Hypothetical “value learning” AIs learn human values and then try to act according to those values. The design of such AIs, however, is hampered by the fact that there exists no satisfactory definition of what exactly human values are. After arguing that the standard concept of preference is insufficient as a definition, I draw on reinforcement learning theory, emotion research, and moral psychology to offer an alternative definition. In this definition, human values are conceptualized as mental representations that encode the brain’s value function (in the reinforcement learning sense) by being imbued with a context-sensitive affective gloss. I finish with a discussion of the implications that this hypothesis has on the design of value learners.

Hashtags are useful in many applications, such as tweet classification, clustering, searching, indexing and social network analysis. This study seeks to recommend relevant Twitter hashtags for health-related keywords based on distributed language representations, generated by the state-of-the-art Deep Learning technology. The word embeddings are built from billions of tweet words without supervision. To the best of our knowledge, this is the first study of applying distributed language representations to recommending hashtags for keywords. The experiment showed that this approach outperformed the baseline approach that is based on keyword and hashtag co-occurrence in tweets.

The capability of collaboration is critical in the design of symbiotic cognitive systems. To obtain this functional capability, a cognitive system should possess evaluative and communicative processes. Emotions and their underlying processes provide such functions in social and collaborative environments. We investigate the mutual influence of affective and collaboration processes in a cognitive theory to support the interaction between humans and robots or virtual agents. We have developed new algorithms for these processes, as well as a new overall computational model for implementing collaborative robots and agents. We build primarily on the cognitive appraisal theory of emotions and the SharedPlans theory of collaboration to investigate the structure, fundamental processes and functions of emotions in a collaboration context.

To enable effective collaborations between humans and cognitive robots, it is important for robots to continuously acquire task knowledge from human partners. To address this issue, we are currently developing a framework that supports task learning through visual demonstration and natural language dialogue. One core component of this framework is the integration of language and vision that is driven by dialogue for task knowledge learning. This paper describes our on-going effort, particularly, grounded task learning through joint processing of video and dialogue using And-Or-Graphs (AOG).

Home environments constitute a main target location where to deploy robots, which are expected to help humans in completing their tasks. However, modern robots do not meet yet user's expectations in terms of both knowledge and skills. In this scenario, users can provide robots with knowledge and help them in performing tasks, through a continuous human-robot interaction. This human-robot cooperation setting in shared environments is known as Symbiotic Autonomy or Symbiotic Robotics. In this paper, we address the problem of an effective coexistence of robots and humans, by analyzing the proposed approaches in literature and by presenting our perspective on the topic. In particular, our focus is on specific contexts that can be embraced within Symbiotic Autonomy: Human Augmented Semantic Mapping, Task Teaching and Social Robotics. Finally, we sketch our view on the problem of knowledge acquisition in robotic platforms by introducing three essential aspects that are to be dealt with: environmental, procedural and social knowledge.

Many practical applications (e.g., plannning for cyber-physical systems) require reasoning about hybrid domains that contain both probabilistic and nondeterministic parametric uncertainty. In general, this is an undecidable problem. We use delta-satisfiability to sidestep undecidability, and we develop an algorithm that computes an enclosure for the range of probability of reaching a goal region in a given number of discrete steps. We utilize SMT techniques that enable reasoning in a safe way, i.e., the computed enclosure is formally guaranteed to contain the reachability probability. We demonstrate the usefulness of our technique on challenging nonlinear hybrid domains.

Reasoning with commonsense knowledge plays an important role in various NLU tasks. Often the commonsense knowledge is needed to be extracted separately. In this paper we present our work of automatically extracting a certain type of commonsense knowledge. The knowledge resembles the kind that humans have about the events and the entities that participate in those events. One example of such knowledge is that "IF A bullying B causes T rescued Z THEN (possibly) Z = B ''. We call this knowledge an event-based conditional commonsense. Our approach involves semantic parsing of natural language sentences by using the Knowledge Parser (K-Parser) and extracting the knowledge, if found. We extracted about 19000 instances of such knowledge from the Open American National Corpus.

While there exist several approaches in the constraint programming community to learn a constraint theory, few of them have considered the learning of constraint optimization problems.To alleviate this situation, we introduce an initial approach to learning first-order weighted MAX-SAT theories. It employs inductive logic programming techniques to learn a set of first-order clauses and then uses preference learning techniques to learn the weights of the clauses.In order to learn these weighted clauses, the clausal optimization system uses examples of possible worlds and a set of preferences that state which examples are preferred over other ones.The technique is also empirically evaluated on a number of examples.These experiments show that the system is capable of learning clauses and weights that accurately capture underlying models.

This paper describes a collection of ideas that allow large games of imperfect information to be solved with counterfactual regret minimization (CFR) using little memory. We replace the regret matching component of CFR with a simple approach known as "follow-the-leader." This helps us quantize the regret values computed in CFR to a single byte. We also investigate not maintaining the accumulated strategy, which saves additional memory. Ultimately, our collection of techniques allows CFR to be run with only 1/16 of the memory required by classic approaches. We present experimental results on poker.

Model counting and weighted model counting are key problems in artificial intelligence. Marginal inference can be reduced to model counting in many statistical-relational systems, such as Markov Logic. One common approach used by model counters is splitting a theory into disjoint subtheories, performing model counting on the subtheories, and then caching the result. If an identical subtheory is encountered again in the search, the cached result is used, greatly reducing runtime. In this work we introduce a way to cache symmetric subtheories compactly, which could potentially decrease required cache size, increase cache hits, and decrease runtime of solving.