Decision-making is crucially important at all levels of biological complexity, from within single-celled organisms, through neural populations within the vertebrate brain, to collections of social organisms such as colonies of ants and honeybees, or societies of humans. What are the prospects for unifying the study of these apparently disparate systems? All can be conceptualised as voting systems at the appropriate level. In this review I will argue that optimality theory can be of fundamental importance in understanding all these systems. In particular I will argue that for groups without conflict of interests, such as neurons and social insect colonies, similar mechanisms could implement statistically optimal decision-making in apparently highly different systems at different levels of biological complexity. I will consider what currency these systems should optimize, and speculate about the possible application of this understanding to the design of voting systems where individual group members' interests are aligned, such as in certain types of human group, and in collectives of robots. I will also consider how established results from economics and political science, notably Arrow's Impossibility Theorem and Condorcet’s ‘jury theorem’, might relate to what we know of social insect voting systems, where interesting effects such as the emergence of collective rationality from the voting of irrational individuals have recently been demonstrated.

A multi-robot system, like a robot formation, contains information that is distributed throughout the system. As the collective increases in numbers or explores distant or difficult areas, obtaining collective situational awareness becomes critical. We propose a method for extracting system and environmental information distributed over a collective of robots.

We introduce CIM, a Commonsense Inference Memory system utilizing both Extended Semantic Networks and Bayesian Networks that builds upon the commonsense knowledgebase ConceptNet. CIM introduces a new technique for self-assembling Bayesian Networks that allows only relevant parts of the commonsense database to affect the inference. The Bayesian Network include the activity in the input sentences and the related activities appearing in the commonsense database. They are used to interpret and infer the meaning of the set of sentences input. Without self-assembled networks, only relevant inference is performed, speeding up performance of reasoning with commonsense knowledge. We demonstrate that our system can disambiguate the needs of the user even if they do not state them directly, and do not use keywords. This ability would not be possible without either the use of commonsense or significant training. Eventually this approach may be applied to increase the effectiveness of other natural language understanding techniques as well.

Trust is an approach to managing the uncertainty about autonomous entities and the information they store, and so can play an important role in any decentralized system. As a result, trust has been widely studied in multiagent systems and related fields such as the semantic web. Here we introduce a simple approach to reasoning about trust with logi

Granularity is the concept of breaking down an event into smaller parts or granules such that each individual granule plays a part in the higher level event. Humans can seamlessly shift their granularity perspectives while reading or understanding a text. To emulate such a mechanism, we describe a theory for inferring this information automatically from raw input text descriptions and some background knowledge to learn the global behavior of event descriptions from local behavior of components. We also elaborate on the importance of discovering granularity structures for solving NLP problems such as — automated question answering and text summarization.

We present an approach to map Robocup-soccer narratives (in natural language) to a sequence of meaningful events. Our approach takes advantage of an action-centered framework, an inference subroutine, and an iterative learning algorithm. Our framework represents the narrative as a sequence of sentences and each sentence as a probability distribution over deterministic events. Our learning algorithm maps sentences to meaningful events without any annotated labeled data. Instead, it uses a prior knowledge about event descriptions and an inference subroutine to estimate initial training labels. The algorithm further improves the training labels at next iterations. In our experiments we demonstrate that with no labeled data our algorithm achieves higher accuracy compared to the state of the art that uses labeled data.

I begin by considering examples of practical reasoning. In the remainder of the paper, I try to say something about what Example 8. Playing soccer. Soccer is like table tennis, but a logical approach that begins to do justice to the subject with the added dimension of teamwork and the need to might be like. This task was selected as a benchmark problem in robotics, and has been extensively Example 1. Ordering a meal at a restaurant. Here, the problem is deciding what to eat and drink. Typing an email message, Even if the only relevant factors are price and preferences composing it as you go along, starts perhaps with a general about food, the number of possible combinations is very idea of what to say.

Starting from a general characterization of roles, we focus on the ways in which roles are specified, we examine the formal constraints on their definitions, and propose definitional schemas motivating different kinds of roles. This classification, in addition to clarify the notion of role itself, helps us to reconsider the two standard solutions that have been proposed for the famous counting problem, and to suggest that a third mixed approach may be considered.

This work introduces Human-Agent Transfer (HAT), an algorithm that combines transfer learning, learning from demonstration and reinforcement learning to achieve rapid learning and high performance in complex domains. Using experiments in a simulated robot soccer domain, we show that human demonstrations transferred into a baseline policy for an agent and refined using reinforcement learning significantly improve both learning time and policy performance. Our evaluation compares three algorithmic approaches to incorporating demonstration rule summaries into transfer learning, and studies the impact of demonstration quality and quantity. Our results show that all three transfer methods lead to statistically significant improvement in performance over learning without demonstration.

As computational agents are increasingly used beyond research labs, their success will depend on their ability to learn new skills and adapt to their dynamic, complex environments. If human users — without programming skills — can transfer their task knowledge to the agents, learning rates can increase dramatically, reducing costly trials. The TAMER framework guides the design of agents whose behavior can be shaped through signals of approval and disapproval, a natural form of human feedback. Whereas early work on TAMER assumed that the agent's only feedback was from the human teacher, this paper considers the scenario of an agent within a Markov decision process (MDP), receiving and simultaneously learning from both MDP reward and human reinforcement signals. Preserving MDP reward as the determinant of optimal behavior, we test two methods of combining human reinforcement and MDP reward and analyze their respective performances. Both methods create a predictive model, H-hat, of human reinforcement and use that model in different ways to augment a reinforcement learning (RL) algorithm. We additionally introduce a technique for appropriately determining the magnitude of the model's influence on the RL algorithm throughout time and the state space.