Brent Harrison

Md Sultan Al Nahian, Spencer Frazier, Brent Harrison et al.

Value alignment is the task of creating autonomous systems whose values align with those of humans. Past work has shown that stories are a potentially rich source of information on human values; however, past work has been limited to considering values in a binary sense. In this work, we explore the use of machine learning models for the task of normative principle prediction on naturally occurring story data. To do this, we extend a dataset that has been previously used to train a binary normative classifier with annotations of moral principles. We then use this dataset to train a variety of machine learning models, evaluate these models and compare their results against humans who were asked to perform the same task. We show that while individual principles can be classified, the ambiguity of what "moral principles" represent, poses a challenge for both human participants and autonomous systems which are faced with the same task.

Maryam Shoaeinaeini, Brent Harrison

Human guidance in reinforcement learning (RL) is often impractical for large-scale applications due to high costs and time constraints. Large Language Models (LLMs) offer a promising alternative to mitigate RL sample inefficiency and potentially replace human trainers. However, applying LLMs as RL trainers is challenging due to their overconfidence and less reliable solutions in sequential tasks. We address this limitation by introducing a calibrated guidance system that uses Monte Carlo Dropout to enhance LLM advice reliability by assessing prediction variances from multiple forward passes. Additionally, we develop a novel RL policy shaping method based on dynamic model average entropy to adjust the LLM's influence on RL policies according to guidance uncertainty. This approach ensures robust RL training by relying on reliable LLM guidance. To validate our contributions, we conduct extensive experiments in a Minigrid environment with three goals in varying environment sizes. The results showcase superior model performance compared to uncalibrated LLMs, unguided RL, and calibrated LLMs with different shaping policies. Moreover, we analyze various uncertainty estimation methods, demonstrating the effectiveness of average entropy in reflecting higher uncertainty in incorrect guidance. These findings highlight the persistent overconfidence in fine-tuned LLMs and underscore the importance of effective calibration in sequential decision-making problems.

Md Sultan Al Nahian, Tasmia Tasrin, Spencer Frazier et al.

Values or principles are key elements of human society that influence people to behave and function according to an accepted standard set of social rules to maintain social order. As AI systems are becoming ubiquitous in human society, it is a major concern that they could violate these norms or values and potentially cause harm. Thus, to prevent intentional or unintentional harm, AI systems are expected to take actions that align with these principles. Training systems to exhibit this type of behavior is difficult and often requires a specialized dataset. This work presents a multi-modal dataset illustrating normative and non-normative behavior in real-life situations described through natural language and artistic images. This training set contains curated sets of images that are designed to teach young children about social principles. We argue that this is an ideal dataset to use for training socially normative agents given this fact.

Chengxi Li, Brent Harrison

In this paper, we build two automatic evaluation metrics for evaluating the association between a machine-generated caption and a ground truth stylized caption: OnlyStyle and StyleCIDEr.

Kshitija Taywade, Brent Harrison, Judy Goldsmith

Many past attempts at modeling repeated Cournot games assume that demand is stationary. This does not align with real-world scenarios in which market demands can evolve over a product's lifetime for a myriad of reasons. In this paper, we model repeated Cournot games with non-stationary demand such that firms/agents face separate instances of non-stationary multi-armed bandit problem. The set of arms/actions that an agent can choose from represents discrete production quantities; here, the action space is ordered. Agents are independent and autonomous, and cannot observe anything from the environment; they can only see their own rewards after taking an action, and only work towards maximizing these rewards. We propose a novel algorithm 'Adaptive with Weighted Exploration (AWE) $ε$-greedy' which is remotely based on the well-known $ε$-greedy approach. This algorithm detects and quantifies changes in rewards due to varying market demand and varies learning rate and exploration rate in proportion to the degree of changes in demand, thus enabling agents to better identify new optimal actions. For efficient exploration, it also deploys a mechanism for weighing actions that takes advantage of the ordered action space. We use simulations to study the emergence of various equilibria in the market. In addition, we study the scalability of our approach in terms number of total agents in the system and the size of action space. We consider both symmetric and asymmetric firms in our models. We found that using our proposed method, agents are able to swiftly change their course of action according to the changes in demand, and they also engage in collusive behavior in many simulations.

Kshitija Taywade, Brent Harrison, Adib Bagh

We investigate the use of a multi-agent multi-armed bandit (MA-MAB) setting for modeling repeated Cournot oligopoly games, where the firms acting as agents choose from the set of arms representing production quantity (a discrete value). Agents interact with separate and independent bandit problems. In this formulation, each agent makes sequential choices among arms to maximize its own reward. Agents do not have any information about the environment; they can only see their own rewards after taking an action. However, the market demand is a stationary function of total industry output, and random entry or exit from the market is not allowed. Given these assumptions, we found that an $ε$-greedy approach offers a more viable learning mechanism than other traditional MAB approaches, as it does not require any additional knowledge of the system to operate. We also propose two novel approaches that take advantage of the ordered action space: $ε$-greedy+HL and $ε$-greedy+EL. These new approaches help firms to focus on more profitable actions by eliminating less profitable choices and hence are designed to optimize the exploration. We use computer simulations to study the emergence of various equilibria in the outcomes and do the empirical analysis of joint cumulative regrets.

Chengxi Li, Brent Harrison

In this paper, we propose to build a stylish image captioning model through a Multi-style Multi modality mechanism (2M). We demonstrate that with 2M, we can build an effective stylish captioner and that multi-references produced by the model can also support explaining the model through identifying erroneous input features on faulty examples. We show how this 2M mechanism can be used to build stylish captioning models and show how these models can be utilized to provide explanations of likely errors in the models.

Md Sultan Al Nahian, Spencer Frazier, Brent Harrison et al.

As more machine learning agents interact with humans, it is increasingly a prospect that an agent trained to perform a task optimally, using only a measure of task performance as feedback, can violate societal norms for acceptable behavior or cause harm. Value alignment is a property of intelligent agents wherein they solely pursue non-harmful behaviors or human-beneficial goals. We introduce an approach to value-aligned reinforcement learning, in which we train an agent with two reward signals: a standard task performance reward, plus a normative behavior reward. The normative behavior reward is derived from a value-aligned prior model previously shown to classify text as normative or non-normative. We show how variations on a policy shaping technique can balance these two sources of reward and produce policies that are both effective and perceived as being more normative. We test our value-alignment technique on three interactive text-based worlds; each world is designed specifically to challenge agents with a task as well as provide opportunities to deviate from the task to engage in normative and/or altruistic behavior.

Tasmia Tasrin, Md Sultan Al Nahian, Habarakadage Perera et al.

Interactive reinforcement learning agents use human feedback or instruction to help them learn in complex environments. Often, this feedback comes in the form of a discrete signal that is either positive or negative. While informative, this information can be difficult to generalize on its own. In this work, we explore how natural language advice can be used to provide a richer feedback signal to a reinforcement learning agent by extending policy shaping, a well-known Interactive reinforcement learning technique. Usually policy shaping employs a human feedback policy to help an agent to learn more about how to achieve its goal. In our case, we replace this human feedback policy with policy generated based on natural language advice. We aim to inspect if the generated natural language reasoning provides support to a deep reinforcement learning agent to decide its actions successfully in any given environment. So, we design our model with three networks: first one is the experience driven, next is the advice generator and third one is the advice driven. While the experience driven reinforcement learning agent chooses its actions being influenced by the environmental reward, the advice driven neural network with generated feedback by the advice generator for any new state selects its actions to assist the reinforcement learning agent to better policy shaping.

Chengxi Li, Brent Harrison

In this paper, we build a multi-style generative model for stylish image captioning which uses multi-modality image features, ResNeXt features and text features generated by DenseCap. We propose the 3M model, a Multi-UPDOWN caption model that encodes multi-modality features and decode them to captions. We demonstrate the effectiveness of our model on generating human-like captions by examining its performance on two datasets, the PERSONALITY-CAPTIONS dataset and the FlickrStyle10K dataset. We compare against a variety of state-of-the-art baselines on various automatic NLP metrics such as BLEU, ROUGE-L, CIDEr, SPICE, etc. A qualitative study has also been done to verify our 3M model can be used for generating different stylized captions.

Kshitija Taywade, Judy Goldsmith, Brent Harrison

In the real world, people/entities usually find matches independently and autonomously, such as finding jobs, partners, roommates, etc. It is possible that this search for matches starts with no initial knowledge of the environment. We propose the use of a multi-agent reinforcement learning (MARL) paradigm for a spatially formulated decentralized two-sided matching market with independent and autonomous agents. Having autonomous agents acting independently makes our environment very dynamic and uncertain. Moreover, agents lack the knowledge of preferences of other agents and have to explore the environment and interact with other agents to discover their own preferences through noisy rewards. We think such a setting better approximates the real world and we study the usefulness of our MARL approach for it. Along with conventional stable matching case where agents have strictly ordered preferences, we check the applicability of our approach for stable matching with incomplete lists and ties. We investigate our results for stability, level of instability (for unstable results), and fairness. Our MARL approach mostly yields stable and fair outcomes.

Tasmia Tasrin, Md Sultan Al Nahian, Brent Harrison

In this work, we propose a deep neural architecture that uses an attention mechanism which utilizes region based image features, the natural language question asked, and semantic knowledge extracted from the regions of an image to produce open-ended answers for questions asked in a visual question answering (VQA) task. The combination of both region based features and region based textual information about the image bolsters a model to more accurately respond to questions and potentially do so with less required training data. We evaluate our proposed architecture on a VQA task against a strong baseline and show that our method achieves excellent results on this task.

Spencer Frazier, Md Sultan Al Nahian, Mark Riedl et al.

Value alignment is a property of an intelligent agent indicating that it can only pursue goals and activities that are beneficial to humans. Traditional approaches to value alignment use imitation learning or preference learning to infer the values of humans by observing their behavior. We introduce a complementary technique in which a value aligned prior is learned from naturally occurring stories which encode societal norms. Training data is sourced from the childrens educational comic strip, Goofus and Gallant. In this work, we train multiple machine learning models to classify natural language descriptions of situations found in the comic strip as normative or non normative by identifying if they align with the main characters behavior. We also report the models performance when transferring to two unrelated tasks with little to no additional training on the new task.

Md Sultan Al Nahian, Tasmia Tasrin, Sagar Gandhi et al.

One of the primary challenges of visual storytelling is developing techniques that can maintain the context of the story over long event sequences to generate human-like stories. In this paper, we propose a hierarchical deep learning architecture based on encoder-decoder networks to address this problem. To better help our network maintain this context while also generating long and diverse sentences, we incorporate natural language image descriptions along with the images themselves to generate each story sentence. We evaluate our system on the Visual Storytelling (VIST) dataset and show that our method outperforms state-of-the-art techniques on a suite of different automatic evaluation metrics. The empirical results from this evaluation demonstrate the necessities of different components of our proposed architecture and shows the effectiveness of the architecture for visual storytelling.

Alexander Zook, Brent Harrison, Mark O. Riedl

Games are often designed to shape player behavior in a desired way; however, it can be unclear how design decisions affect the space of behaviors in a game. Designers usually explore this space through human playtesting, which can be time-consuming and of limited effectiveness in exhausting the space of possible behaviors. In this paper, we propose the use of automated planning agents to simulate humans of varying skill levels to generate game playthroughs. Metrics can then be gathered from these playthroughs to evaluate the current game design and identify its potential flaws. We demonstrate this technique in two games: the popular word game Scrabble and a collectible card game of our own design named Cardonomicon. Using these case studies, we show how using simulated agents to model humans of varying skill levels allows us to extract metrics to describe game balance (in the case of Scrabble) and highlight potential design flaws (in the case of Cardonomicon).

Upol Ehsan, Pradyumna Tambwekar, Larry Chan et al.

Automated rationale generation is an approach for real-time explanation generation whereby a computational model learns to translate an autonomous agent's internal state and action data representations into natural language. Training on human explanation data can enable agents to learn to generate human-like explanations for their behavior. In this paper, using the context of an agent that plays Frogger, we describe (a) how to collect a corpus of explanations, (b) how to train a neural rationale generator to produce different styles of rationales, and (c) how people perceive these rationales. We conducted two user studies. The first study establishes the plausibility of each type of generated rationale and situates their user perceptions along the dimensions of confidence, humanlike-ness, adequate justification, and understandability. The second study further explores user preferences between the generated rationales with regard to confidence in the autonomous agent, communicating failure and unexpected behavior. Overall, we find alignment between the intended differences in features of the generated rationales and the perceived differences by users. Moreover, context permitting, participants preferred detailed rationales to form a stable mental model of the agent's behavior.

Pradyumna Tambwekar, Murtaza Dhuliawala, Lara J. Martin et al.

Language-modeling--based approaches to story plot generation attempt to construct a plot by sampling from a language model (LM) to predict the next character, word, or sentence to add to the story. LM techniques lack the ability to receive guidance from the user to achieve a specific goal, resulting in stories that don't have a clear sense of progression and lack coherence. We present a reward-shaping technique that analyzes a story corpus and produces intermediate rewards that are backpropagated into a pre-trained LM in order to guide the model towards a given goal. Automated evaluations show our technique can create a model that generates story plots which consistently achieve a specified goal. Human-subject studies show that the generated stories have more plausible event ordering than baseline plot generation techniques.

Zhiyu Lin, Brent Harrison, Aaron Keech et al.

We describe a method to use discrete human feedback to enhance the performance of deep learning agents in virtual three-dimensional environments by extending deep-reinforcement learning to model the confidence and consistency of human feedback. This enables deep reinforcement learning algorithms to determine the most appropriate time to listen to the human feedback, exploit the current policy model, or explore the agent's environment. Managing the trade-off between these three strategies allows DRL agents to be robust to inconsistent or intermittent human feedback. Through experimentation using a synthetic oracle, we show that our technique improves the training speed and overall performance of deep reinforcement learning in navigating three-dimensional environments using Minecraft. We further show that our technique is robust to highly innacurate human feedback and can also operate when no human feedback is given.

Brent Harrison, Upol Ehsan, Mark O. Riedl

In this work we present a technique to use natural language to help reinforcement learning generalize to unseen environments. This technique uses neural machine translation, specifically the use of encoder-decoder networks, to learn associations between natural language behavior descriptions and state-action information. We then use this learned model to guide agent exploration using a modified version of policy shaping to make it more effective at learning in unseen environments. We evaluate this technique using the popular arcade game, Frogger, under ideal and non-ideal conditions. This evaluation shows that our modified policy shaping algorithm improves over a Q-learning agent as well as a baseline version of policy shaping.

Lara J. Martin, Prithviraj Ammanabrolu, Xinyu Wang et al.

Automated story generation is the problem of automatically selecting a sequence of events, actions, or words that can be told as a story. We seek to develop a system that can generate stories by learning everything it needs to know from textual story corpora. To date, recurrent neural networks that learn language models at character, word, or sentence levels have had little success generating coherent stories. We explore the question of event representations that provide a mid-level of abstraction between words and sentences in order to retain the semantic information of the original data while minimizing event sparsity. We present a technique for preprocessing textual story data into event sequences. We then present a technique for automated story generation whereby we decompose the problem into the generation of successive events (event2event) and the generation of natural language sentences from events (event2sentence). We give empirical results comparing different event representations and their effects on event successor generation and the translation of events to natural language.

Mark O. Riedl, Brent Harrison

Autonomous systems that operate around humans will likely always rely on kill switches that stop their execution and allow them to be remote-controlled for the safety of humans or to prevent damage to the system. It is theoretically possible for an autonomous system with sufficient sensor and effector capability that learn online using reinforcement learning to discover that the kill switch deprives it of long-term reward and thus learn to disable the switch or otherwise prevent a human operator from using the switch. This is referred to as the big red button problem. We present a technique that prevents a reinforcement learning agent from learning to disable the kill switch. We introduce an interruption process in which the agent's sensors and effectors are redirected to a virtual simulation where it continues to believe it is receiving reward. We illustrate our technique in a simple grid world environment.

Upol Ehsan, Brent Harrison, Larry Chan et al.

We introduce AI rationalization, an approach for generating explanations of autonomous system behavior as if a human had performed the behavior. We describe a rationalization technique that uses neural machine translation to translate internal state-action representations of an autonomous agent into natural language. We evaluate our technique in the Frogger game environment, training an autonomous game playing agent to rationalize its action choices using natural language. A natural language training corpus is collected from human players thinking out loud as they play the game. We motivate the use of rationalization as an approach to explanation generation and show the results of two experiments evaluating the effectiveness of rationalization. Results of these evaluations show that neural machine translation is able to accurately generate rationalizations that describe agent behavior, and that rationalizations are more satisfying to humans than other alternative methods of explanation.