects of our daily lives. We propose a method for controlling access to data and knowledge through fine-grained, user-specified explicitly represented policies. We present an overview of a policy formalism and mechanisms to facilitate distributed data sharing. We provide a breakdown of how our approach defines compliance and violation, specifically providing a new outlook on violation of permissions within the context of data sharing. We also examine how our mechanisms have been adapted to support socially responsible interactions between participants, whilst still providing them with control over their own data. We also explore a series of planned experiments investigating how users understand and interact with policies in a simplified version of our formalism.
In recent work Xiong and Seligman [1,2] introduced a logic for reasoning about preferences and decision making in a setting where the agent focuses on a subset of all alternatives (such as possible workshops to submit her paper to), and where new alternatives can come into focus as the result of questions of the type “have you considered IJCAI workshops?” being asked. An axiomatization of the logic is presented and proved complete. The purpose of the current paper is twofold. First, we present an alternative semantics for the key language constructs proposed in [2], basically differing in representing the alternatives under consideration semantically rather than syntactically. We argue that in some cases this semantics captures the intended meaning better. Second, we present a “direct” sound and complete axiomatization of a preference logic based on this semantics, without using hybrid logic techniques.
In this paper we show how to argue about agreements based on dependence. First, we introduce a formal theory of arguing about agreements by instantiating Dung's abstract theory of argumentation with proposals for agreements represented as dependence networks. Second, we show that acceptable agreements are exchange based--satisfying the so-called do-ut-des principle--and not redundant. Third, to further decrease the number of proposals, we define a notion of minimal proposals. Roughly, all proposals can be split into a number of minimal sub-proposals such that if the proposal is acceptable, then its minimal sub-proposals are acceptable too. We show that minimal proposals satisfy the indecomposable do-ut-des property, i.e., they cannot be split into two nonempty sub-proposals with at most one shared agent.
