Spoken English proficiency is a powerful driver of economic mobility for low-income Indian youth, yet opportunities for spoken practice remain scarce in schools. We investigate the deployment of a voice-based chatbot for English conversation practice across four low-resource schools in Delhi. Through a six-day field study combining observations and interviews, we captured the perspectives of students, teachers, and principals. Findings confirm high demand across all groups, with notable gains in student speaking confidence. Our multi-stakeholder analysis surfaced a tension in long-term adoption vision: students favored open-ended conversational practice, while administrators emphasized curriculum-aligned assessment. We offer design recommendations for voice-enabled chatbots in low-resource multilingual contexts, highlighting the need for more intelligible speech output for non-native learners, one-tap interactions with simplified interfaces, and actionable analytics for educators. Beyond language learning, our findings inform the co-design of future AI-based educational technologies that are socially sustainable within the complex ecosystem of low-resource schools.

To learn the complex norms and behaviors of society, social robots will need human teachers. But teachers must trust their learners so they will continue teaching them. The present study experimentally assigned different teaching methods (instruction, evaluation, or free choice between them) to human teachers of virtual robots. Human trust formation (and recovery from initial trust loss) was robust over these methods as long as robots markedly improved over the course of their training. Teaching methods elicited different initial expectations in teachers, but in the end, robots’ improving performance made all teachers converge at high levels of trust.

Building and maintaining trust is critically important for continued human-robot teaching and the prospect of robots learning social skills from natural environments. Whereas previous work often explored strategies to reduce system errors, mitigate trust loss, or enhance learning by interactive teaching, few studies have investigated the possible benefits of fully engaged, interactive teaching on human trust. Motivated by a pair of discrepant previous investigations, the present studies for the first time directly tested the causal impact of interactivity on the loss and recovery of trust in a human-robot social skills training context. Building on a previously developed experimental paradigm, we randomly assigned participants to one of two modes of interaction: interactive teacher vs. supervisor of an experimentally controlled virtual robot. The robot was engaged in learning norm-appropriate behavior in a healthcare setting and improved from mistake-prone to near-flawless performance. Participants indicated their changing trust during the 15-trial training session and how much they attributed the robot's improvement to their own training contributions. Interactive teachers were more resilient to initial trust loss, showed increased trust in the robot's performance on additional tasks, and attributed more of the robot's improvement to themselves than did supervisors, even when the robots were slow learners.

Human-robot trust research often measures people's trust in robots in individual scenarios. However, humans may update their trust dynamically as they continuously interact with a robot. In a well-powered study (n = 220), we investigate the trust updating process across a 15-trial interaction. In a novel paradigm, participants act in the role of teacher to a simulated robot on a smartphone-based platform, and we assess trust at multiple levels (momentary trust feelings, perceptions of trustworthiness, and intended reliance). Results reveal that people are highly sensitive to the robot's learning progress trial by trial: they take into account both previous-task performance, current-task difficulty, and cumulative learning across training. More integrative perceptions of robot trustworthiness steadily grow as people gather more evidence from observing robot performance, especially of faster-learning robots. Intended reliance on the robot in novel tasks increased only for faster-learning robots.

Robots deployed in social communities must act according to the communities’ social and moral norms. To acquire the large number of nuanced norms, robots can rely on human teaching. While humans tend to naturally use more than one teaching method when training a novice, current human-in-the-loop teaching frameworks have typically relied on single teaching methods (e.g., instruction or reward). To gain insight into how humans would teach robots to master social and moral norms, we present a novel paradigm in which participants interactively teach a simulated robot to behave appropriately in a healthcare setting, choosing to either instruct the robot or evaluate its proposed actions. We demonstrate that 89.5% of human teachers naturally use both teaching methods. Importantly, they adapt their teaching method as they observe the robot’s task performance, responding dynamically to the task’s difficulty, the robot’s most recent action, and the accumulated evidence of the robot’s learning progress.

Norms are indispensable for human communities, and so they will be for robot-human communities. We analyze some of the requirements for a robot to represent norms and conform its actions to them. These requirements include both cognitive and social properties that human norms instantiate. We examine which of these properties can be implemented in a robot’s architecture and review some previous computational approaches. We then introduce an approach using behavior trees, argue for its promise to implement properties of norms, and discuss unsolved challenges.

Robots deployed in social settings must act appropriately-that is, in compliance with social and moral norms. However, efforts of teaching norms to robots have typically relied on single teaching methods (e.g., instruction, reward). By contrast, humans may naturally use more than one teaching method when training a novice. To test this claim in the domain of human-robot teaching, we present a novel paradigm in which participants interactively teach a simulated robot to behave appropriately in a healthcare setting, choosing to either instruct the robot or evaluate its proposed actions. We demonstrate that 89% of human teachers naturally adopt mixed teaching strategies. We further identify some of the factors that influence people's choices. Results reveal that human teachers dynamically update their impression of the robot from early to late in the teaching session, and they choose their teaching strategy based on the robot's specific actions and their accumulated perceptions of the robot's learning progress.

As robots become more prevalent in society, they will need to learn to act appropriately under diverse human teaching styles. We present a human-centered approach for teaching robots reward functions by using a mixture of teaching strategies when communicating action appropriateness and goal success. Our method incorporates two teaching strategies for learning: explicit action instruction and evaluative, scalar-based feedback. We demonstrate that a robot instantiating our method can learn from humans who use both kinds of strategies to train the robot in a complex navigation task that includes norm-like constraints.

People are increasingly interacting with artificial agents in social settings, and as these agents become more sophisticated, people will have to teach them social norms. Two prominent teaching methods include instructing the learner how to act, and giving evaluative feedback on the learner’s actions. Our empirical findings indicate that people naturally adopt both methods when teaching norms to a simulated robot, and they use the methods selectively as a function of the robot’s perceived expertise and learning progress. In our algorithmic work, we conceptualize a set of context-specific norms as a reward function and integrate learning from the two teaching methods under a single likelihood-based algorithm, which estimates a reward function that induces policies maximally likely to satisfy the teacher’s intended norms. We compare robot learning under various teacher models and demonstrate that a robot responsive to both teaching methods can learn to reach its goal and minimize norm violations in a navigation task for a grid world. We improve the robot’s learning speed and performance by enabling teachers to give feedback at an abstract level (which rooms are acceptable to navigate) rather than at a low level (how to navigate any particular room).

Human behavior is powerfully guided by social and moral norms. Robots that enter human societies must therefore behave in norm-conforming ways as well. However, there is currently no cognitive, let alone computational model available of how humans represent, activate, and learn norms. We offer first steps toward such a model and apply it to the design of a norm-competent social robot. We propose a general methodology for such a design, from empirical identification of relevant norms to computational implementations of norm learning to thorough and iterative evaluation of the robot's norm compliance by means of community feedback.