Both the ethics of autonomous systems and the problems of their technical implementation have by now been studied in some detail. Less attention has been given to the areas in which these two separate concerns meet. This paper, written by both philosophers and engineers of autonomous systems, addresses a number of issues in machine ethics that are located at precisely the intersection between ethics and engineering. We first discuss the main challenges which, in our view, machine ethics posses to moral philosophy. We them consider different approaches towards the conceptual design of autonomous systems and their implications on the ethics implementation in such systems. Then we examine problematic areas regarding the specification and verification of ethical behavior in autonomous systems, particularly with a view towards the requirements of future legislation. We discuss transparency and accountability issues that will be crucial for any future wide deployment of autonomous systems in society. Finally we consider the, often overlooked, possibility of intentional misuse of AI systems and the possible dangers arising out of deliberately unethical design, implementation, and use of autonomous robots.

The ability to act upon a new stimulus, based on previous experience with similar, but distinct, stimuli, sometimes denoted as generalization, is used extensively in our daily life. As a simple example, consider a driver's response to traffic lights: The driver need not recognize the details of a particular traffic light in order to respond to it correctly, even though traffic lights may appear different from one another. The only property that matters is the color, whereas neither shape nor size should play any role in the driver's reaction. Learning how to react to traffic lights thus involves an aspect of generalization. A learning agent, capable of a meaningful and useful generalization is expected to have the following characteristics: (a) an ability for categorization (recognizing that all red signals have a common property, which we can refer to as redness); (b) an ability to classify (a new red object is to be related to the group of objects with the redness property); (c) ideally, only generalizations that are relevant for the success of the agent should be learned (red signals should be treated the same, whereas squareshaped signals should not, as they share no property that is of relevance in this context); (d) correct actions should be associated with relevant generalized properties (the driver should stop whenever a red signal is shown); and (e) the generalization mechanism should be flexible. To illustrate what we mean by "flexible generalization", let us go back to our driver. After learning how to handle traffic lights correctly, the driver tries to follow arrow signs to, say, a nearby airport. Clearly, it is now the shape category of the signal that should guide the driver, rather than the color category.

In this paper we consider distributed optimization problems over a multi-agent network, where each agent can only partially evaluate the objective function, and it is allowed to exchange messages with its immediate neighbors. Differently from all existing works on distributed optimization, our focus is given to optimizing a class of difficult non-convex problems, and under the challenging setting where each agent can only access the zeroth-order information (i.e., the functional values) of its local functions. For different types of network topologies such as undirected connected networks or star networks, we develop efficient distributed algorithms and rigorously analyze their convergence and rate of convergence (to the set of stationary solutions). Numerical results are provided to demonstrate the efficiency of the proposed algorithms.

I am a poet and artist, and I make Twitter bots. The term "bot" encompasses many different kinds of software agents, from conversation simulators like Eliza, to programs that write stories about sports events without human intervention, to automatically created social media accounts that spam hashtags. The bots I make have an artistic and literary bent: for example, I made @everyword -- which tweeted every word in the English language in alphabetical order over the course of seven years -- and more recently @the_ephemerides, which tweets computer-generated poetry juxtaposed with NASA imagery. I'm part of a community of bot-making artists (loosely known as #botALLY) who are taking the canvas of social media and covering it with computer-generated writing and other kinds of generative art. As part of Sam Woolley's provocateur-in-residence workshop at Data & Society, I was asked to write a provocation regarding automated agents and bots from my perspective as a poet and artist.

Due to the lack of enough generalization in the state-space, common methods in Reinforcement Learning (RL) suffer from slow learning speed especially in the early learning trials. This paper introduces a model-based method in discrete state-spaces for increasing learning speed in terms of required experience (but not required computational time) by exploiting generalization in the experiences of the subspaces. A subspace is formed by choosing a subset of features in the original state representation (full-space). Generalization and faster learning in a subspace are due to many-to-one mapping of experiences from the full-space to each state in the subspace. Nevertheless, due to inherent perceptual aliasing in the subspaces, the policy suggested by each subspace does not generally converge to the optimal policy. Our approach, called Model Based Learning with Subspaces (MoBLeS), calculates confidence intervals of the estimated Q-values in the full-space and in the subspaces. These confidence intervals are used in the decision making, such that the agent benefits the most from the possible generalization while avoiding from detriment of the perceptual aliasing in the subspaces. Convergence of MoBLeS to the optimal policy is theoretically investigated. Additionally, we show through several experiments that MoBLeS improves the learning speed in the early trials.

The use of the agent paradigm to understand and design complex systems occupies an important and growing role in different areas of social and natural sciences and technology. Application areas where the agent paradigm delivers appropriate solutions include online trading,16 disaster management,10 and policy making.11 However, the two main agent approaches, Multi-Agent Systems (MAS) and Agent-Based Modeling (ABM) differ considerably in methodology, applications, and aims. MAS focus on solving specific complex problems using autonomous heterogeneous agents, while ABM is used to capture the dynamics of a (social or technical) system for analytical purposes. ABM is a form of computational modeling whereby a population of individual agents is given simple rules to govern their behavior such that global properties of the whole can be analyzed.9

It's been a long time worry that when AI gains a certain level of autonomy it will see no use for humans or even perceive them as a threat. A new study by Google's DeepMind lab may or may not ease those fears. There are two unmistakable sides to the debate concerning the future of artificial intelligence. The researchers at DeepMind have been working with two games to test whether neural networks are more likely to understand motivations to compete or cooperate. They hope that this research could lead to AI being better at working with other AI in situations that contain imperfect information.

We set up competitions between multiple simulated 3D robots on a range of basic games, trained each agent with simple goals (push the opponent out of the sumo ring, reach the other side of the ring while preventing the other agent from doing the same, kick the ball into the net or prevent the other agent from doing so, and so on), then analyzed the different strategies that emerged. Agents initially receive dense rewards for behaviours that aid exploration like standing and moving forward, which are eventually annealed to zero in favor of being rewarded for just winning and losing. Despite the simple rewards, the agents learn subtle behaviors like tackling, ducking, faking, kicking and catching, and diving for the ball. Each agent's neural network policy is independently trained with Proximal Policy Optimization. To understand how complex behaviors can emerge through a combination of simple goals and competitive pressure, let's analyze the sumo wrestling task.

On Sunday night, a lone gunman armed with 23 powerful weapons opened fire from a window on the 32nd floor of the Mandalay Bay casino in Las Vegas, targeting thousands concertgoers gathered for a country music festival below. Within minutes, the gunman had killed at least 58 people and injured nearly 500. There may be no more contentious issue in America than gun control, and no more emotional time to discuss it than in the days following a national tragedy like the one that unfolded in Las Vegas. But, a subject being difficult to discuss should not preclude us from trying to understand it, and fortunately nature has evolved methods for helping relatively simple organisms work through incredibly complicated problems with life or death consequences. Swarm Intelligence allows groups of bees to converge on the perfect place for their hive nearly 90% of the time, and extending this power to humans through Unanimous AI's Swarm AI technology empowers groups to create similarly optimized insight.