Evaluating Human-Language Model Interaction

Many real-world applications of language models (LMs), such as code autocomplete and writing assistance, involve human-LM interaction. However, the main LM benchmarks are non-interactive in that a system produces output without human involvement. To evaluate human-LM interaction, we develop a new framework, Human-AI Language-based Interaction Evaluation (HALIE), that expands non-interactive evaluation along three dimensions, capturing (i) the interactive process, not only the final output; (ii) the first-person subjective experience, not just a third-party assessment; and (iii) notions of preference beyond quality. We then design five tasks ranging from goal-oriented to open-ended to capture different forms of interaction. On four state-of-the-art LMs (three variants of OpenAI's GPT-3 and AI21's J1-Jumbo), we find that non-interactive performance does not always result in better human-LM interaction and that first-person and third-party metrics can diverge, suggesting the importance of examining the nuances of human-LM interaction.

On Second Thought, Let's Not Think Step by Step! Bias and Toxicity in Zero-Shot Reasoning

Generating a chain of thought (CoT) can increase large language model (LLM) performance on a wide range of tasks. Zero-shot CoT evaluations, however, have been conducted primarily on logical tasks (e.g. arithmetic, commonsense QA). In this paper, we perform a controlled evaluation of zero-shot CoT across two sensitive domains: harmful questions and stereotype benchmarks. We find that using zero-shot CoT reasoning in a prompt can significantly increase a model's likelihood to produce undesirable output. Without future advances in alignment or explicit mitigation instructions, zero-shot CoT should be avoided on tasks where models can make inferences about marginalized groups or harmful topics.

Explanations Can Reduce Overreliance on AI Systems During Decision-Making

Prior work has identified a resilient phenomenon that threatens the performance of human-AI decision-making teams: overreliance, when people agree with an AI, even when it is incorrect. Surprisingly, overreliance does not reduce when the AI produces explanations for its predictions, compared to only providing predictions. Some have argued that overreliance results from cognitive biases or uncalibrated trust, attributing overreliance to an inevitability of human cognition. By contrast, our paper argues that people strategically choose whether or not to engage with an AI explanation, demonstrating empirically that there are scenarios where AI explanations reduce overreliance. To achieve this, we formalize this strategic choice in a cost-benefit framework, where the costs and benefits of engaging with the task are weighed against the costs and benefits of relying on the AI. We manipulate the costs and benefits in a maze task, where participants collaborate with a simulated AI to find the exit of a maze. Through 5 studies (N = 731), we find that costs such as task difficulty (Study 1), explanation difficulty (Study 2, 3), and benefits such as monetary compensation (Study 4) affect overreliance. Finally, Study 5 adapts the Cognitive Effort Discounting paradigm to quantify the utility of different explanations, providing further support for our framework. Our results suggest that some of the null effects found in literature could be due in part to the explanation not sufficiently reducing the costs of verifying the AI's prediction.

ELIGN: Expectation Alignment as a Multi-Agent Intrinsic Reward

Modern multi-agent reinforcement learning frameworks rely on centralized training and reward shaping to perform well. However, centralized training and dense rewards are not readily available in the real world. Current multi-agent algorithms struggle to learn in the alternative setup of decentralized training or sparse rewards. To address these issues, we propose a self-supervised intrinsic reward ELIGN - expectation alignment - inspired by the self-organization principle in Zoology. Similar to how animals collaborate in a decentralized manner with those in their vicinity, agents trained with expectation alignment learn behaviors that match their neighbors' expectations. This allows the agents to learn collaborative behaviors without any external reward or centralized training. We demonstrate the efficacy of our approach across 6 tasks in the multi-agent particle and the complex Google Research football environments, comparing ELIGN to sparse and curiosity-based intrinsic rewards. When the number of agents increases, ELIGN scales well in all multi-agent tasks except for one where agents have different capabilities. We show that agent coordination improves through expectation alignment because agents learn to divide tasks amongst themselves, break coordination symmetries, and confuse adversaries. These results identify tasks where expectation alignment is a more useful strategy than curiosity-driven exploration for multi-agent coordination, enabling agents to do zero-shot coordination.

Visual Intelligence through Human Interaction

Over the last decade, Computer Vision, the branch of Artificial Intelligence aimed at understanding the visual world, has evolved from simply recognizing objects in images to describing pictures, answering questions about images, aiding robots maneuver around physical spaces and even generating novel visual content. As these tasks and applications have modernized, so too has the reliance on more data, either for model training or for evaluation. In this chapter, we demonstrate that novel interaction strategies can enable new forms of data collection and evaluation for Computer Vision. First, we present a crowdsourcing interface for speeding up paid data collection by an order of magnitude, feeding the data-hungry nature of modern vision models. Second, we explore a method to increase volunteer contributions using automated social interventions. Third, we develop a system to ensure human evaluation of generative vision models are reliable, affordable and grounded in psychophysics theory. We conclude with future opportunities for Human-Computer Interaction to aid Computer Vision.

Conceptual Metaphors Impact Perceptions of Human-AI Collaboration

With the emergence of conversational artificial intelligence (AI) agents, it is important to understand the mechanisms that influence users' experiences of these agents. We study a common tool in the designer's toolkit: conceptual metaphors. Metaphors can present an agent as akin to a wry teenager, a toddler, or an experienced butler. How might a choice of metaphor influence our experience of the AI agent? Sampling metaphors along the dimensions of warmth and competence---defined by psychological theories as the primary axes of variation for human social perception---we perform a study (N=260) where we manipulate the metaphor, but not the behavior, of a Wizard-of-Oz conversational agent. Following the experience, participants are surveyed about their intention to use the agent, their desire to cooperate with the agent, and the agent's usability. Contrary to the current tendency of designers to use high competence metaphors to describe AI products, we find that metaphors that signal low competence lead to better evaluations of the agent than metaphors that signal high competence. This effect persists despite both high and low competence agents featuring human-level performance and the wizards being blind to condition. A second study confirms that intention to adopt decreases rapidly as competence projected by the metaphor increases. In a third study, we assess effects of metaphor choices on potential users' desire to try out the system and find that users are drawn to systems that project higher competence and warmth. These results suggest that projecting competence may help attract new users, but those users may discard the agent unless it can quickly correct with a lower competence metaphor. We close with a retrospective analysis that finds similar patterns between metaphors and user attitudes towards past conversational agents such as Xiaoice, Replika, Woebot, Mitsuku, and Tay.

Deep Bayesian Active Learning for Multiple Correct Outputs

Typical active learning strategies are designed for tasks, such as classification, with the assumption that the output space is mutually exclusive. The assumption that these tasks always have exactly one correct answer has resulted in the creation of numerous uncertainty-based measurements, such as entropy and least confidence, which operate over a model's outputs. Unfortunately, many real-world vision tasks, like visual question answering and image captioning, have multiple correct answers, causing these measurements to overestimate uncertainty and sometimes perform worse than a random sampling baseline. In this paper, we propose a new paradigm that estimates uncertainty in the model's internal hidden space instead of the model's output space. We specifically study a manifestation of this problem for visual question answer generation (VQA), where the aim is not to classify the correct answer but to produce a natural language answer, given an image and a question. Our method overcomes the paraphrastic nature of language. It requires a semantic space that structures the model's output concepts and that enables the usage of techniques like dropout-based Bayesian uncertainty. We build a visual-semantic space that embeds paraphrases close together for any existing VQA model. We empirically show state-of-art active learning results on the task of VQA on two datasets, being 5 times more cost-efficient on Visual Genome and 3 times more cost-efficient on VQA 2.0.

Visual Relationships as Functions: Enabling Few-Shot Scene Graph Prediction

Scene graph prediction --- classifying the set of objects and predicates in a visual scene --- requires substantial training data. The long-tailed distribution of relationships can be an obstacle for such approaches, however, as they can only be trained on the small set of predicates that carry sufficient labels. We introduce the first scene graph prediction model that supports few-shot learning of predicates, enabling scene graph approaches to generalize to a set of new predicates. First, we introduce a new model of predicates as functions that operate on object features or image locations. Next, we define a scene graph model where these functions are trained as message passing protocols within a new graph convolution framework. We train the framework with a frequently occurring set of predicates and show that our approach outperforms those that use the same amount of supervision by 1.78 at recall@50 and performs on par with other scene graph models. Next, we extract object representations generated by the trained predicate functions to train few-shot predicate classifiers on rare predicates with as few as 1 labeled example. When compared to strong baselines like transfer learning from existing state-of-the-art representations, we show improved 5-shot performance by 4.16 recall@1. Finally, we show that our predicate functions generate interpretable visualizations, enabling the first interpretable scene graph model.

Scene Graph Prediction with Limited Labels

Visual knowledge bases such as Visual Genome power numerous applications in computer vision, including visual question answering and captioning, but suffer from sparse, incomplete relationships. All scene graph models to date are limited to training on a small set of visual relationships that have thousands of training labels each. Hiring human annotators is expensive, and using textual knowledge base completion methods are incompatible with visual data. In this paper, we introduce a semi-supervised method that assigns probabilistic relationship labels to a large number of unlabeled images using few labeled examples. We analyze visual relationships to suggest two types of image-agnostic features that are used to generate noisy heuristics, whose outputs are aggregated using a factor graph-based generative model. With as few as 10 labeled relationship examples, the generative model creates enough training data to train any existing state-of-the-art scene graph model. We demonstrate that our method for generating training data outperforms all baseline approaches by 5.16 recall@100. Since we only use a few labels, we define a complexity metric for relationships that serves as an indicator (R^2 = 0.778) for conditions under which our method succeeds over transfer learning, the de-facto approach for training with limited labels.