This paper analyzes and compares 11 different proposals for building safe advanced AI under the current machine learning paradigm, including major contenders such as iterated amplification, AI safety via debate, and recursive reward modeling. Each proposal is evaluated on the four components of outer alignment, inner alignment, training competitiveness, and performance competitiveness, of which the distinction between the latter two is introduced in this paper. While prior literature has primarily focused on analyzing individual proposals, or primarily focused on outer alignment at the expense of inner alignment, this analysis seeks to take a comparative look at a wide range of proposals including a comparative analysis across all four previously mentioned components.

There is a need for socially assistive robots (SARs) to provide transparency in their behavior by explaining their reasoning. Additionally, the reasoning and explanation should represent the user's preferences and goals. To work towards satisfying this need for interpretable reasoning and representations, we propose the robot uses Analogical Theory of Mind to infer what the user is trying to do and uses the Hint Engine to find an appropriate assistance based on what the user is trying to do. If the user is unsure or confused, the robot provides the user with an explanation, generated by the Explanation Synthesizer. The explanation helps the user understand what the robot inferred about the user's preferences and why the robot decided to provide the assistance it gave. A knowledge-driven approach provides transparency to reasoning about preferences, assistance, and explanations, thereby facilitating the incorporation of user feedback and allowing the robot to learn and adapt to the user.

In many real world situations, collective decisions are made using voting and, in scenarios such as committee or board elections, employing voting rules that return multiple winners. In multi-winner approval voting (AV), an agent submits a ballot consisting of approvals for as many candidates as they wish, and winners are chosen by tallying up the votes and choosing the top-$k$ candidates receiving the most approvals. In many scenarios, an agent may manipulate the ballot they submit in order to achieve a better outcome by voting in a way that does not reflect their true preferences. In complex and uncertain situations, agents may use heuristics instead of incurring the additional effort required to compute the manipulation which most favors them. In this paper, we examine voting behavior in single-winner and multi-winner approval voting scenarios with varying degrees of uncertainty using behavioral data obtained from Mechanical Turk. We find that people generally manipulate their vote to obtain a better outcome, but often do not identify the optimal manipulation. There are a number of predictive models of agent behavior in the COMSOC and psychology literature that are based on cognitively plausible heuristic strategies. We show that the existing approaches do not adequately model real-world data. We propose a novel model that takes into account the size of the winning set and human cognitive constraints, and demonstrate that this model is more effective at capturing real-world behaviors in multi-winner approval voting scenarios.

We consider a scenario in which two reinforcement learning agents repeatedly play a matrix game against each other and update their parameters after each round. The agents' decision-making is transparent to each other, which allows each agent to predict how their opponent will play against them. To prevent an infinite regress of both agents recursively predicting each other indefinitely, each agent is required to give an opponent-independent response with some probability at least epsilon. Transparency also allows each agent to anticipate and shape the other agent's gradient step, i.e. to move to regions of parameter space in which the opponent's gradient points in a direction favourable to them. We study the resulting dynamics experimentally, using two algorithms from previous literature (LOLA and SOS) for opponent-aware learning. We find that the combination of mutually transparent decision-making and opponent-aware learning robustly leads to mutual cooperation in a single-shot prisoner's dilemma. In a game of chicken, in which both agents try to manoeuvre their opponent towards their preferred equilibrium, converging to a mutually beneficial outcome turns out to be much harder, and opponent-aware learning can even lead to worst-case outcomes for both agents. This highlights the need to develop opponent-aware learning algorithms that achieve acceptable outcomes in social dilemmas involving an equilibrium selection problem.

In crowdsourcing systems, it is important for the crowdsource campaign initiator to incentivize users to share their data to produce results of the desired computational accuracy. This problem becomes especially challenging when users are concerned about the privacy of their data. To overcome this challenge, existing work often aims to provide users with differential privacy guarantees to incentivize privacy-sensitive users to share their data. However, this work neglects the network effect that a user enjoys greater privacy protection when he aligns his participation behaviour with that of other users. To explore this network effect, we formulate the interaction among users regarding their participation decisions as a population game, because a user's welfare from the interaction depends not only on his own participation decision but also the distribution of others' decisions. We show that the Nash equilibrium of this game consists of a threshold strategy, where all users whose privacy sensitivity is below a certain threshold will participate and the remaining users will not. We characterize the existence and uniqueness of this equilibrium, which depends on the privacy guarantee, the reward provided by the initiator and the population size. Based on this equilibria analysis, we design the PINE (Privacy Incentivization with Network Effects) mechanism and prove that it maximizes the initiator's payoff while providing participating users with a guaranteed degree of privacy protection. Numerical simulations, on both real and synthetic data, show that (i) PINE improves the initiator's expected payoff by up to 75%, compared to state of the art mechanisms that do not consider this effect; (ii) the performance gain by exploiting the network effect is particularly good when the majority of users are flexible over their privacy attitudes and when there are a large number of low quality task performers.

The planning domain has experienced increased interest in the formal synthesis of decision-making policies. This formal synthesis typically entails finding a policy which satisfies formal specifications in the form of some well-defined logic, such as Linear Temporal Logic (LTL) or Computation Tree Logic (CTL), among others. While such logics are very powerful and expressive in their capacity to capture desirable agent behavior, their value is limited when deriving decision-making policies which satisfy certain types of asymptotic behavior. In particular, we are interested in specifying constraints on the steady-state behavior of an agent, which captures the proportion of time an agent spends in each state as it interacts for an indefinite period of time with its environment. This is sometimes called the average or expected behavior of the agent. In this paper, we explore the steady-state planning problem of deriving a decision-making policy for an agent such that constraints on its steady-state behavior are satisfied. A linear programming solution for the general case of multichain Markov Decision Processes (MDPs) is proposed and we prove that optimal solutions to the proposed programs yield stationary policies with rigorous guarantees of behavior.

A wide range of reinforcement learning (RL) problems -- including robustness, transfer learning, unsupervised RL, and emergent complexity -- require specifying a distribution of tasks or environments in which a policy will be trained. However, creating a useful distribution of environments is error prone, and takes a significant amount of developer time and effort. We propose Unsupervised Environment Design (UED) as an alternative paradigm, where developers provide environments with unknown parameters, and these parameters are used to automatically produce a distribution over valid, solvable environments. Existing approaches to automatically generating environments suffer from common failure modes: domain randomization cannot generate structure or adapt the difficulty of the environment to the agent's learning progress, and minimax adversarial training leads to worst-case environments that are often unsolvable. To generate structured, solvable environments for our protagonist agent, we introduce a second, antagonist agent that is allied with the environment-generating adversary. The adversary is motivated to generate environments which maximize regret, defined as the difference between the protagonist and antagonist agent's return. We call our technique Protagonist Antagonist Induced Regret Environment Design (PAIRED). Our experiments demonstrate that PAIRED produces a natural curriculum of increasingly complex environments, and PAIRED agents achieve higher zero-shot transfer performance when tested in highly novel environments.

In this paper we introduce DeepCrawl, a fully-playable Roguelike prototype for iOS and Android in which all agents are controlled by policy networks trained using Deep Reinforcement Learning (DRL). Our aim is to understand whether recent advances in DRL can be used to develop convincing behavioral models for non-player characters in videogames. We begin with an analysis of requirements that such an AI system should satisfy in order to be practically applicable in video game development, and identify the elements of the DRL model used in the DeepCrawl prototype. The successes and limitations of DeepCrawl are documented through a series of playability tests performed on the final game. We believe that the techniques we propose offer insight into innovative new avenues for the development of behaviors for non-player characters in video games, as they offer the potential to overcome critical issues with

We study a cooperative multi-agent multi-armed bandits with M agents and K arms. The goal of the agents is to minimized the cumulative regret. We adapt a traditional Thompson Sampling algoirthm under the distributed setting. However, with agent's ability to communicate, we note that communication may further reduce the upper bound of the regret for a distributed Thompson Sampling approach. To further improve the performance of distributed Thompson Sampling, we propose a distributed Elimination based Thompson Sampling algorithm that allow the agents to learn collaboratively. We analyse the algorithm under Bernoulli reward and derived a problem dependent upper bound on the cumulative regret.

Power systems are becoming more complex to operate in the digital age. As a result, real-time decision-making is getting more challenging as the human operator has to deal with more information, more uncertainty, more applications and more coordination. While supervision has been primarily used to help them make decisions over the last decades, it cannot reasonably scale up anymore. There is a great need for rethinking the human-machine interface under more unified and interactive frameworks. Taking advantage of the latest developments in Human-machine Interactions and Artificial intelligence, we share the vision of a new assistant framework relying on an hypervision interface and greater bidirectional interactions. We review the known principles of decision-making that drives the assistant design and supporting assistance functions we present. We finally share some guidelines to make progress towards the development of such an assistant.