In this interview series, we're meeting some of the AAAI/SIGAI Doctoral Consortium participants to find out more about their research. Kate Candon is a PhD student at Yale University interested in understanding how we can create interactive agents that are more effectively able to help people. We spoke to Kate to find out more about how she is leveraging explicit and implicit feedback in human-robot interactions. Specifically I'm interested in how we can get robots to better learn from humans in the way that they naturally teach. Typically, a lot of work in robot learning is with a human teacher who is only tasked with giving explicit feedback to the robot, but they're not necessarily engaged in the task.

Research on human-AI collaboration often prioritizes objective performance. However, understanding human subjective preferences is essential to improving human-AI complementarity and human experiences. We investigate human preferences for controllability in a shared workspace task with AI partners using Behavior Shaping (BS), a reinforcement learning algorithm that allows humans explicit control over AI behavior. In one experiment, we validate the robustness of BS in producing effective AI policies relative to self-play policies, when controls are hidden. In another experiment, we enable human control, showing that participants perceive AI partners as more effective and enjoyable when they can directly dictate AI behavior. Our findings highlight the need to design AI that prioritizes both task performance and subjective human preferences. By aligning AI behavior with human preferences, we demonstrate how human-AI complementarity can extend beyond objective outcomes to include subjective preferences.

The development of largely human-annotated benchmarks has driven the success of deep neural networks in various NLP tasks. To enhance the effectiveness of existing benchmarks, collecting new additional input-output pairs is often too costly and challenging, particularly considering their marginal impact on improving the current model accuracy. Instead, additional or complementary annotations on the existing input texts in the benchmarks can be preferable as an efficient way to pay the additional human cost. In this paper, we investigate task-specific preferences between pairs of input texts as a new alternative way for such auxiliary data annotation. From 'pair-wise' comparisons with respect to the task, the auxiliary preference learning enables the model to learn an additional informative training signal that cannot be captured with 'instance-wise' task labels. To this end, we propose a novel multi-task learning framework, called prefer-to-classify (P2C), which can enjoy the cooperative effect of learning both the given classification task and the auxiliary preferences. Here, we provide three different ways to collect preference signals in practice: (a) implicitly extracting from annotation records (for free, but often unavailable), (b) collecting explicitly from crowd workers (high paid), or (c) pre-trained large language models such as GPT-3 (low paid). Given existing classification NLP benchmarks, we demonstrate that the proposed auxiliary preference learning via P2C on them is effective in improving text classifiers. Our codes are publicly available.

Existing multimodal task-oriented dialog data fails to demonstrate the diverse expressions of user subjective preferences and recommendation acts in the real-life shopping scenario. This paper introduces a new dataset SURE (Multimodal Recommendation Dialog with SUbjective PREference), which contains 12K shopping dialogs in complex store scenes. The data is built in two phases with human annotations to ensure quality and diversity. SURE is well-annotated with subjective preferences and recommendation acts proposed by sales experts. A comprehensive analysis is given to reveal the distinguishing features of SURE. Three benchmark tasks are then proposed on the data to evaluate the capability of multimodal recommendation agents. Based on the SURE, we propose a baseline model, powered by a state-of-the-art multimodal model, for these tasks.

Interaction and cooperation with humans are overarching aspirations of artificial intelligence (AI) research. Recent studies demonstrate that AI agents trained with deep reinforcement learning are capable of collaborating with humans. These studies primarily evaluate human compatibility through "objective" metrics such as task performance, obscuring potential variation in the levels of trust and subjective preference that different agents garner. To better understand the factors shaping subjective preferences in human-agent cooperation, we train deep reinforcement learning agents in Coins, a two-player social dilemma. We recruit participants for a human-agent cooperation study and measure their impressions of the agents they encounter. Participants' perceptions of warmth and competence predict their stated preferences for different agents, above and beyond objective performance metrics. Drawing inspiration from social science and biology research, we subsequently implement a new "partner choice" framework to elicit revealed preferences: after playing an episode with an agent, participants are asked whether they would like to play the next round with the same agent or to play alone. As with stated preferences, social perception better predicts participants' revealed preferences than does objective performance. Given these results, we recommend human-agent interaction researchers routinely incorporate the measurement of social perception and subjective preferences into their studies.

One reason for the widespread interest in AI is that it has the potential to reduce the degree of bias underpinning human decisions. For example, meta-analytic studies have long highlighted the pervasive nature of bias in hiring and recruitment. Even in the rich and liberal world, there are many biases at play in the workplace, which account for the unmeritocratic or unfair advantage that some groups have over others, irrespective of their actual talent or potential: sexism, racism, and ageism, to name just a few. But one of the most prominent biases is hardly ever discussed or acknowledged, namely the beauty bias -- also known as "lookism." Indeed, the existence of a beauty premium in the labor market is well-documented.

The more AI agents are deployed in scenarios with possibly unexpected situations, the more they need to be flexible, adaptive, and creative in achieving the goal we have given them. Thus, a certain level of freedom to choose the best path to the goal is inherent in making AI robust and flexible enough. At the same time, however, the pervasive deployment of AI in our life, whether AI is autonomous or collaborating with humans, raises several ethical challenges. AI agents should be aware and follow appropriate ethical principles and should thus exhibit properties such as fairness or other virtues. These ethical principles should define the boundaries of AI's freedom and creativity. However, it is still a challenge to understand how to specify and reason with ethical boundaries in AI agents and how to combine them appropriately with subjective preferences and goal specifications. Some initial attempts employ either a data-driven example-based approach for both, or a symbolic rule-based approach for both. We envision a modular approach where any AI technique can be used for any of these essential ingredients in decision making or decision support systems, paired with a contextual approach to define their combination and relative weight. In a world where neither humans nor AI systems work in isolation, but are tightly interconnected, e.g., the Internet of Things, we also envision a compositional approach to building ethically bounded AI, where the ethical properties of each component can be fruitfully exploited to derive those of the overall system. In this paper we define and motivate the notion of ethically-bounded AI, we describe two concrete examples, and we outline some outstanding challenges.

A paradigmatic problem in social choice theory deals with the aggregation of subjective preferences of individuals --- represented as rankings of alternatives --- into a social ranking. We are interested in settings where individuals are uncertain about their own preferences, and represent their uncertainty as distributions over rankings. Under the classic objective of minimizing the (expected) sum of Kendall tau distances between the input rankings and the output ranking, we establish that preference elicitation is surprisingly straightforward and near-optimal solutions can be obtained in polynomial time. We show, both in theory and using real data, that ignoring uncertainty altogether can lead to suboptimal outcomes.