Intelligenza Artificiale

Plans for emergency response are complex collaborations in which actors take roles and responsibilities. They are generally long textual documents containing practical instructions, in natural language, for hazard responses. A more rigorous structured-text would be useful for a twofold audience. From one side, it can be useful for quickly understanding the plan and on the other side it can be used to improve the modelling phase and delivering an automatic emergency-support system. This paper proposes an approach, conceived for humans, for converting a free-form plan document into a structured version of the same document. The approach is based on a linguistic and semantic analysis that are strictly correlated and materialize in a metamodel. It contains the essential elements of an emergency plan, and it aids in interpreting the input document also reducing inconsistencies, redundancies, and ambiguities.

A common practice in modern explainable AI is to post-hoc explain black-box machine learning (ML) predictors – such as neural networks – by extracting symbolic knowledge out of them, in the form of either rule lists or decision trees. By acting as a surrogate model, the extracted knowledge aims at revealing the inner working of the black box, thus enabling its inspection, representation, and explanation.

Various knowledge-extraction algorithms have been presented in the literature so far. Unfortunately, running implementations of most of them are currently either proofs of concept or unavailable. In any case, a unified, coherent software framework supporting them all – as well as their interchange, comparison, and exploitation in arbitrary ML workflows – is currently missing.

Accordingly, in this paper we discuss the design of PSyKE, a platform providing general-purpose support to symbolic knowledge extraction from different sorts of black-box predictors via many extraction algorithms. Notably, PSyKE targets symbolic knowledge in logic form, allowing the extraction of first-order logic clauses. The extracted knowledge is thus both machine- and human-interpretable, and can be used as a starting point for further symbolic processing—e.g. automated reasoning.

Nowadays, micro-mobility sharing systems have become extremely popular. Such systems consist in fleets of dockless electric vehicles which are deployed in cities, and used by citizens to move in a more ecological and flexible way. Unfortunately, one of the issues related to such technologies is its intrinsic load imbalance, since users can pick up and drop off the electric vehicles where they prefer.

In this paper we present ESB-DQN, a multi-agent system for E-Scooter Balancing (ESB) based on Deep Reinforcement Learning where agents are implemented as Deep Q-Networks (DQN). ESB-DQN offers suggestions to pick or return e-scooters in order to make the fleet usage and sharing as balanced as possible, still ensuring that the original plans of the user undergo only minor changes.

The main contributions of this paper include a careful analysis of the state of the art, an innovative customer-oriented rebalancing strategy, the integration of state-of-the-art libraries for deep Reinforcement Learning into the existing ODySSEUS simulator of mobility sharing systems, and preliminary but promising experiments that suggest that our approach is worth further exploration.

Autonomy is crucial in cooperation. The complexity of HRI scenarios requires autonomous robots able to exploit their superhuman computations (based on DNN, Machine Learning techniques and Big Data) in a trustworthy way. Trustworthiness is not only a matter of accuracy, privacy or security, but it is becoming more and more a matter of adaptation to humans agency. As claimed by Falcone and Castelfranchi, autonomy means the possibility of dislaying or providing an unexpected behavior (including refusal) that departs from a requested (agreed upon or not) behavior. In this sense, the autonomy to decide how to adopt a task delegated by the user, with respect to her/his own real needs and goals, distinguishes intelligent and trustworthy robots from highly performing robots. This kind of smart help can be provided only by cognitive robots able to represent and ascribe mental states (beliefs, goals, intentions, desires etc.) to their interlocutors. The mental states attribution can be the result of complex reasoning mechanisms or can be fast and automatic, based on scripts, roles, categories or stereotypes typically exploited by humans every time they interact in everyday life. In all these cases, robots that build and use cognitive models of humans (that have a Theory of Mind of their interlocutors), have to operate also a meta-evaluation of their own predictive skills to build those models. Robots have to be endowed with the capability to self-trust their skills to interpret the interlocutors and the context, for producing smart and effective decisions towards humans.

After exploring the main concepts that make collaboration between humans and robots trustworthy and effective, we present the first of a series of experiments draw for testing different aspects of a designed cognitive architecture for trustworthy HRI. This architecture, based on consolidated theoretical principles (theory of social adjustable autonomy, theory of mind, theory of trust) has the main goal to build cognitive robots that provide smart, trustworthy collaboration, every time a human requires their help. In particular, the experiment has been designed in order to demonstrate how the robot’s capability to learn its own level of self-trust on its predictive abilities in perceiving the user and building a model of her/him, allows it to establish a trustworthy collaboration and to maintain a high level of user’s satisfaction, with respect to the robot’s performance, also when these abilities progressively degrade.

The movie industry is a highly differentiated context where production studios compete in non-price product attributes, which influences the box office results of a motion picture. Because of the short life cycle and the constant entrance of new competitive products, temporal decisions play a crucial role. Time series of the number of movies on release and the sum of the box office results of the ten top motion pictures (ranked by box office result for that week) present a counterphased seasonality in the US movie market. We suggest that a possible reason is a risk sensitivity adaptation in the behaviour of the movie’s distributors. This paper provides a model supporting this hypothesis. We developed an agent-based model of a movie market, and we simulated it for 15 years. A comparable global behaviour exists when producers schedule the movies according to given risk-sensitive strategies. This research improves the knowledge of the US motion picture market, analyzing a real-world scenario and providing insight into the behaviour of existing firms in a complex environment.

The interoperability of devices from distinct brands on the Internet of Things (IoT) domain is still an open issue. The main reason is that pioneer companies always deliberately neglected to deploy devices able to interoperate with competitors products. The key factors that may invert such a trend derive, on one hand, from the abstraction of communication protocols that facilitates the migration from vertical to horizontal paradigms and, on the other hand, from the introduction of common and shared ontologies encoding devices specifications. The Semantic Web, with all its layers, can be considered the main framework for delivering ontologies, and by virtue of its features, it is surely the ideal means for providing shared knowledge. In this paper we present a framework that instantiates cognitive agents operating in IoT context, endowed with meta-reasoning in the Semantic Web. The framework, called SW-Caspar, is also provided with a module that performs semi-automatic ontology learning from sentences expressed in natural language; such a learning process generates a conceptual space reflecting the domain of discourse with an instance of a novel foundational ontology called Linguistic Oriented Davidsonian Ontology (LODO), whose main feature is to increase the deepness of reasoning without compromising linguistic-related features. LODO is inspired by the First-Order Logic Davidsonian notation and is serialized in OWL 2. Well-known examples derived from the theory of logical reasoning and a case-study applied to automation on health scenarios are also provided.

Recent years have witnessed the rise of accurate but obscure classification models that hide the logic of their internal decision processes. In this paper, we present a framework to locally explain any type of black-box classifiers working on any data type through a rule-based model. In the literature already exists local explanation approaches able to accomplish this task. However, they suffer from a significant limitation that implies representing data as a binary vectors and constraining the local surrogate model to be trained on synthetic instances that are not representative of the real world. We overcome these deficiencies by using autoencoder-based approaches. The proposed framework first allows to generate synthetic instances in the latent feature space and learn a latent decision tree classifier. After that, it selects and decodes the synthetic instances respecting local decision rules. Independently from the data type under analysis, such synthetic instances belonging to different classes can unveil the reasons for the classification. Also, depending on the data type, they can be exploited to provide the most useful kind of explanation. Experiments show that the proposed framework advances the state-of-the-art towards a comprehensive and widely usable approach that is able to successfully guarantee various properties besides interpretability.

Like other custom-built machinery, elevators are charecterized by a design process which includes selection, sizing and placement of components to fit a given configuration, satisfy users’ requirements and adhere to stringent normative regulations. Unlike mass-produced items, the design process needs to be repeated almost from scratch each time a new configuration is considered. Since elevators are still designed mostly manually, project engineers must engage in time-consuming and error-prone activities over and over again, leaving little to be reused from one design to the next. Computer automated design can provide a cost-effective solution as it relieves the project engineer from such burdens. However, it introduces new challenges both in terms of efficiency — the search space for solutions grows exponentially in the number of component choices — and effectiveness — the perceived quality of the final design may not be as good as in the manual process. In this paper we compare three mainstream AI techniques that can provide problem-solving capabilities inside our tool LiftCreate for automated elevator design, namely Genetic Algorithms (GAs), Constraint Programming (CP) and Satisfiability Modulo Theories (SMT). A special-purpose heuristic search technique embedded in LiftCreate provides us with a yardstick to evaluate the solutions obtained with GAs, CP and SMT and to assess their feasibility for practical applications.