The recent field of Machine Ethics is experiencing rapid growth to answer the societal need for Artificial Intelligence (AI) algorithms imbued with ethical considerations, such as benevolence toward human users and actors. Several approaches already exist for this purpose, mostly either by reasoning over a set of predefined ethical principles (Top-Down), or by learning new principles (Bottom-Up). While both methods have their own advantages and drawbacks, only few works have explored hybrid approaches, such as using symbolic rules to guide the learning process for instance, combining the advantages of each. This paper draws upon existing works to propose a novel hybrid method using symbolic judging agents to evaluate the ethics of learning agents' behaviors, and accordingly improve their ability to ethically behave in dynamic multi-agent environments. Multiple benefits ensue from this separation between judging and learning agents: agents can evolve (or be updated by human designers) separately, benefiting from co-construction processes; judging agents can act as accessible proxies for non-expert human stakeholders or regulators; and finally, multiple points of view (one per judging agent) can be adopted to judge the behavior of the same agent, which produces a richer feedback. Our proposed approach is applied to an energy distribution problem, in the context of a Smart Grid simulator, with continuous and multi-dimensional states and actions. The experiments and results show the ability of learning agents to correctly adapt their behaviors to comply with the judging agents' rules, including when rules evolve over time.

To be accepted in our everyday life and to be valuable interaction partners, robots should be able to display emotional and empathic behaviors. That is why there has been a great focus on developing empathy in robots in recent years. However, there is no consensus on how to measure how much a robot is considered to be empathic. In this context, we decided to construct a questionnaire which specifically measures the perception of a robot's empathy in human-robot interaction (HRI). Therefore we conducted pretests to generate items. These were validated by experts and will be further validated in an experimental setting.

Designing a public urban policy is a demanding process which requires both time and money with no warranty of its efficiency. It involves knowledge about the purpose of urban design, behaviors of users and needs in terms of mobility. We believe that in the near future, decision makers will have to react and more frequently adapt public policies, based on the huge amount of available data, feedbacks from both target users and stakeholders. In this paper, we propose a generic agent-based architecture to model and simulate urban policies, which could facilitate the co-design and assessment of public policies in a specific environment. Two agent-based models are coupled with a micromacro dynamic loop, and they can be adapted either by the system using reinforcement learning, or by the stakeholders using simulation results. A generic formalism is elaborated to represent urban policies, which can be instantiated in a co-design approach between the policymaker and our system. An experimentation is conducted on an urban mobility policy, related to the configuration of parking price system in downtown area. The agent’s behavior and environment are developed to be as realistic as possible, based on a real-world source of modeling. However, our architecture has been designed to be generic, exploiting infrastructure data from any city using available community data (Open Street Map). The scenario of parking pricing shows that the system learns postpolicy behaviors and can propose some adjustments (e.g., specific actions to apply, when and how) to better meet the stakeholders’ objectives (e.g., maximize parking gains). The policy maker can then choose to validate the provided policies, or modified them for additional simulations.

The relevance of decision making in autonomous systems is intrinsically related to the system capacity to discriminate its perception-action states. This is particularly challenging in unknown and changing complex environments, where providing a complete a priori representation to the system is not possible. To illustrate the problem, let us consider a decentralized control of road traffic, where a control device of the distributed infrastructure locally controls traffic, by learning to construct a precise representation (perception-action states) of the traffic state. In this context, it is challenging to define from prior knowledge a relevant representation of the traffic state that enables an efficient recommendation-based control. Without considering a prior domain-knowledge representation, we propose an approach able to combine a set of existing traditional unsupervised learning methods that collaborate as a population of agents in order to build an efficient representation. Our approach follows a constructivist learning perspective, where each agent produces a possible discretization of the raw sensed data. Thanks to a multi-agent reinforcement learning process, the population is able to collectively build a representation that combines the good capacities of the individual ones.

Cooperative Intelligent Transportation Systems (C-ITS) are complex systems well-suited to a multi-agent modeling. We propose a multi-agent based modeling of a C-ITS, that couples 3 dynamics (physical, informational and control dynamics) in order to ensure a smooth cooperation between non cooperative and cooperative vehicles, that communicate with each other (V2V communication) and the infrastructure (I2V and V2I communication). We present our multi-agent model, tested through simulations using real traffic data and integrated into our extension of the Multi-model Open-source Vehicular-traffic SIMulator (MovSim).

Human is surrounded by a tremendous amount of information on the web. That highlights the continuous need of recommendation systems in the different domains. Unfortunately cold start problem is still an important issue in these systems on new users and new items. The problem becomes more critical in systems that contain resources that lives too shortly like offers on products which stays only for few days (short life resources - SLiR), or news in a news site. From the other side social networks are very rich with users' information, unfortunately most of the proposed social recommender are applied on domain specific social networks like flickers and epinions which are much less used in the day to day life, because dealing with General Purpose Social Network (GPSN) like Facebook and Twitter needs to transform these GPSN into a useful source of recommendation dealing with them as row, implicit or unary data. In this work we highlight how iSoNTRE (the intelligent Social Network Transformer into Recommendation Engine) addresses this challenge by transforming the GPSN into useful information for recommendation based on middle layer of domain concepts. iSoNTRE overcomes the cold start problem on new users and items. It has been evaluated over Twitter, on new users, recommending offers as a kind of SLiR, results showed that iSoNTRE succeeded in recommending good offers with 14% of click on recommended offers, which is high compared to general open rate in social media, especially when we have nothing about users and we are recommending SLiR resources.

International audience

MCTS (Monte Carlo Tree Search) is a well-known and efficient process to cover and evaluate a large range of states for combinatorial problems. We choose to study MCTS for the Computer Go problem, which is one of the most challenging problem in the field in Artificial Intelligence. For this game, a single combinatorial approach does not always lead to a reliable evaluation of the game states. In order to enhance MCTS ability to tackle such problems, one can benefit from game specific knowledge in order to increase the accuracy of the game state evaluation. Such a knowledge is not easy to acquire. It is the result of a constructivist learning mechanism based on the experience of the player. That is why we explore the idea to endow the MCTS with a process inspired by constructivist learning, to self-acquire knowledge from playing experience. In this paper, we propose a complementary process for MCTS called BHRF (Background History Reply Forest), which allows to memorize efficient patterns in order to promote their use through the MCTS process. Our experimental results lead to promising results and underline how self-acquired data can be useful for MCTS based algorithms.

Human is surrounded by a tremendous and scary amount of information on the web. That highlights the continuous need of recommendation systems in the different domains. Unfortunately cold start problem is still an important issue in these systems on new users and new items. The problem becomes more critical in systems that contain resources that lives too shortly like offers on products which stays only for few days (short life resources-SLiR). In this work we highlight how iSoNTRE (the intelligent Social Network Transformer into Recommendation Engine) solves this problem by using users' information in online social networks to overcome the cold start problem on new users, as well as iSoNTRE uses conceptual similarity, this overcomes the problem on new items, and on short life resources also. The work has been evaluated on Twitter on real users and results show that iSoNTRE succeeded in recommending offers to users with 14% of open rate on recommended offers, which is high compared to general open rate in social media, especially when we have nothing about users or offers before.

With the proliferation of Web services, it is becoming increasingly important to support the users in selecting the most appropriate compositions of services for a task. We propose a new service discovery and selection framework that utilises the concept of task memories and a social network of task memories. A task memory captures the service composition history and their meta-data such as associated context and user rating. A network of task memories is formed to realise an effective task memory sharing platform among the users.

The Web continues to grow at a phenomenal rate and the amount of information on the Web is overwhelming. Finding the relevant information remains a big challenge. Due to its wide distribution, its openness and high dynamics, the WWW is a complex system, for which we have to imagine mechanisms of content maintaining, filtering and organizing that are able to deal with its evolving, dynamics and distribution. Integrating mechanisms of self-organization of the Web content is an attractive perspective, to match with these requirements. This paper explores the Web from a complex adaptive system (CAS) perspective. It reviews some characteristic behaviours of CAS and shows how the Web exhibits similar behaviours. We illustrate with a model for Web content organization adopting the CAS vision and using the multi-agent paradigm

The World Wide Web contains a huge amount of unstructured, distributed, multi-media data. This content provides a great potential source for knowledge acquisition that needs to be filtered, organized, and maintained in order to permit an efficient use. The wide distribution of the Web, its openness and high dynamics make any tentative of content organization or maintaining, a task very hard to achieve. The WWW is thus, a complex system, for which we have to imagine mechanisms of content maintaining, filtering and organizing, that are able to deal with its content evolving, dynamics and distribution. Integrating mechanisms of self-organization of the Web content is an attractive perspective, to match with these requirements. We present in this paper a new approach, inspired by social insects, to organize dynamically the Web content. Our approach combines foraging behavior and collective sorting behavior, and uses the stigmergy mechanism. We present in this paper some experiments and results that show its effectiveness.