I model the belief formation and decision making processes of economic agents during a monetary policy regime change (an acceleration in the money supply) with a deep reinforcement learning algorithm in the AI literature. I show that when the money supply accelerates, the learning agents only adjust their actions, which include consumption and demand for real balance, after gathering learning experience for many periods. This delayed adjustments leads to low returns during transition periods. Once they start adjusting to the new environment, their welfare improves. Their changes in beliefs and actions lead to temporary inflation volatility. I also show that, 1. the AI agents who explores their environment more adapt to the policy regime change quicker, which leads to welfare improvements and less inflation volatility, and 2. the AI agents who have experienced a structural change adjust their beliefs and behaviours quicker than an inexperienced learning agent.
The ability of an AI agent to assist other agents, such as humans, is an important and challenging goal, which requires the assisting agent to reason about the behavior and infer the goals of the assisted agent. Training such an ability by using reinforcement learning usually requires large amounts of online training, which is difficult and costly. On the other hand, offline data about the behavior of the assisted agent might be available, but is non-trivial to take advantage of by methods such as offline reinforcement learning. We introduce methods where the capability to create a representation of the behavior is first pre-trained with offline data, after which only a small amount of interaction data is needed to learn an assisting policy. We test the setting in a gridworld where the helper agent has the capability to manipulate the environment of the assisted artificial agents, and introduce three different scenarios where the assistance considerably improves the performance of the assisted agents.
A big part of achieving Artificial General Intelligence(AGI) is to build a machine that can see and listen like humans. Much work has focused on designing models for image classification, video classification, object detection, pose estimation, speech recognition, etc., and has achieved significant progress in recent years thanks to deep learning. However, understanding the world is not enough. An AI agent also needs to know how to talk, especially how to communicate with a human. While perception (vision, for example) is more common across animal species, the use of complicated language is unique to humans and is one of the most important aspects of intelligence. In this thesis, we focus on generating textual output given visual input. In Chapter 3, we focus on generating the referring expression, a text description for an object in the image so that a receiver can infer which object is being described. We use a comprehension machine to directly guide the generated referring expressions to be more discriminative. In Chapter 4, we introduce a method that encourages discriminability in image caption generation. We show that more discriminative captioning models generate more descriptive captions. In Chapter 5, we study how training objectives and sampling methods affect the models' ability to generate diverse captions. We find that a popular captioning training strategy will be detrimental to the diversity of generated captions. In Chapter 6, we propose a model that can control the length of generated captions. By changing the desired length, one can influence the style and descriptiveness of the captions. Finally, in Chapter 7, we rank/generate informative image tags according to their information utility. The proposed method better matches what humans think are the most important tags for the images.
Recognizing and localizing objects in the 3D space is a crucial ability for an AI agent to perceive its surrounding environment. While significant progress has been achieved with expensive LiDAR point clouds, it poses a great challenge for 3D object detection given only a monocular image. While there exist different alternatives for tackling this problem, it is found that they are either equipped with heavy networks to fuse RGB and depth information or empirically ineffective to process millions of pseudo-LiDAR points. With in-depth examination, we realize that these limitations are rooted in inaccurate object localization. In this paper, we propose a novel and lightweight approach, dubbed {\em Progressive Coordinate Transforms} (PCT) to facilitate learning coordinate representations. Specifically, a localization boosting mechanism with confidence-aware loss is introduced to progressively refine the localization prediction. In addition, semantic image representation is also exploited to compensate for the usage of patch proposals. Despite being lightweight and simple, our strategy leads to superior improvements on the KITTI and Waymo Open Dataset monocular 3D detection benchmarks. At the same time, our proposed PCT shows great generalization to most coordinate-based 3D detection frameworks. The code is available at: https://github.com/amazon-research/progressive-coordinate-transforms .
Several approaches have been developed for answering users' specific questions about AI behavior and for assessing their core functionality in terms of primitive executable actions. However, the problem of summarizing an AI agent's broad capabilities for a user is comparatively new. This paper presents an algorithm for discovering from scratch the suite of high-level "capabilities" that an AI system with arbitrary internal planning algorithms/policies can perform. It computes conditions describing the applicability and effects of these capabilities in user-interpretable terms. Starting from a set of user-interpretable state properties, an AI agent, and a simulator that the agent can interact with, our algorithm returns a set of high-level capabilities with their parameterized descriptions. Empirical evaluation on several game-based scenarios shows that this approach efficiently learns descriptions of various types of AI agents in deterministic, fully observable settings. User studies show that such descriptions are easier to understand and reason with than the agent's primitive actions.
This paper presents a novel approach to automated playtesting for the prediction of human player behavior and experience. It has previously been demonstrated that Deep Reinforcement Learning (DRL) game-playing agents can predict both game difficulty and player engagement, operationalized as average pass and churn rates. We improve this approach by enhancing DRL with Monte Carlo Tree Search (MCTS). We also motivate an enhanced selection strategy for predictor features, based on the observation that an AI agent's best-case performance can yield stronger correlations with human data than the agent's average performance. Both additions consistently improve the prediction accuracy, and the DRL-enhanced MCTS outperforms both DRL and vanilla MCTS in the hardest levels. We conclude that player modelling via automated playtesting can benefit from combining DRL and MCTS. Moreover, it can be worthwhile to investigate a subset of repeated best AI agent runs, if AI gameplay does not yield good predictions on average.
Video Question Answering is a task which requires an AI agent to answer questions grounded in video. This task entails three key challenges: (1) understand the intention of various questions, (2) capturing various elements of the input video (e.g., object, action, causality), and (3) cross-modal grounding between language and vision information. We propose Motion-Appearance Synergistic Networks (MASN), which embed two cross-modal features grounded on motion and appearance information and selectively utilize them depending on the question's intentions. MASN consists of a motion module, an appearance module, and a motion-appearance fusion module. The motion module computes the action-oriented cross-modal joint representations, while the appearance module focuses on the appearance aspect of the input video. Finally, the motion-appearance fusion module takes each output of the motion module and the appearance module as input, and performs question-guided fusion. As a result, MASN achieves new state-of-the-art performance on the TGIF-QA and MSVD-QA datasets. We also conduct qualitative analysis by visualizing the inference results of MASN. The code is available at https://github.com/ahjeongseo/MASN-pytorch.
There is a growing interest in the community in making an embodied AI agent perform a complicated task while interacting with an environment following natural language directives. Recent studies have tackled the problem using ALFRED, a well-designed dataset for the task, but achieved only very low accuracy. This paper proposes a new method, which outperforms the previous methods by a large margin. It is based on a combination of several new ideas. One is a two-stage interpretation of the provided instructions. The method first selects and interprets an instruction without using visual information, yielding a tentative action sequence prediction. It then integrates the prediction with the visual information etc., yielding the final prediction of an action and an object. As the object's class to interact is identified in the first stage, it can accurately select the correct object from the input image. Moreover, our method considers multiple egocentric views of the environment and extracts essential information by applying hierarchical attention conditioned on the current instruction. This contributes to the accurate prediction of actions for navigation. A preliminary version of the method won the ALFRED Challenge 2020. The current version achieves the unseen environment's success rate of 4.45% with a single view, which is further improved to 8.37% with multiple views.
This exercise proposes a learning mechanism to model economic agent's decision-making process using an actor-critic structure in the literature of artificial intelligence. It is motivated by the psychology literature of learning through reinforcing good or bad decisions. In a model of an environment, to learn to make decisions, this AI agent needs to interact with its environment and make explorative actions. Each action in a given state brings a reward signal to the agent. These interactive experience is saved in the agent's memory, which is then used to update its subjective belief of the world. The agent's decision-making strategy is formed and adjusted based on this evolving subjective belief. This agent does not only take an action that it knows would bring a high reward, it also explores other possibilities. This is the process of taking explorative actions, and it ensures that the agent notices changes in its environment and adapt its subjective belief and decisions accordingly. Through a model of stochastic optimal growth, I illustrate that the economic agent under this proposed learning structure is adaptive to changes in an underlying stochastic process of the economy. AI agents can differ in their levels of exploration, which leads to different experience in the same environment. This reflects on to their different learning behaviours and welfare obtained. The chosen economic structure possesses the fundamental decision making problems of macroeconomic models, i.e., how to make consumption-saving decisions in a lifetime, and it can be generalised to other decision-making processes and economic models.
Affordances describe the possibilities for an agent to perform actions with an object. While the significance of the affordance concept has been previously studied from varied perspectives, such as psychology and cognitive science, these approaches are not always sufficient to enable direct transfer, in the sense of implementations, to artificial intelligence (AI)-based systems and robotics. However, many efforts have been made to pragmatically employ the concept of affordances, as it represents great potential for AI agents to effectively bridge perception to action. In this survey, we review and find common ground amongst different strategies that use the concept of affordances within robotic tasks, and build on these methods to provide guidance for including affordances as a mechanism to improve autonomy. To this end, we outline common design choices for building representations of affordance relations, and their implications on the generalisation capabilities of an agent when facing previously unseen scenarios. Finally, we identify and discuss a range of interesting research directions involving affordances that have the potential to improve the capabilities of an AI agent.
This paper addresses the problem of synthesizing the behavior of an AI agent that provides proactive task assistance to a human in settings like factory floors where they may coexist in a common environment. Unlike in the case of requested assistance, the human may not be expecting proactive assistance and hence it is crucial for the agent to ensure that the human is aware of how the assistance affects her task. This becomes harder when there is a possibility that the human may neither have full knowledge of the AI agent's capabilities nor have full observability of its activities. Therefore, our \textit{proactive assistant} is guided by the following three principles: \textbf{(1)} its activity decreases the human's cost towards her goal; \textbf{(2)} the human is able to recognize the potential reduction in her cost; \textbf{(3)} its activity optimizes the human's overall cost (time/resources) of achieving her goal. Through empirical evaluation and user studies, we demonstrate the usefulness of our approach.
Assessing an AI agent that can converse in human language and understand visual content is challenging. Generation metrics, such as BLEU scores favor correct syntax over semantics. Hence a discriminative approach is often used, where an agent ranks a set of candidate options. The mean reciprocal rank (MRR) metric evaluates the model performance by taking into account the rank of a single human-derived answer. This approach, however, raises a new challenge: the ambiguity and synonymy of answers, for instance, semantic equivalence (e.g., `yeah' and `yes'). To address this, the normalized discounted cumulative gain (NDCG) metric has been used to capture the relevance of all the correct answers via dense annotations. However, the NDCG metric favors the usually applicable uncertain answers such as `I don't know. Crafting a model that excels on both MRR and NDCG metrics is challenging. Ideally, an AI agent should answer a human-like reply and validate the correctness of any answer. To address this issue, we describe a two-step non-parametric ranking approach that can merge strong MRR and NDCG models. Using our approach, we manage to keep most MRR state-of-the-art performance (70.41% vs. 71.24%) and the NDCG state-of-the-art performance (72.16% vs. 75.35%). Moreover, our approach won the recent Visual Dialog 2020 challenge. Source code is available at https://github.com/idansc/mrr-ndcg.
Reinforcement learning, which acquires a policy maximizing long-term rewards, has been actively studied. Unfortunately, this learning type is too slow and difficult to use in practical situations because the state-action space becomes huge in real environments. Many studies have incorporated human knowledge into reinforcement Learning. Though human knowledge on trajectories is often used, a human could be asked to control an AI agent, which can be difficult. Knowledge on subgoals may lessen this requirement because humans need only to consider a few representative states on an optimal trajectory in their minds. The essential factor for learning efficiency is rewards. Potential-based reward shaping is a basic method for enriching rewards. However, it is often difficult to incorporate subgoals for accelerating learning over potential-based reward shaping. This is because the appropriate potentials are not intuitive for humans. We extend potential-based reward shaping and propose a subgoal-based reward shaping. The method makes it easier for human trainers to share their knowledge of subgoals. To evaluate our method, we obtained a subgoal series from participants and conducted experiments in three domains, four-rooms(discrete states and discrete actions), pinball(continuous and discrete), and picking(both continuous). We compared our method with a baseline reinforcement learning algorithm and other subgoal-based methods, including random subgoal and naive subgoal-based reward shaping. As a result, we found out that our reward shaping outperformed all other methods in learning efficiency.
This paper describes an AI agent that plays the popular first-person-shooter (FPS) video game `Counter-Strike; Global Offensive' (CSGO) from pixel input. The agent, a deep neural network, matches the performance of the medium difficulty built-in AI on the deathmatch game mode, whilst adopting a humanlike play style. Unlike much prior work in games, no API is available for CSGO, so algorithms must train and run in real-time. This limits the quantity of on-policy data that can be generated, precluding many reinforcement learning algorithms. Our solution uses behavioural cloning - training on a large noisy dataset scraped from human play on online servers (4 million frames, comparable in size to ImageNet), and a smaller dataset of high-quality expert demonstrations. This scale is an order of magnitude larger than prior work on imitation learning in FPS games.
This paper addresses ontology and ethics of an AI agent called digital me. We define digital me as autonomous, decision-making, and learning agent, representing an individual and having practically immortal own life. It is assumed that digital me is equipped with the big-five personality model, ensuring that it provides a model of some aspects of a strong AI: consciousness, free will, and intentionality. As computer-based personality judgments are more accurate than those made by humans, digital me can judge the personality of the individual represented by the digital me, other individuals' personalities, and other digital me-s. We describe seven ontological qualities of digital me: a) double-layer status of Digital Being versus digital me, b) digital me versus real me, c) mind-digital me and body-digital me, d) digital me versus doppelganger (shadow digital me), e) non-human time concept, f) social quality, g) practical immortality. We argue that with the advancement of AI's sciences and technologies, there exist two digital me thresholds. The first threshold defines digital me having some (rudimentarily) form of consciousness, free will, and intentionality. The second threshold assumes that digital me is equipped with moral learning capabilities, implying that, in principle, digital me could develop their own ethics which significantly differs from human's understanding of ethics. Finally we discuss the implications of digital me metaethics, normative and applied ethics, the implementation of the Golden Rule in digital me-s, and we suggest two sets of normative principles for digital me: consequentialist and duty based digital me principles.
Providing explanations is considered an imperative ability for an AI agent in a human-robot teaming framework. The right explanation provides the rationale behind an AI agent's decision-making. However, to maintain the human teammate's cognitive demand to comprehend the provided explanations, prior works have focused on providing explanations in a specific order or intertwining the explanation generation with plan execution. Moreover, these approaches do not consider the degree of details required to share throughout the provided explanations. In this work, we argue that the agent-generated explanations, especially the complex ones, should be abstracted to be aligned with the level of details the human teammate desires to maintain the recipient's cognitive load. Therefore, learning a hierarchical explanations model is a challenging task. Moreover, the agent needs to follow a consistent high-level policy to transfer the learned teammate preferences to a new scenario while lower-level detailed plans are different. Our evaluation confirmed the process of understanding an explanation, especially a complex and detailed explanation, is hierarchical. The human preference that reflected this aspect corresponded exactly to creating and employing abstraction for knowledge assimilation hidden deeper in our cognitive process. We showed that hierarchical explanations achieved better task performance and behavior interpretability while reduced cognitive load. These results shed light on designing explainable agents utilizing reinforcement learning and planning across various domains.
The purpose of this paper is to discuss the possibilities for computing machinery, or AI agents, to know and to possess knowledge. This is done mainly from a virtue epistemology perspective and definition of knowledge. However, this inquiry also shed light on the human condition, what it means for a human to know, and to possess knowledge. The paper argues that it is possible for an AI agent to know and examines this from both current state-of-the-art in artificial intelligence as well as from the perspective of what the future AI development might bring in terms of superintelligent AI agents.
StarCraft, one of the most difficult esport games with long-standing history of professional tournaments, has attracted generations of players and fans, and also, intense attentions in artificial intelligence research. Recently, Google's DeepMind announced AlphaStar, a grandmaster level AI in StarCraft II that can play with humans using comparable action space and operations. In this paper, we introduce a new AI agent, named TStarBot-X, that is trained under orders of less computations and can play competitively with expert human players. TStarBot-X takes advantage of important techniques introduced in AlphaStar, and also benefits from substantial innovations including new league training methods, novel multi-agent roles, rule-guided policy search, stabilized policy improvement, lightweight neural network architecture, and importance sampling in imitation learning, etc. We show that with orders of less computation scale, a faithful reimplementation of AlphaStar's methods can not succeed and the proposed techniques are necessary to ensure TStarBot-X's competitive performance. We reveal all technical details that are complementary to those mentioned in AlphaStar, showing the most sensitive parts in league training, reinforcement learning and imitation learning that affect the performance of the agents. Most importantly, this is an open-sourced study that all codes and resources (including the trained model parameters) are publicly accessible via \url{https://github.com/tencent-ailab/tleague_projpage}. We expect this study could be beneficial for both academic and industrial future research in solving complex problems like StarCraft, and also, might provide a sparring partner for all StarCraft II players and other AI agents.
MOBA games, e.g., Honor of Kings, League of Legends, and Dota 2, pose grand challenges to AI systems such as multi-agent, enormous state-action space, complex action control, etc. Developing AI for playing MOBA games has raised much attention accordingly. However, existing work falls short in handling the raw game complexity caused by the explosion of agent combinations, i.e., lineups, when expanding the hero pool in case that OpenAI's Dota AI limits the play to a pool of only 17 heroes. As a result, full MOBA games without restrictions are far from being mastered by any existing AI system. In this paper, we propose a MOBA AI learning paradigm that methodologically enables playing full MOBA games with deep reinforcement learning. Specifically, we develop a combination of novel and existing learning techniques, including curriculum self-play learning, policy distillation, off-policy adaption, multi-head value estimation, and Monte-Carlo tree-search, in training and playing a large pool of heroes, meanwhile addressing the scalability issue skillfully. Tested on Honor of Kings, a popular MOBA game, we show how to build superhuman AI agents that can defeat top esports players. The superiority of our AI is demonstrated by the first large-scale performance test of MOBA AI agent in the literature.
Conversational AI agents are becoming ubiquitous and provide assistance to us in our everyday activities. In recent years, researchers have explored the migration of these agents across different embodiments in order to maintain the continuity of the task and improve user experience. In this paper, we investigate user's affective responses in different configurations of the migration parameters. We present a 2x2 between-subjects study in a task-based scenario using information migration and identity migration as parameters. We outline the affect processing pipeline from the video footage collected during the study and report user's responses in each condition. Our results show that users reported highest joy and were most surprised when both the information and identity was migrated; and reported most anger when the information was migrated without the identity of their agent.