The Atari 2600 games supported in the Arcade Learning Environment (Bellemare et al. 2013) all feature aknown initial (RAM) state and actions that have deterministic effects. Classical planners, however, cannot be used for selecting actions for two reasons: first, nocompact PDDL-model of the games is given, and more importantly, the action effects and goals are not known a priori. Moreover, in these games there is usually no set of goals to be achieved but rewards to be collected. These features do not preclude the use of classical algorithms like breadth-first search or Dijkstra’s algorithm, but these methods are not effective over large state spaces. We thus turn to a different class of classical planning algorithms introduced recently that perform a structured exploration of the state space; namely, like breadth-first search and Dijkstra’s algorithm they are“blind” and hence do not require prior knowledge of state transitions, costs (rewards) or goals, and yet, like heuristic search algorithms, they have been shown to be effective for solving problems over huge state spaces.The simplest such algorithm, called Iterated Width or IW, consists of a sequence of calls IW(1), IW(2), . . . ,IW(k) where IW(i) is a breadth-first search in which a state is pruned when it is not the first state in the search to make true some subset of i atoms. The empirical results over 54 games suggest that the performance of IW with the k parameter fixed to 1, i.e., IW(1), is at the level of the state of the art represented by UCT. A simple best-first variation of IW that combines exploration and exploitation proves to be very competitive as well.

Residential and commercial buildings are responsible for about 40% of primary energy consumption in the United States, hence improving their energy efficiency could have important sustainability benefits. The design of a building has tremendous effect on its energy profile, and recently there has been an increased interest in developing optimization methods that support the design of high performance buildings. Previous approaches are either based on simulation optimization or on training an accurate predictive model that is queried during the optimization. We propose a method that more tightly integrates the machine learning and optimization components, by employing active learning during optimization. In particular, we use a Gaussian Process (GP) model for the prediction and active learning and multi-objective genetic algorithm NSGA-II for the optimization. We develop a comprehensive and publicly available benchmark for building design optimization. We evaluate 5 machine learning approaches on our dataset, and show that the GP model is competitive, in addition to being well-suited for the active learning setting. We compare our optimization approach against the 2-stage approach and simulation optimization. Our results show that our approach produces solutions at the Pareto frontier compared to the other two approaches, while using only a fraction of the simulations and time.

Ensuring that autonomous systems work ethically is both complex and difficult. However, the idea of having an additional ‘governor’ that assesses options the system has, and prunes them to select the most ethical choices is well understood. Recent work has produced such a governor consisting of a ‘consequence engine’ that assesses the likely future outcomes of actions then applies a Safety/Ethical logic to select actions. Although this is appealing, it is impossible to be certain that the most ethical options are actually taken. In this paper we extend and apply a well-known agent verification approach to our consequence engine, allowing us to verify the correctness of its ethical decision-making.

Many experts believe that in the coming decades, artificial intelligence will change, and perhaps significantly reduce, the demand for human labor in the economy, but there remains much uncertainty about the accuracy of this claim and what to do about it. This paper identifies several ways in which the artificial intelligence community can help society to anticipate and shape such outcomes in a socially beneficial direction. First, different technical aspirations for the field of AI may be associated with different social outcomes, increasing the stakes of decisions made in the AI community. Second, the extent of researchers' efforts to apply AI to different social and economic domains will influence the distribution of cognition between humans and machines in those domains. Third, the AI community can play a key role in initiating a more nuanced and inclusive public discussion of the social and economic possibilities afforded by AI technologies. To pave the way for such dialogue, we suggest a line of research aimed at better understanding the nature, pace, and drivers of progress in AI in order to more effectively anticipate and shape AI's role in society.

The cultural and political implications of modern AI research are not some far off concern, they are things that affect the world in the here and now. From advanced control systems with advanced visualizations and image processing techniques that drive the machines of the modern military to the slow creep of a mechanized workforce, ethical questions surround us. Part of dealing with these ethical questions is not just speculating on what could be but teaching our students how to engage with these ethical questions. We explore the use of science fiction as an appropriate tool to enable AI researchers to help engage students and the public on the current state and potential impacts of AI.

The Smart Grid allows users to monitor power usage through the use of Smart Meter technology. In principle, this information can be used to modify usage habits in a way that reduces consumer costs as well as greenhouse emissions. However, in an urban environment, many users are restricted by the same constaints: they work during the day, and they are home at night. This creates spikes in power cost at peak usage times, and it may also lead to increased emissions in scenarios where sustainable resources are limited. An individual user can avoid these spikes by using an electric car as a storage device; it can be charged at the cheapest times, and then discharged to the home at the most expensive times. While this idea is intuitively appealing, it turns out that the benefits vary greatly depending on the storage algorithm used. In this paper, we describe the Power Storage Simulator, a tool for experimenting with storage algorithms to improve the efficiency of vehicle to grid systems. We suggest that this tool is also useful for educating power consumers about load balancing on the Smart Grid through an engaging, visual simulation.

We analyze the results of assistance to individuals with autism in reasoning about mental world. This assistance is provided by a natural language multiagent simulator of mental states, NL_MAMS (Galitsky 2013b). It assists in the tasks which are the hardest for autistic reasoning: operating with mental states and actions. Autistic patients are trained to perform a number of reasoning exercises. We conduct both short term and long term evaluations including the behavior in real world and confirm that the system has a positive effect on their rehabilitation.

One approach to conserving endangered species is to purchase and protect a set of land parcels in a way that maximizes the expected future population spread. Unfortunately, an ideal set of parcels may have a cost that is beyond the immediate budget constraints and must thus be purchased incrementally. This raises the challenge of deciding how to schedule the parcel purchases in a way that maximizes the flexibility of budget usage while keeping population spread loss in control. In this paper, we introduce a formulation of this scheduling problem that does not rely on knowing the future budgets of an organization. In particular, we consider scheduling purchases in a way that achieves a population spread no less than desired but delays purchases as long as possible. Such schedules offer conservation planners maximum flexibility and use available budgets in the most efficient way. We develop the problem formally as a stochastic optimization problem over a network cascade model describing a commonly used model of population spread. Our solution approach is based on reducing the stochastic problem to a novel variant of the directed Steiner tree problem, which we call the set-weighted directed Steiner graph problem. We show that this problem is computationally hard, motivating the development of a primal-dual algorithm for the problem that computes both a feasible solution and a bound on the quality of an optimal solution. We evaluate the approach on both real and synthetic conservation data with a standard population spread model. The algorithm is shown to produce near optimal results and is much more scalable than more generic off-the-shelf optimizers. Finally, we evaluate a variant of the algorithm to explore the trade-offs between budget savings and population growth.

This paper focuses on a preliminary experimental protocol that aims at assessing a robot operator’s behavior when the robot is equipped with what appears as moral decision capabilities. The protocol is derived from the trolley dilemma, a well-known decision making paradigm. Indeed the participants, acting as operators of simulated aerial robots via a computer screen, are faced to impersonal moral dilemmas, i.e. decide to crash a damaged robot on one of two inhabited areas, and to non-moral choices, i.e. decide to crash a damaged robot on one of two uninhabited areas. In each situation, the robot has a default crash behavior which is displayed to the participant who will have to decide whether to follow it or not. The participants are equipped with fNIRS and eye-tracking and answer a post-experimental questionnaire. As some of the behavioral and physiological results do not match the hypotheses we had set, we give the features of the further experiments that we are planning.

Maintaining an accurate set of beliefs in a partially observable scenario, particularly with respect to other agents operating in the same space, is a vital aspect of multiagent planning. We analyze how the beliefs of an agent can be updated for fast adaptivity to changes in the behavior of an unknown teammate. The main contribution of this paper is the empirical evaluation of an agent cooperating with a teammate whose goals change periodically. We test our approach in a collaborative multiagent domain where identification of goals is necessary for successful completion. The belief revision technique we propose outperforms the traditional approach in a majority of test cases. Additionally, our results suggest the ability to approximate a higher level model by utilizing a belief distribution over a set of lower level behaviors, particularly when the belief update strategy identifies changes in the behavior in a responsive manner.