Search

Machine ethics is an uprising sub-field of artificial intelligence fueled by the interest and concerns about the deployment of automated agents in our everyday life. As these agents gain independence from human intervention and make decisions with possible impact on human welfare, real concerns are rising across domains.Due to those reasons, various approaches have been proposed to imbue automated agents with ethical considerations. Several research currents have developed models stemming from psychology and philosophy in an effort to adapt decision-making algorithms to consider ethical values so that the impact of agents on people is bounded and guided by these notions.Most of these approaches consist of either reasoning and applying a set of well-known ethical restrictions, also known as principles (top-down), or inferring them based on carefully crafted datasets through learning algorithms (bottom-up).In this thesis, we look at the problem of implementing these ethical principles in the context of tasks involving sequences of interdependent decisions, i.e: automated planning. We show how certain notions can be modeled using preference-based frameworks, as in top-down approaches, and how these preferences can be inferred from a corpus of data like bottom-up methodologies, to develop a hybrid approach that can be applied to planning problems. An implementation for each facet of our approach is provided in order to test our ideas in practical scenarios., L'éthique des machines est un sous-domaine en plein essor de l'intelligence artificielle qui suscite intérêt et inquiétudes, en particulier en ce qui concerne le déploiement d'agents automatisés dansnotre vie quotidienne. À mesure que ces agents gagnent en indépendance vis-à-vis de l'intervention humaine et prennent des décisions susceptibles d'avoir un impact sur le bien-être humain, de réelles inquiétudes appraissent dans plusieurs domaines.Pour ces raisons, diverses approches ont été proposées pour apporter les agents automatisés de considérations éthiques. Plusieurs courants de recherche ont développé des modèles issus de la psychologie et de la philosophie dans le but d'adapter les algorithmes de prise de décision pour tenir compte des valeurs éthiques afin que l'impact des agents sur les personnes soit délimité et guidé par ces notions.La plupart de ces approches consistent soit à raisonner et à appliquer un ensemble de restrictions éthiques bien connues, également appelées principes (top-down), soit à les inférer sur la base d'ensembles de données soigneusement élaborés grâce à des algorithmes d'apprentissage (bottom-up).Dans cette thèse, nous examinons la mise en œuvre de ces principes éthiques dans le contexte de problèmes impliquant des séquences de décisions, c'est-à-dire : la planification automatique. Nous montrons comment certaines notions peuvent être modélisées à l'aide de cadres formels basés sur les préférences, comme dans les approches `top-down', et comment ces préférences peuvent être déduites d'un corpus de données comme les méthodologies `bottom-up', pour développer une approche hybride applicable à la planification automatique. Un logiciel pour chaque facette de notre approche est fournie afin de tester nos idées sur des scénarios pratiques.

Demand-based, local production will gain relevance in the context of sustainability and circular economy. One way to implement local value creation is through establishing highly dynamic networks that consolidate the competencies of regional manufacturers. Consequently, the structure of the value chains needs to be determined ad hoc dependent on demand. This is a rather challenging task due to the dynamics within such networks and the flat hierarchies. Traditionally, value chains are defined and controlled in a centralized form by a lead firm or a separate stakeholder (e.g. Intermediary, Broker). However, to accommodate the dynamics of demand and the increasing complexity of products, we propose a decentralized form of coordination. The basic idea is to upscale Routing Planning, used in Process Planning, to a network level. Meaning instead of a centralized instance within a company defining the production steps, the stakeholders will collaboratively determine the cross-company Routing Plan, effectively building the value chain. Thus, the accumulated experience and knowledge of all stakeholders can be utilized to efficiently fulfil current customer demand, since the value chain will be executed by the same stakeholders that created it. But in order to coordinate the sequencing of operations by multiple stakeholders, suitable methods need to be implemented. We look at a strategy to facilitate such a collaboration between companies and demonstrate one possible technical implementation based on AI planning using Planning Domain Definition Language (PDDL).

Synthesizing plans for a deformable object to transit from initial observations to goal observations, both of which are represented by high-dimensional data (namely “raw” data), is challenging due to the difficulty of learning abstract state representations of raw data and transition models of continuous states and continuous actions. Even though there have been some approaches making remarkable progress regarding the planning problem, they often neglect actions between observations and are unable to generate action sequences from initial observations to goal observations. In this paper, we propose a novel algorithm framework, namely AGN. We first learn a state-abstractor model to abstract states from raw observations, a state-generator model to generate raw observations from states, a heuristic model to predict actions to be executed in current states, and a transition model to transform current states to next states after executing specific actions. After that, we directly generate plans for a deformable object by performing the four models. We evaluate our approach in continuous domains and show that our approach is effective with comparison to state-of-the-art algorithms.

For social robots to be brought more into widespread use in the fields of companionship, care taking and domestic help, they must be capable of demonstrating social intelligence. In order to be acceptable, they must exhibit socio-communicative skills. Classic approaches to program HRI from observed human-human interactions fails to capture the subtlety of multimodal interactions as well as the key structural differences between robots and humans. The former arises due to a difficulty in quantifying and coding multimodal behaviours, while the latter due to a difference of the degrees of liberty between a robot and a human. However, the notion of reverse engineering from multimodal HRI traces to learn the underlying behavioral blueprint of the robot given multimodal traces seems an option worth exploring. With this spirit, the entire HRI can be seen as a sequence of exchanges of speech acts between the robot and human, each act treated as an action, bearing in mind that the entire sequence is goal-driven. Thus, this entire interaction can be treated as a sequence of actions propelling the interaction from its initial to goal state, also known as a plan in the domain of AI planning. In the same domain, this action sequence that stems from plan execution can be represented as a trace. AI techniques, such as machine learning, can be used to learn behavioral models (also known as symbolic action models in AI), intended to be reusable for AI planning, from the aforementioned multimodal traces. This article reviews recent machine learning techniques for learning planning action models which can be applied to the field of HRI with the intent of rendering robots as socio-communicative., Workshop on Affect, Artifcial Compagnon and Interaction, 2016

A reinforcement learning agent that autonomously explores its environment can utilize a curiosity drive to enable continual learning of skills, in the absence of any external rewards. We formulate curiosity-driven exploration, and eventual skill acquisition, as a selective sampling problem. Each environment setting provides the agent with a stream of instances. An instance is a sensory observation that, when queried, causes an outcome that the agent is trying to predict. After an instance is observed, a query condition, derived herein, tells whether its outcome is statistically known or unknown to the agent, based on the confidence interval of an online linear classifier. Upon encountering the first unknown instance, the agent "queries'' the environment to observe the outcome, which is expected to improve its confidence in the corresponding predictor. If the environment is in a setting where all instances are known, the agent generates a plan of actions to reach a new setting, where an unknown instance is likely to be encountered. The desired setting is a self-generated goal, and the plan of action, essentially a program to solve a problem, is a skill. The success of the plan depends on the quality of the agent's predictors, which are improved as mentioned above. For validation, this method is applied to both a simulated and real Katana robot arm in its "blocks-world'' environment. Results show that the proposed method generates sample-efficient curious exploration behavior, which exhibits developmental stages, continual learning, and skill acquisition, in an intrinsically-motivated playful agent.