Cynthia Breazeal

Yubin Kim, Huili Chen, Sharifa Alghowinem et al. · mit

Affect understanding capability is essential for social robots to autonomously interact with a group of users in an intuitive and reciprocal way. However, the challenge of multi-person affect understanding comes from not only the accurate perception of each user's affective state (e.g., engagement) but also the recognition of the affect interplay between the members (e.g., joint engagement) that presents as complex, but subtle, nonverbal exchanges between them. Here we present a novel hybrid framework for identifying a parent-child dyad's joint engagement by combining a deep learning framework with various video augmentation techniques. Using a dataset of parent-child dyads reading storybooks together with a social robot at home, we first train RGB frame- and skeleton-based joint engagement recognition models with four video augmentation techniques (General Aug, DeepFake, CutOut, and Mixed) applied datasets to improve joint engagement classification performance. Second, we demonstrate experimental results on the use of trained models in the robot-parent-child interaction context. Third, we introduce a behavior-based metric for evaluating the learned representation of the models to investigate the model interpretability when recognizing joint engagement. This work serves as the first step toward fully unlocking the potential of end-to-end video understanding models pre-trained on large public datasets and augmented with data augmentation and visualization techniques for affect recognition in the multi-person human-robot interaction in the wild.

Kate Pearce, Sharifa Alghowinem, Cynthia Breazeal

As artificial intelligence (AI) becomes a prominent part of modern life, AI literacy is becoming important for all citizens, not just those in technology careers. Previous research in AI education materials has largely focused on the introduction of terminology as well as AI use cases and ethics, but few allow students to learn by creating their own machine learning models. Therefore, there is a need for enriching AI educational tools with more adaptable and flexible platforms for interested educators with any level of technical experience to utilize within their teaching material. As such, we propose the development of an open-source tool (Build-a-Bot) for students and teachers to not only create their own transformer-based chatbots based on their own course material, but also learn the fundamentals of AI through the model creation process. The primary concern of this paper is the creation of an interface for students to learn the principles of artificial intelligence by using a natural language pipeline to train a customized model to answer questions based on their own school curriculums. The model uses contexts given by their instructor, such as chapters of a textbook, to answer questions and is deployed on an interactive chatbot/voice agent. The pipeline teaches students data collection, data augmentation, intent recognition, and question answering by having them work through each of these processes while creating their AI agent, diverging from previous chatbot work where students and teachers use the bots as black-boxes with no abilities for customization or the bots lack AI capabilities, with the majority of dialogue scripts being rule-based. In addition, our tool is designed to make each step of this pipeline intuitive for students at a middle-school level. Further work primarily lies in providing our tool to schools and seeking student and teacher evaluations.

Dong Won Lee, Yubin Kim, Rosalind Picard et al. · mit

As we move closer to real-world AI systems, AI agents must be able to deal with multiparty (group) conversations. Recognizing and interpreting multiparty behaviors is challenging, as the system must recognize individual behavioral cues, deal with the complexity of multiple streams of data from multiple people, and recognize the subtle contingent social exchanges that take place amongst group members. To tackle this challenge, we propose the Multiparty-Transformer (Multipar-T), a transformer model for multiparty behavior modeling. The core component of our proposed approach is the Crossperson Attention, which is specifically designed to detect contingent behavior between pairs of people. We verify the effectiveness of Multipar-T on a publicly available video-based group engagement detection benchmark, where it outperforms state-of-the-art approaches in average F-1 scores by 5.2% and individual class F-1 scores by up to 10.0%. Through qualitative analysis, we show that our Crossperson Attention module is able to discover contingent behavior.

Yubin Kim, Salman Rahman, Samuel Schmidgall et al.

Scientific discovery in digital health requires converting continuous physiological signals from wearable devices into clinically actionable biomarkers. We introduce CoDaS (AI Co-Data-Scientist), a multi-agent system that structures biomarker discovery as an iterative process combining hypothesis generation, statistical analysis, adversarial validation, and literature-grounded reasoning with human oversight using large-scale wearable datasets. Across three cohorts totaling 9,279 participant-observations, CoDaS identified 41 candidate digital biomarkers for mental health and 25 for metabolic outcomes, each subjected to an internal validation battery spanning replication, stability, robustness, and discriminative power. Across two independent depression cohorts, CoDaS surfaced circadian instability-related features in both datasets, reflected in sleep duration variability (DWB, ρ= 0.252, p < 0.001) and sleep onset variability (GLOBEM, ρ= 0.126, p < 0.001). In a metabolic cohort, CoDaS derived a cardiovascular fitness index (steps/resting heart rate; ρ= -0.374, p < 0.001), and recovered established clinical associations, including the hepatic function ratio (AST/ALT; ρ= -0.375, p < 0.001), a known correlate of insulin resistance. Incorporating CoDaS-derived features alongside demographic variables led to modest but consistent improvements in predictive performance, with cross-validated ΔR^2 increases of 0.040 for depression and 0.021 for insulin resistance. These findings suggest that CoDaS enables systematic and traceable hypothesis generation and prioritization for biomarker discovery from large-scale wearable data.

Pedro Colon-Hernandez, Henry Lieberman, Yida Xin et al.

Contextualized or discourse aware commonsense inference is the task of generating coherent commonsense assertions (i.e., facts) from a given story, and a particular sentence from that story. Some problems with the task are: lack of controllability for topics of the inferred facts; lack of commonsense knowledge during training; and, possibly, hallucinated or false facts. In this work, we utilize a transformer model for this task and develop techniques to address the aforementioned problems in the task. We control the inference by introducing a new technique we call "hinting". Hinting is a kind of language model prompting, that utilizes both hard prompts (specific words) and soft prompts (virtual learnable templates). This serves as a control signal to advise the language model "what to talk about". Next, we establish a methodology for performing joint inference with multiple commonsense knowledge bases. Joint inference of commonsense requires care, because it is imprecise and the level of generality is more flexible. You want to be sure that the results "still make sense" for the context. To this end, we align the textual version of assertions from three knowledge graphs (ConceptNet, ATOMIC2020, and GLUCOSE) with a story and a target sentence. This combination allows us to train a single model to perform joint inference with multiple knowledge graphs. We show experimental results for the three knowledge graphs on joint inference. Our final contribution is exploring a GAN architecture that generates the contextualized commonsense assertions and scores them as to their plausibility through a discriminator. The result is an integrated system for contextual commonsense inference in stories, that can controllably generate plausible commonsense assertions, and takes advantage of joint inference between multiple commonsense knowledge bases.

Sigurdur Orn Adalgeirsson, Cynthia Breazeal

Partially Observable Markov Decision Processes (POMDPs) offer a promising world representation for autonomous agents, as they can model both transitional and perceptual uncertainties. Calculating the optimal solution to POMDP problems can be computationally expensive as they require reasoning over the (possibly infinite) space of beliefs. Several approaches have been proposed to overcome this difficulty, such as discretizing the belief space, point-based belief sampling, and Monte Carlo tree search. The Real-Time Dynamic Programming approach of the RTDP-Bel algorithm approximates the value function by storing it in a hashtable with discretized belief keys. We propose an extension to the RTDP-Bel algorithm which we call Belief Branch and Bound RTDP (B$^3$RTDP). Our algorithm uses a bounded value function representation and takes advantage of this in two novel ways: a search-bounding technique based on action selection convergence probabilities, and a method for leveraging early action convergence called the \textit{Convergence Frontier}. Lastly, we empirically demonstrate that B$^3$RTDP can achieve greater returns in less time than the state-of-the-art SARSOP solver on known POMDP problems.

Safinah Ali, Cynthia Breazeal

Art created using generated Artificial Intelligence has taken the world by storm and generated excitement for many digital creators and technologists. However, the reception and reaction from artists have been mixed. Concerns about plagiarizing their artworks and styles for datasets and uncertainty around the future of digital art sparked movements in artist communities shunning the use of AI for generating art and protecting artists' rights. Collaborating with these tools for novel creative use cases also sparked hope from some creators. Artists are an integral stakeholder in the rapidly evolving digital creativity industry and understanding their concerns and hopes inform responsible development and use of creativity support tools. In this work, we study artists' sentiments about AI-generated art. We interviewed 7 artists and analyzed public posts from artists on social media platforms Reddit, Twitter and Artstation. We report artists' main concerns and hopes around AI-generated artwork, informing a way forward for inclusive development of these tools.

Yubin Kim, Chanwoo Park, Hyewon Jeong et al. · mit

Foundation models are becoming valuable tools in medicine. Yet despite their promise, the best way to leverage Large Language Models (LLMs) in complex medical tasks remains an open question. We introduce a novel multi-agent framework, named Medical Decision-making Agents (MDAgents) that helps address this gap by automatically assigning a collaboration structure to a team of LLMs. The assigned solo or group collaboration structure is tailored to the medical task at hand, emulating real-world medical decision-making processes adapted to tasks of varying complexities. We evaluate our framework and baseline methods using state-of-the-art LLMs across a suite of real-world medical knowledge and medical diagnosis benchmarks, including a comparison of LLMs' medical complexity classification against human physicians. MDAgents achieved the best performance in seven out of ten benchmarks on tasks requiring an understanding of medical knowledge and multi-modal reasoning, showing a significant improvement of up to 4.2% (p < 0.05) compared to previous methods' best performances. Ablation studies reveal that MDAgents effectively determines medical complexity to optimize for efficiency and accuracy across diverse medical tasks. Notably, the combination of moderator review and external medical knowledge in group collaboration resulted in an average accuracy improvement of 11.8%. Our code can be found at https://github.com/mitmedialab/MDAgents.

Jocelyn Shen, Amina Luvsanchultem, Jessica Kim et al.

As users increasingly turn to LLMs for practical and personal advice, they become vulnerable to being subtly steered toward hidden incentives misaligned with their own interests. Prior works have benchmarked persuasion and manipulation detection, but these efforts rely on simulated or debate-style settings, remain uncorrelated with real human belief shifts, and overlook a critical dimension: the morality of hidden incentives driving the manipulation. We introduce PUPPET, a theoretical taxonomy of personalized emotional manipulation in LLM-human dialogues that centers around incentive morality, and conduct a human study with N=1,035 participants across realistic everyday queries, varying personalization and incentive direction (harmful versus prosocial). We find that harmful hidden incentives produce significantly larger belief shifts than prosocial ones. Finally, we benchmark LLMs on the task of belief prediction, finding that models exhibit moderate predictive ability of belief change based on conversational contexts (r=0.3 - 0.5), but they also systematically underestimate the magnitude of belief shift. Together, this work establishes a theoretically grounded and behaviorally validated foundation for studying, and ultimately combatting, incentive-driven manipulation in LLMs during everyday, practical user queries.

Zhiyuan Hu, Yunhai Hu, Juncheng Liu et al.

Multi-agent systems have evolved into practical LLM-driven collaborators for many applications, gaining robustness from diversity and cross-checking. However, multi-agent RL (MARL) training is resource-intensive and unstable: co-adapting teammates induce non-stationarity, and rewards are often sparse and high-variance. Therefore, we introduce \textbf{Multi-Agent Test-Time Reinforcement Learning (MATTRL)}, a framework that injects structured textual experience into multi-agent deliberation at inference time. MATTRL forms a multi-expert team of specialists for multi-turn discussions, retrieves and integrates test-time experiences, and reaches consensus for final decision-making. We also study credit assignment for constructing a turn-level experience pool, then reinjecting it into the dialogue. Across challenging benchmarks in medicine, math, and education, MATTRL improves accuracy by an average of 3.67\% over a multi-agent baseline, and by 8.67\% over comparable single-agent baselines. Ablation studies examine different credit-assignment schemes and provide a detailed comparison of how they affect training outcomes. MATTRL offers a stable, effective and efficient path to distribution-shift-robust multi-agent reasoning without tuning.

Zhiyuan Hu, Yucheng Wang, Yufei He et al.

Reinforcement learning (RL) has become a central paradigm for post-training large language models (LLMs), particularly for complex reasoning tasks, yet it often suffers from exploration collapse: policies prematurely concentrate on a small set of dominant reasoning patterns, improving pass@1 while limiting rollout-level diversity and gains in pass@k. We argue that this failure stems from regularizing local token behavior rather than diversity over sets of solutions. To address this, we propose Uniqueness-Aware Reinforcement Learning, a rollout-level objective that explicitly rewards correct solutions that exhibit rare high-level strategies. Our method uses an LLM-based judge to cluster rollouts for the same problem according to their high-level solution strategies, ignoring superficial variations, and reweights policy advantages inversely with cluster size. As a result, correct but novel strategies receive higher rewards than redundant ones. Across mathematics, physics, and medical reasoning benchmarks, our approach consistently improves pass@$k$ across large sampling budgets and increases the area under the pass@$k$ curve (AUC@$K$) without sacrificing pass@1, while sustaining exploration and uncovering more diverse solution strategies at scale.

Yubin Kim, Xuhai Xu, Daniel McDuff et al. · mit

Large language models (LLMs) are capable of many natural language tasks, yet they are far from perfect. In health applications, grounding and interpreting domain-specific and non-linguistic data is crucial. This paper investigates the capacity of LLMs to make inferences about health based on contextual information (e.g. user demographics, health knowledge) and physiological data (e.g. resting heart rate, sleep minutes). We present a comprehensive evaluation of 12 state-of-the-art LLMs with prompting and fine-tuning techniques on four public health datasets (PMData, LifeSnaps, GLOBEM and AW_FB). Our experiments cover 10 consumer health prediction tasks in mental health, activity, metabolic, and sleep assessment. Our fine-tuned model, HealthAlpaca exhibits comparable performance to much larger models (GPT-3.5, GPT-4 and Gemini-Pro), achieving the best performance in 8 out of 10 tasks. Ablation studies highlight the effectiveness of context enhancement strategies. Notably, we observe that our context enhancement can yield up to 23.8% improvement in performance. While constructing contextually rich prompts (combining user context, health knowledge and temporal information) exhibits synergistic improvement, the inclusion of health knowledge context in prompts significantly enhances overall performance.

Yubin Kim, Chanwoo Park, Taehan Kim et al.

Agent systems often decompose a task across multiple roles, but these roles are typically specified by prompts rather than enforced by access controls. Without enforcement, a team pass rate can mask whether agents actually coordinated or whether one role effectively did another role's work. We present TeamBench, a benchmark with 851 task templates and 931 seeded instances for evaluating agent coordination under operating system-enforced role separation. TeamBench separates specification access, workspace editing, and final certification across Planner, Executor, and Verifier roles, so that no role can read the full requirements, modify the workspace, and certify the final answer. Prompt-only and sandbox-enforced teams reach statistically indistinguishable pass rates, but prompt-only runs produce 3.6 times more cases where the verifier attempts to edit the executor's code. Verifiers approve 49% of submissions that fail the deterministic grader, and removing the verifier improves mean partial score in the ablation. Team value is also conditional. Teams benefit when single agents struggle, but hurt when single agents already perform well. A 40-session human study under the same role separation shows that our benchmark exposes interaction patterns that pass rate misses. Solo participants work through the task directly, human participants paired with agents often collapse into quick approval, and human teams spend more effort coordinating missing information across roles.

Yubin Kim, Hyewon Jeong, Shan Chen et al.

Hallucinations in foundation models arise from autoregressive training objectives that prioritize token-likelihood optimization over epistemic accuracy, fostering overconfidence and poorly calibrated uncertainty. We define medical hallucination as any model-generated output that is factually incorrect, logically inconsistent, or unsupported by authoritative clinical evidence in ways that could alter clinical decisions. We evaluated 11 foundation models (7 general-purpose, 4 medical-specialized) across seven medical hallucination tasks spanning medical reasoning and biomedical information retrieval. General-purpose models achieved significantly higher proportions of hallucination-free responses than medical-specialized models (median: 76.6% vs 51.3%, difference = 25.2%, 95% CI: 18.7-31.3%, Mann-Whitney U = 27.0, p = 0.012, rank-biserial r = -0.64). Top-performing models such as Gemini-2.5 Pro exceeded 97% accuracy when augmented with chain-of-thought prompting (base: 87.6%), while medical-specialized models like MedGemma ranged from 28.6-61.9% despite explicit training on medical corpora. Chain-of-thought reasoning significantly reduced hallucinations in 86.4% of tested comparisons after FDR correction (q < 0.05), demonstrating that explicit reasoning traces enable self-verification and error detection. Physician audits confirmed that 64-72% of residual hallucinations stemmed from causal or temporal reasoning failures rather than knowledge gaps. A global survey of clinicians (n = 70) validated real-world impact: 91.8% had encountered medical hallucinations, and 84.7% considered them capable of causing patient harm. The underperformance of medical-specialized models despite domain training indicates that safety emerges from sophisticated reasoning capabilities and broad knowledge integration developed during large-scale pre-training, not from narrow optimization.

Jocelyn Shen, Yubin Kim, Mohit Hulse et al. · mit

Modeling empathy is a complex endeavor that is rooted in interpersonal and experiential dimensions of human interaction, and remains an open problem within AI. Existing empathy datasets fall short in capturing the richness of empathy responses, often being confined to in-lab or acted scenarios, lacking longitudinal data, and missing self-reported labels. We introduce a new multimodal dataset for empathy during personal experience sharing: the EmpathicStories++ dataset (https://mitmedialab.github.io/empathic-stories-multimodal/) containing 53 hours of video, audio, and text data of 41 participants sharing vulnerable experiences and reading empathically resonant stories with an AI agent. EmpathicStories++ is the first longitudinal dataset on empathy, collected over a month-long deployment of social robots in participants' homes, as participants engage in natural, empathic storytelling interactions with AI agents. We then introduce a novel task of predicting individuals' empathy toward others' stories based on their personal experiences, evaluated in two contexts: participants' own personal shared story context and their reflections on stories they read. We benchmark this task using state-of-the-art models to pave the way for future improvements in contextualized and longitudinal empathy modeling. Our work provides a valuable resource for further research in developing empathetic AI systems and understanding the intricacies of human empathy within genuine, real-world settings.

Prerna Ravi, Annalisa Broski, Glenda Stump et al.

Artificial Intelligence (AI) and its associated applications are ubiquitous in today's world, making it imperative that students and their teachers understand how it works and the ramifications arising from its usage. In this study, we investigate the experiences of seven teachers following their implementation of modules from the MIT RAICA (Responsible AI for Computational Action) curriculum. Through semi-structured interviews, we investigated their instructional strategies as they engaged with the AI curriculum in their classroom, how their teaching and learning beliefs about AI evolved with the curriculum as well as how those beliefs impacted their implementation of the curriculum. Our analysis suggests that the AI modules not only expanded our teachers' knowledge in the field, but also prompted them to recognize its daily applications and their ethical and societal implications, so that they could better engage with the content they deliver to students. Teachers were able to leverage their own interdisciplinary backgrounds to creatively introduce foundational AI topics to students to maximize engagement and playful learning. Our teachers advocated their need for better external support when navigating technological resources, additional time for preparation given the novelty of the curriculum, more flexibility within curriculum timelines, and additional accommodations for students of determination. Our findings provide valuable insights for enhancing future iterations of AI literacy curricula and teacher professional development (PD) resources.

Prerna Ravi, John Masla, Gisella Kakoti et al.

The emergence of generative AI, particularly large language models (LLMs), has opened the door for student-centered and active learning methods like project-based learning (PBL). However, PBL poses practical implementation challenges for educators around project design and management, assessment, and balancing student guidance with student autonomy. The following research documents a co-design process with interdisciplinary K-12 teachers to explore and address the current PBL challenges they face. Through teacher-driven interviews, collaborative workshops, and iterative design of wireframes, we gathered evidence for ways LLMs can support teachers in implementing high-quality PBL pedagogy by automating routine tasks and enhancing personalized learning. Teachers in the study advocated for supporting their professional growth and augmenting their current roles without replacing them. They also identified affordances and challenges around classroom integration, including resource requirements and constraints, ethical concerns, and potential immediate and long-term impacts. Drawing on these, we propose design guidelines for future deployment of LLM tools in PBL.

Yubin Kim, Chanwoo Park, Hyewon Jeong et al.

Medical Decision-Making (MDM) is a multi-faceted process that requires clinicians to assess complex multi-modal patient data patient, often collaboratively. Large Language Models (LLMs) promise to streamline this process by synthesizing vast medical knowledge and multi-modal health data. However, single-agent are often ill-suited for nuanced medical contexts requiring adaptable, collaborative problem-solving. Our MDAgents addresses this need by dynamically assigning collaboration structures to LLMs based on task complexity, mimicking real-world clinical collaboration and decision-making. This framework improves diagnostic accuracy and supports adaptive responses in complex, real-world medical scenarios, making it a valuable tool for clinicians in various healthcare settings, and at the same time, being more efficient in terms of computing cost than static multi-agent decision making methods.

Yubin Kim, Zhiyuan Hu, Hyewon Jeong et al.

Large Language Models (LLMs) as clinical agents require careful behavioral adaptation. While adept at reactive tasks (e.g., diagnosis reasoning), LLMs often struggle with proactive engagement, like unprompted identification of critical missing information or risks. We introduce BehaviorBench, a comprehensive dataset to evaluate agent behaviors across a clinical assistance spectrum, ranging from reactive query responses to proactive interventions (e.g., clarifying ambiguities, flagging overlooked critical data). Our BehaviorBench experiments reveal LLMs' inconsistent proactivity. To address this, we propose BehaviorSFT, a novel training strategy using behavioral tokens to explicitly condition LLMs for dynamic behavioral selection along this spectrum. BehaviorSFT boosts performance, achieving up to 97.3% overall Macro F1 on BehaviorBench and improving proactive task scores (e.g., from 95.0% to 96.5% for Qwen2.5-7B-Ins). Crucially, blind clinician evaluations confirmed BehaviorSFT-trained agents exhibit more realistic clinical behavior, striking a superior balance between helpful proactivity (e.g., timely, relevant suggestions) and necessary restraint (e.g., avoiding over-intervention) versus standard fine-tuning or explicit instructed agents.

Yubin Kim, Taehan Kim, Wonjune Kang et al.

Vocal health plays a crucial role in peoples' lives, significantly impacting their communicative abilities and interactions. However, despite the global prevalence of voice disorders, many lack access to convenient diagnosis and treatment. This paper introduces VocalAgent, an audio large language model (LLM) to address these challenges through vocal health diagnosis. We leverage Qwen-Audio-Chat fine-tuned on three datasets collected in-situ from hospital patients, and present a multifaceted evaluation framework encompassing a safety assessment to mitigate diagnostic biases, cross-lingual performance analysis, and modality ablation studies. VocalAgent demonstrates superior accuracy on voice disorder classification compared to state-of-the-art baselines. Its LLM-based method offers a scalable solution for broader adoption of health diagnostics, while underscoring the importance of ethical and technical validation.

Dong Won Lee, Yubin Kim, Denison Guvenoz et al.

Our work aims to advance the social reasoning of embodied artificial intelligence (AI) agents in real-world social interactions. Recently, language models (LMs) and foundational models (FMs) are being utilized as automatic evaluators of human-AI interactions with the goal of eventually being used to improve the policy of the AI agent. To enable further research in this direction, we introduce a large-scale real-world Human Robot Social Interaction (HSRI) Dataset to benchmark the capabilities of LMs and FMs to identify and reason about social interactions, specifically with regard to robot social errors and competencies . Our dataset consists of 400 real-world human social robot interaction videos and over 10K annotations, detailing the robot's social errors, competencies, rationale, and corrective actions, capturing unique aspects of human-AI interaction only present in real-world interactions. To further assess AI models' ability to reason about social interactions, we propose eight new benchmark tasks for evaluating centered around whether AI models can (1) evaluate social interactions via detecting social errors and competencies, (2) identify the explanatory factors associated to errors and competencies, (3) understand the flow of real-world social interactions, and (4) provide reasons and corrective actions for social errors. Human studies and experiments with modern LMs and FMs reveal that current models struggle with these tasks, demonstrating that our dataset and benchmark provides a step forward towards socially intelligent AI.

Serena Bono, Spandan Madan, Ishaan Grover et al.

Is it better to perform tennis training in a pristine indoor environment or a noisy outdoor one? To model this problem, here we investigate whether shifts in the transition probabilities between the training and testing environments in reinforcement learning problems can lead to better performance under certain conditions. We generate new Markov Decision Processes (MDPs) starting from a given MDP, by adding quantifiable, parametric noise into the transition function. We refer to this process as Noise Injection and the resulting environments as δ-environments. This process allows us to create variations of the same environment with quantitative control over noise serving as a metric of distance between environments. Conventional wisdom suggests that training and testing on the same MDP should yield the best results. In stark contrast, we observe that agents can perform better when trained on the noise-free environment and tested on the noisy δ-environments, compared to training and testing on the same δ-environments. We confirm that this finding extends beyond noise variations: it is possible to showcase the same phenomenon in ATARI game variations including varying Ghost behaviour in PacMan, and Paddle behaviour in Pong. We demonstrate this intriguing behaviour across 60 different variations of ATARI games, including PacMan, Pong, and Breakout. We refer to this phenomenon as the Indoor-Training Effect. Code to reproduce our experiments and to implement Noise Injection can be found at https://bit.ly/3X6CTYk.

Prerna Ravi, Dong Won Lee, Beatriz Flamia et al.

Understanding how ideas develop and flow in small-group conversations is critical for analyzing collaborative learning. A key structural feature of these interactions is threading, the way discourse talk naturally organizes into interwoven topical strands that evolve over time. While threading has been widely studied in asynchronous text settings, detecting threads in synchronous spoken dialogue remains challenging due to overlapping turns and implicit cues. At the same time, large language models (LLMs) show promise for automating discourse analysis but often struggle with long-context tasks that depend on tracing these conversational links. In this paper, we investigate whether explicit thread linkages can improve LLM-based coding of relational moves in group talk. We contribute a systematic guidebook for identifying threads in synchronous multi-party transcripts and benchmark different LLM prompting strategies for automated threading. We then test how threading influences performance on downstream coding of conversational analysis frameworks, that capture core collaborative actions such as agreeing, building, and eliciting. Our results show that providing clear conversational thread information improves LLM coding performance and underscores the heavy reliance of downstream analysis on well-structured dialogue. We also discuss practical trade-offs in time and cost, emphasizing where human-AI hybrid approaches can yield the best value. Together, this work advances methods for combining LLMs and robust conversational thread structures to make sense of complex, real-time group interactions.

Chenyu Zhang, Sharifa Alghowinem, Cynthia Breazeal

While recent studies have examined the leaning impact of large language model (LLM) in educational contexts, the affective dynamics of LLM-mediated tutoring remain insufficiently understood. This work introduces the first ensemble-LLM framework for large-scale affect sensing in tutoring dialogues, advancing the conversation on responsible pathways for integrating generative AI into education by attending to learners' evolving affective states. To achieve this, we analyzed two semesters' worth of 16,986 conversational turns exchanged between PyTutor, an LLM-powered AI tutor, and 261 undergraduate learners across three U.S. institutions. To investigate learners' emotional experiences, we generate zero-shot affect annotations from three frontier LLMs (Gemini, GPT-4o, Claude), including scalar ratings of valence, arousal, and learning-helpfulness, along with free-text emotion labels. These estimates are fused through rank-weighted intra-model pooling and plurality consensus across models to produce robust emotion profiles. Our analysis shows that during interaction with the AI tutor, students typically report mildly positive affect and moderate arousal. Yet learning is not uniformly smooth: confusion and curiosity are frequent companions to problem solving, and frustration, while less common, still surfaces in ways that can derail progress. Emotional states are short-lived--positive moments last slightly longer than neutral or negative ones, but they are fragile and easily disrupted. Encouragingly, negative emotions often resolve quickly, sometimes rebounding directly into positive states. Neutral moments frequently act as turning points, more often steering students upward than downward, suggesting opportunities for tutors to intervene at precisely these junctures.

Yubin Kim, Taehan Kim, Eugene Park et al.

We present InvThink, a simple yet powerful approach that gives large language models (LLMs) the capability of inverse thinking: reasoning through failure modes before generating responses. Unlike existing safety alignment methods that optimize directly for safe response, InvThink instructs models to 1) enumerate potential harms, 2) analyze their consequences, and 3) generate safe outputs that proactively avoid these risks. Our method reveals three key findings: (i) safety improvements show stronger scaling with model size compared to existing safety methods. (ii) InvThink mitigates safety tax; by training models to systematically consider failure modes, it preserves general reasoning capabilities on standard benchmarks. (iii) beyond general safety tasks, InvThink excels in high-stakes domains including external-facing (medicine, finance, law) and agentic (blackmail, murder) risk scenarios, achieving up to 15.7% reduction in harmful responses compared to baseline methods like SafetyPrompt. We further implement InvThink via supervised fine-tuning, and reinforcement learning across three LLM families. These results suggest that inverse reasoning provides a scalable and generalizable path toward safer, more capable language models.

Yubin Kim, Hyewon Jeong, Chanwoo Park et al.

Large language models (LLMs) deployed as agents introduce significant safety risks in clinical settings due to their potential for error and single points of failure. We introduce Tiered Agentic Oversight (TAO), a hierarchical multi-agent system that enhances AI safety through layered, automated supervision. Inspired by clinical hierarchies (e.g., nurse-physician-specialist) in hospital, TAO routes tasks to specialized agents based on complexity, creating a robust safety framework through automated inter- and intra-tier communication and role-playing. Crucially, this hierarchical structure functions as an effective error-correction mechanism, absorbing up to 24% of individual agent errors before they can compound. Our experiments reveal TAO outperforms single-agent and other multi-agent systems on 4 out of 5 healthcare safety benchmarks, with up to an 8.2% improvement. Ablation studies confirm key design principles of the system: (i) its adaptive architecture is over 3% safer than static, single-tier configurations, and (ii) its lower tiers are indispensable, as their removal causes the most significant degradation in overall safety. Finally, we validated the system's synergy with human doctors in a user study where a physician, acting as the highest tier agent, provided corrective feedback that improved medical triage accuracy from 40% to 60%. Project Page: https://tiered-agentic-oversight.github.io/

Jocelyn Shen, Akhila Yerukola, Xuhui Zhou et al. · mit

Conversational breakdowns in close relationships are deeply shaped by personal histories and emotional context, yet most NLP research treats conflict detection as a general task, overlooking the relational dynamics that influence how messages are perceived. In this work, we leverage nonviolent communication (NVC) theory to evaluate LLMs in detecting conversational breakdowns and assessing how relationship backstory influences both human and model perception of conflicts. Given the sensitivity and scarcity of real-world datasets featuring conflict between familiar social partners with rich personal backstories, we contribute the PersonaConflicts Corpus, a dataset of N=5,772 naturalistic simulated dialogues spanning diverse conflict scenarios between friends, family members, and romantic partners. Through a controlled human study, we annotate a subset of dialogues and obtain fine-grained labels of communication breakdown types on individual turns, and assess the impact of backstory on human and model perception of conflict in conversation. We find that the polarity of relationship backstories significantly shifted human perception of communication breakdowns and impressions of the social partners, yet models struggle to meaningfully leverage those backstories in the detection task. Additionally, we find that models consistently overestimate how positively a message will make a listener feel. Our findings underscore the critical role of personalization to relationship contexts in enabling LLMs to serve as effective mediators in human communication for authentic connection.

Dong Won Lee, Hae Won Park, Cynthia Breazeal et al.

We propose a large language model based reward decomposition framework for aligning dialogue agents using only a single session-level feedback signal. We leverage the reasoning capabilities of a frozen, pretrained large language model (LLM) to infer fine-grained local implicit rewards by decomposing global, session-level feedback. Our first text-only variant prompts the LLM to perform reward decomposition using only the dialogue transcript. The second multimodal variant incorporates additional behavioral cues, such as pitch, gaze, and facial affect, expressed as natural language descriptions. These inferred turn-level rewards are distilled into a lightweight reward model, which we utilize for RL-based fine-tuning for dialogue generation. We evaluate both text-only and multimodal variants against state-of-the-art reward decomposition methods and demonstrate notable improvements in human evaluations of conversation quality, suggesting that LLMs are strong reward decomposers that obviate the need for manual reward shaping and granular human feedback.

Yilin Qi, Dong Won Lee, Cynthia Breazeal et al.

Cognitive Reframing, a core element of Cognitive Behavioral Therapy (CBT), helps individuals reinterpret negative experiences by finding positive meaning. Recent advances in Large Language Models (LLMs) have demonstrated improved performance through reasoning-based strategies. This inspires a promising direction of leveraging the reasoning capabilities of LLMs to improve CBT and mental reframing by simulating the process of critical thinking, potentially enabling more effective recognition, generation, and reframing of cognitive distortions. In this work, we investigate the role of various reasoning methods, including pre-trained reasoning LLMs and augmented reasoning strategies such as CoT and self-consistency in enhancing LLMs' ability to perform cognitive reframing tasks. We find that augmented reasoning methods, even when applied to "outdated" LLMs like GPT-3.5, consistently outperform state-of-the-art pretrained reasoning models on recognizing, generating and reframing unhelpful thoughts.

Dong Won Lee, Hae Won Park, Yoon Kim et al. · mit

We describe an approach for aligning an LLM-based dialogue agent based on global (i.e., dialogue-level) rewards, while also taking into account naturally-occurring multimodal signals. At a high level, our approach (dubbed GELI) learns a local, turn-level reward model by decomposing the human-provided Global Explicit (GE) session-level reward, using Local Implicit (LI) multimodal reward signals to crossmodally shape the reward decomposition step. This decomposed reward model is then used as part of the standard RHLF pipeline improve an LLM-based dialog agent. We run quantitative and qualitative human studies to evaluate the performance of our GELI approach, and find that it shows consistent improvements across various conversational metrics compared to baseline methods.

Safinah Ali, Vishesh Kumar, Cynthia Breazeal

An essential element of K-12 AI literacy is educating learners about the ethical and societal implications of AI systems. Previous work in AI ethics literacy have developed curriculum and classroom activities that engage learners in reflecting on the ethical implications of AI systems and developing responsible AI. There is little work in using game-based learning methods in AI literacy. Games are known to be compelling media to teach children about complex STEM concepts. In this work, we developed a competitive card game for middle and high school students called "AI Audit" where they play as AI start-up founders building novel AI-powered technology. Players can challenge other players with potential harms of their technology or defend their own businesses by features that mitigate these harms. The game mechanics reward systems that are ethically developed or that take steps to mitigate potential harms. In this paper, we present the game design, teacher resources for classroom deployment and early playtesting results. We discuss our reflections about using games as teaching tools for AI literacy in K-12 classrooms.

Jocelyn Shen, Maarten Sap, Pedro Colon-Hernandez et al.

The most meaningful connections between people are often fostered through expression of shared vulnerability and emotional experiences in personal narratives. We introduce a new task of identifying similarity in personal stories based on empathic resonance, i.e., the extent to which two people empathize with each others' experiences, as opposed to raw semantic or lexical similarity, as has predominantly been studied in NLP. Using insights from social psychology, we craft a framework that operationalizes empathic similarity in terms of three key features of stories: main events, emotional trajectories, and overall morals or takeaways. We create EmpathicStories, a dataset of 1,500 personal stories annotated with our empathic similarity features, and 2,000 pairs of stories annotated with empathic similarity scores. Using our dataset, we fine-tune a model to compute empathic similarity of story pairs, and show that this outperforms semantic similarity models on automated correlation and retrieval metrics. Through a user study with 150 participants, we also assess the effect our model has on retrieving stories that users empathize with, compared to naive semantic similarity-based retrieval, and find that participants empathized significantly more with stories retrieved by our model. Our work has strong implications for the use of empathy-aware models to foster human connection and empathy between people.

Yubin Kim, Dong Won Lee, Paul Pu Liang et al.

Accurately modeling affect dynamics, which refers to the changes and fluctuations in emotions and affective displays during human conversations, is crucial for understanding human interactions. By analyzing affect dynamics, we can gain insights into how people communicate, respond to different situations, and form relationships. However, modeling affect dynamics is challenging due to contextual factors, such as the complex and nuanced nature of interpersonal relationships, the situation, and other factors that influence affective displays. To address this challenge, we propose a Cross-person Memory Transformer (CPM-T) framework which is able to explicitly model affective dynamics (intrapersonal and interpersonal influences) by identifying verbal and non-verbal cues, and with a large language model to utilize the pre-trained knowledge and perform verbal reasoning. The CPM-T framework maintains memory modules to store and update the contexts within the conversation window, enabling the model to capture dependencies between earlier and later parts of a conversation. Additionally, our framework employs cross-modal attention to effectively align information from multi-modalities and leverage cross-person attention to align behaviors in multi-party interactions. We evaluate the effectiveness and generalizability of our approach on three publicly available datasets for joint engagement, rapport, and human beliefs prediction tasks. Remarkably, the CPM-T framework outperforms baseline models in average F1-scores by up to 7.3%, 9.3%, and 2.0% respectively. Finally, we demonstrate the importance of each component in the framework via ablation studies with respect to multimodal temporal behavior.

Safinah Ali, Daniella DiPaola, Randi Williams et al.

Generative AI tools introduce new and accessible forms of media creation for youth. They also raise ethical concerns about the generation of fake media, data protection, privacy and ownership of AI-generated art. Since generative AI is already being used in products used by youth, it is critical that they understand how these tools work and how they can be used or misused. In this work, we facilitated students' generative AI learning through expression of their imagined future identities. We designed a learning workshop - Dreaming with AI - where students learned about the inner workings of generative AI tools, used text-to-image generation algorithms to create their imaged future dreams, reflected on the potential benefits and harms of generative AI tools and voiced their opinions about policies for the use of these tools in classrooms. In this paper, we present the learning activities and experiences of 34 high school students who engaged in our workshops. Students reached creative learning objectives by using prompt engineering to create their future dreams, gained technical knowledge by learning the abilities, limitations, text-visual mappings and applications of generative AI, and identified most potential societal benefits and harms of generative AI.

Hang Jiang, Xiajie Zhang, Xubo Cao et al.

Despite the many use cases for large language models (LLMs) in creating personalized chatbots, there has been limited research on evaluating the extent to which the behaviors of personalized LLMs accurately and consistently reflect specific personality traits. We consider studying the behavior of LLM-based agents which we refer to as LLM personas and present a case study with GPT-3.5 and GPT-4 to investigate whether LLMs can generate content that aligns with their assigned personality profiles. To this end, we simulate distinct LLM personas based on the Big Five personality model, have them complete the 44-item Big Five Inventory (BFI) personality test and a story writing task, and then assess their essays with automatic and human evaluations. Results show that LLM personas' self-reported BFI scores are consistent with their designated personality types, with large effect sizes observed across five traits. Additionally, LLM personas' writings have emerging representative linguistic patterns for personality traits when compared with a human writing corpus. Furthermore, human evaluation shows that humans can perceive some personality traits with an accuracy of up to 80%. Interestingly, the accuracy drops significantly when the annotators were informed of AI authorship.

Zhuoyue Lyu, Safinah Ali, Cynthia Breazeal

Generative Artificial Intelligence (AI) models are a compelling way to introduce K-12 students to AI education using an artistic medium, and hence have drawn attention from K-12 AI educators. Previous Creative AI curricula mainly focus on Generative Adversarial Networks (GANs) while paying less attention to Autoregressive Models, Variational Autoencoders (VAEs), or other generative models, which have since become common in the field of generative AI. VAEs' latent-space structure and interpolation ability could effectively ground the interdisciplinary learning of AI, creative arts, and philosophy. Thus, we designed a lesson to teach high school students about VAEs. We developed a web-based game and used Plato's cave, a philosophical metaphor, to introduce how VAEs work. We used a Google Colab notebook for students to re-train VAEs with their hand-written digits to consolidate their understandings. Finally, we guided the exploration of creative VAE tools such as SketchRNN and MusicVAE to draw the connection between what they learned and real-world applications. This paper describes the lesson design and shares insights from the pilot studies with 22 students. We found that our approach was effective in teaching students about a novel AI concept.

Matthew Huggins, Anastasia K. Ostrowski, Andrew Rapo et al.

Human-robot interaction can be regarded as a flow between users and robots. Designing good interaction flows takes a lot of effort and needs to be field tested. Unfortunately, the interaction flow design process is often very disjointed, with users experiencing prototypes, designers forming those prototypes, and developers implementing them as independent processes. In this paper, we present the Interaction Flow Editor (IFE), a new human-robot interaction prototyping tool that enables everyday users to create and modify their own interactions, while still providing a full suite of features that is powerful enough for developers and designers to create complex interactions. We also discuss the Flow Engine, a flexible and adaptable framework for executing robot interaction flows authors through the IFE. Finally, we present our case study results that demonstrates how older adults, aged 70 and above, can design and iterate interactions in real-time on a robot using the IFE.

Pedro Colon-Hernandez, Yida Xin, Henry Lieberman et al.

Retrofitting is a technique used to move word vectors closer together or further apart in their space to reflect their relationships in a Knowledge Base (KB). However, retrofitting only works on concepts that are present in that KB. RetroGAN uses a pair of Generative Adversarial Networks (GANs) to learn a one-to-one mapping between concepts and their retrofitted counterparts. It applies that mapping (post-specializes) to handle concepts that do not appear in the original KB in a manner similar to how some natural language systems handle out-of-vocabulary entries. We test our system on three word-similarity benchmarks and a downstream sentence simplification task and achieve the state of the art (CARD-660). Altogether, our results demonstrate our system's effectiveness for out-of-knowledge and rare word generalization.

Huili Chen, Cynthia Breazeal

In planning for future human space exploration, it is important to consider how to design for uplifting interpersonal communications and social dynamics among crew members. What if embodied social robots could help to improve the overall team interaction experience in space? On Earth, social robots have been shown effective in providing companionship, relieving stress and anxiety, fostering connection among people, enhancing team performance, and mediating conflicts in human groups. In this paper, we introduce a set of novel research questions exploring social human-robot interactions in long-duration space exploration missions.

Safinah Ali, Nisha Devasia, Cynthia Breazeal

Digital tools have long been used for supporting children's creativity. Digital games that allow children to create artifacts and express themselves in a playful environment serve as efficient Creativity Support Tools (or CSTs). Creativity is also scaffolded by social interactions with others in their environment. In our work, we explore the use of game-based interactions with a social agent to scaffold children's creative expression as game players. We designed three collaborative games and play-tested with 146 5-10 year old children played with the social robot Jibo, which affords three different kinds of creativity: verbal creativity, figural creativity and divergent thinking during creative problem solving. In this paper, we reflect on game mechanic practices that we incorporated to design for stimulating creativity in children. These strategies may be valuable to game designers and HCI researchers designing games and social agents for supporting children's creativity.

Pedro Colon-Hernandez, Catherine Havasi, Jason Alonso et al.

In recent years, transformer-based language models have achieved state of the art performance in various NLP benchmarks. These models are able to extract mostly distributional information with some semantics from unstructured text, however it has proven challenging to integrate structured information, such as knowledge graphs into these models. We examine a variety of approaches to integrate structured knowledge into current language models and determine challenges, and possible opportunities to leverage both structured and unstructured information sources. From our survey, we find that there are still opportunities at exploiting adapter-based injections and that it may be possible to further combine various of the explored approaches into one system.

Ravi Tejwani, Boris Katz, Cynthia Breazeal

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.

Ravi Tejwani, Boris Katz, Cynthia Breazeal

The migration of conversational AI agents across different embodiments in order to maintain the continuity of the task has been recently explored to further improve user experience. However, these migratable agents lack contextual understanding of the user information and the migrated device during the dialog conversations with the user. This opens the question of how an agent might behave when migrated into an embodiment for contextually predicting the next utterance. We collected a dataset from the dialog conversations between crowdsourced workers with the migration context involving personal and non-personal utterances in different settings (public or private) of embodiment into which the agent migrated. We trained the generative and information retrieval models on the dataset using with and without migration context and report the results of both qualitative metrics and human evaluation. We believe that the migration dataset would be useful for training future migratable AI systems.

Sooyeon Jeong, Sharifa Alghowinem, Laura Aymerich-Franch et al.

A significant number of college students suffer from mental health issues that impact their physical, social, and occupational outcomes. Various scalable technologies have been proposed in order to mitigate the negative impact of mental health disorders. However, the evaluation for these technologies, if done at all, often reports mixed results on improving users' mental health. We need to better understand the factors that align a user's attributes and needs with technology-based interventions for positive outcomes. In psychotherapy theory, therapeutic alliance and rapport between a therapist and a client is regarded as the basis for therapeutic success. In prior works, social robots have shown the potential to build rapport and a working alliance with users in various settings. In this work, we explore the use of a social robot coach to deliver positive psychology interventions to college students living in on-campus dormitories. We recruited 35 college students to participate in our study and deployed a social robot coach in their room. The robot delivered daily positive psychology sessions among other useful skills like delivering the weather forecast, scheduling reminders, etc. We found a statistically significant improvement in participants' psychological wellbeing, mood, and readiness to change behavior for improved wellbeing after they completed the study. Furthermore, students' personality traits were found to have a significant association with intervention efficacy. Analysis of the post-study interview revealed students' appreciation of the robot's companionship and their concerns for privacy.

Huili Chen, Yue Zhang, Felix Weninger et al.

Automatic speech-based affect recognition of individuals in dyadic conversation is a challenging task, in part because of its heavy reliance on manual pre-processing. Traditional approaches frequently require hand-crafted speech features and segmentation of speaker turns. In this work, we design end-to-end deep learning methods to recognize each person's affective expression in an audio stream with two speakers, automatically discovering features and time regions relevant to the target speaker's affect. We integrate a local attention mechanism into the end-to-end architecture and compare the performance of three attention implementations -- one mean pooling and two weighted pooling methods. Our results show that the proposed weighted-pooling attention solutions are able to learn to focus on the regions containing target speaker's affective information and successfully extract the individual's valence and arousal intensity. Here we introduce and use a "dyadic affect in multimodal interaction - parent to child" (DAMI-P2C) dataset collected in a study of 34 families, where a parent and a child (3-7 years old) engage in reading storybooks together. In contrast to existing public datasets for affect recognition, each instance for both speakers in the DAMI-P2C dataset is annotated for the perceived affect by three labelers. To encourage more research on the challenging task of multi-speaker affect sensing, we make the annotated DAMI-P2C dataset publicly available, including acoustic features of the dyads' raw audios, affect annotations, and a diverse set of developmental, social, and demographic profiles of each dyad.

Ravi Tejwani, Felipe Moreno, Sooyeon Jeong et al.

Conversational AI agents are proliferating, embodying a range of devices such as smart speakers, smart displays, robots, cars, and more. We can envision a future where a personal conversational agent could migrate across different form factors and environments to always accompany and assist its user to support a far more continuous, personalized, and collaborative experience. This opens the question of what properties of a conversational AI agent migrates across forms, and how it would impact user perception. To explore this, we developed a Migratable AI system where a user's information and/or the agent's identity can be preserved as it migrates across form factors to help its user with a task. We designed a 2x2 between-subjects study to explore the effects of information migration and identity migration on user perceptions of trust, competence, likeability, and social presence. Our results suggest that identity migration had a positive effect on trust, competence, and social presence, while information migration had a positive effect on trust, competence, and likeability. Overall, users report the highest trust, competence, likeability, and social presence towards the conversational agent when both identity and information were migrated across embodiments.

Nick DePalma, Cynthia Breazeal

This work proposes a biologically inspired approach that focuses on attention systems that are able to inhibit or constrain what is relevant at any one moment. We propose a radically new approach to making progress in human-robot joint attention called "the joint guided search task". Visual guided search is the activity of the eye as it saccades from position to position recognizing objects in each fixation location until the target object is found. Our research focuses on the exchange of nonverbal behavior toward changing the fixation location while also performing object recognition. Our main goal is a very ambitious goal of sharing attention through probing synthetic foreground maps (i.e. what is being considered by the robotic agent) and the biological attention system of the human.

Nick DePalma, Cynthia Breazeal

Robotic architectures that incorporate cloud-based resources are just now gaining popularity. However, researchers have very few investigations into their capabilities to support claims of their feasibility. We propose a novel method to exchange quality for speed of response. Further, we back this assertion with empirical findings from experiments performed with Amazon Mechanical Turk and find that our method improves quality in exchange for response time in our cognitive architecture.