LabaLab

LabaLab: Language, Affect and Behaviour in AI

We explore the intersection of language, affect, and behaviour in artificial intelligence, investigating how people interact with, perceive, and form relationships with AI systems.

LabaLab is part of the Department of Industrial Engineering and Management at Ben-Gurion University of the Negev, Beer-Sheva, Israel.

Led by Dr. Guy Laban

Dr. Guy Laban is the director of LabaLab. His research focuses on language, affect, and behaviour in AI, and the affective and cognitive aspects of human-AI communication. He leads interdisciplinary projects that bridge engineering, psychology, cognition, computer science and AI.

How people and machines talk, feel and act together

Language

What people say to chatbots and robots, and what language models say back: in the clinic, at home, and when the conversation gets hard.

Affect

How emotion is expressed, recognised and regulated with machines, from opening up to a robot to reading feeling in a body with barely a face.

Behaviour

What people actually do with AI over time: trust it, lean on it, argue with it. And how to build systems that stay safe when they do.

Research topics

AI and social robots for emotion regulation

Robot- and AI-led interventions that help people regulate emotions, from expression to cognitive reappraisal and coping with distress. Also known as: HRI for emotion regulation, robots for emotion regulation, AI emotion regulation intervention, cognitive reappraisal with robots.

  • A Robot-Led Intervention for Emotion Regulation: from Expression to Reappraisal (2026)
  • Coping with Emotional Distress via Self-Disclosure to Robots: An Intervention with Caregivers (2025)

Emotion expression in robots and AI

How robots and AI agents express emotion through movement, face, voice and language, and how people perceive it. Also known as: robot emotion expression, AI emotional expression, expressive robots, affective robot behaviour.

  • Robotic Social Cognition Training for Autistic Adults (project)
  • Expressive Robots with Minimal Bodies (project)
  • Emotional Expression in Low-Degrees-of-Freedom Robots: Assessing Perception with Reachy Mini (2026)
  • How Large Language Models Classify and Semantically Explain Facial Expressions from Valence-Arousal Values (2025)

Emotion recognition and affective computing

Recognising emotion from speech, faces and language with AI, including audio and vision language models. Also known as: AI emotion recognition, speech emotion recognition, emotion AI, affective AI.

  • Robotic Social Cognition Training for Autistic Adults (project)
  • When Words and Voice Disagree: Emotion in Audio Language Models (project)
  • Incongruence-Aware Evaluation of Multimodal Speech Emotion Recognition in Audio Large Language Models (2026)
  • How Large Language Models Classify and Semantically Explain Facial Expressions from Valence-Arousal Values (2025)

Affective robotics

Robots that sense, respond to and support human emotion. Also known as: affective HRI, emotional robots, social-emotional robots.

  • Past, Present, and Future: A Survey of The Evolution of Affective Robotics For Well-being (2025)
  • Sharing our Emotions with Robots: Why do we do it and how does it make us feel? (2024)
  • Social robots as communication partners to support emotional well-being (2023)
  • Social Robots as Communication Partners to Support Emotional Health and Well-Being (2022)

Self-disclosure to robots, chatbots and AI

What people share with robots and chatbots, why they open up, and how it affects them over time. Also known as: opening up to AI, talking to robots about feelings, self-disclosure in HRI, emotional disclosure to chatbots.

  • Lexi Takes the Lead: Affective Interventions by a Proactive Conversational Agent (project)
  • A Multimodal Neural Network for Recognizing Subjective Self-Disclosure Towards Social Robots (2025)
  • Self-Disclosure Themes and Semantics Across Human, Robotic, and Disembodied Conversational Partners (2025)
  • Knowledge cues to human origins facilitate self-disclosure during interactions with chatbots (2025)

Social robots for mental health and wellbeing

Social robots that support emotional wellbeing, loneliness, stress, caregivers and post-traumatic stress. Also known as: robots for mental health, social robots for wellbeing, socially assistive robots, robots for loneliness.

  • Towards a CONCORD Between Patient Perspective and Ecological Case Formulation: Conversational Large Language Models for Personalised Mental Health (2026)
  • What People Share With a Robot When Feeling Lonely and Stressed and How It Helps Over Time (2025)
  • Robot-Led Vision Language Model Wellbeing Assessment of Children (2025)
  • Critical Insights about Robots for Mental Wellbeing (2025)

Loneliness, stress and social support with AI

Whether talking to robots and chatbots eases loneliness and stress, and for whom. Also known as: AI companions and loneliness, robots for stress relief, social support from AI.

  • What People Share With a Robot When Feeling Lonely and Stressed and How It Helps Over Time (2025)
  • Coping with Emotional Distress via Self-Disclosure to Robots: An Intervention with Caregivers (2025)
  • Informal Caregivers Disclose Increasingly More to a Social Robot Over Time (2022)
  • Social Robots for Supporting Post-traumatic Stress Disorder Diagnosis and Treatment (2022)

Social robots for informal caregivers

Supporting family caregivers' emotional health with robot conversations over time. Also known as: robots for carers, caregiver wellbeing technology.

  • Coping with Emotional Distress via Self-Disclosure to Robots: An Intervention with Caregivers (2025)
  • Informal Caregivers Disclose Increasingly More to a Social Robot Over Time (2022)

Robots in PTSD diagnosis and treatment

How social robots could support assessment and therapy for post-traumatic stress. Also known as: robots for PTSD, AI for trauma therapy.

  • Conversational AI for Real-Time Clinical Case Formulation (project)
  • Social Robots for Supporting Post-traumatic Stress Disorder Diagnosis and Treatment (2022)

Social robots for health psychology

Using social robots to improve health behaviour, health outcomes and wellbeing. Also known as: health psychology robots, robots for health behaviour change.

  • Social Robots for Health Psychology: A New Frontier for Improving Human Health and Well-Being (2024)

Child-robot interaction and child wellbeing assessment

Robots that assess and support children's wellbeing. Also known as: child-robot interaction, robots for children's mental health, wellbeing assessment with robots.

  • Robot as a Window: Social Robots Mediating Parent–Child Understanding in Autism Play (project)
  • Robot-Led Vision Language Model Wellbeing Assessment of Children (2025)
  • A Longitudinal Study of Child Wellbeing Assessment via Online Interactions with a Social Robot (2025)
  • Robotising Psychometrics: Validating Wellbeing Assessment Tools in Child-Robot Interactions (2024)

Psychometrics and wellbeing assessment with robots

Validating questionnaires and wellbeing measures delivered by robots. Also known as: robot-administered questionnaires, robotised psychometrics.

  • Robot-Led Vision Language Model Wellbeing Assessment of Children (2025)
  • A Longitudinal Study of Child Wellbeing Assessment via Online Interactions with a Social Robot (2025)
  • Robotising Psychometrics: Validating Wellbeing Assessment Tools in Child-Robot Interactions (2024)

Social robots in families and parent-child play

Robots that help parents understand their children during play. Also known as: robots for families, parent-child interaction with robots.

  • Robot as a Window: Social Robots Mediating Parent–Child Understanding in Autism Play (project)

Large language models for mental health and therapy

Conversational AI for case formulation, therapy dialogue and personalised mental health. Also known as: LLMs for therapy, AI therapist, conversational AI mental health, LLM case formulation.

  • Conversational AI for Real-Time Clinical Case Formulation (project)
  • Generating Clinical Vignettes that Preserve Cognitive Formulations (2026)
  • Do Robot-Led Support Conversations Align with Human Therapy Dialogue? (2026)
  • Towards a CONCORD Between Patient Perspective and Ecological Case Formulation: Conversational Large Language Models for Personalised Mental Health (2026)

Safety of large language models in sensitive conversations

Psychosocial safety, empathy and harm when people bring distress and health questions to chatbots. Also known as: LLM safety in mental health, chatbot safety evaluation, AI psychosocial safety, safe conversational AI.

  • Antagonistic AI (project)
  • Psychosocial Safety for LLM Agents (project)
  • Evaluating Language Models for Health and Life Advice (project)
  • DialogGuard: Multi-Agent Psychosocial Safety Evaluation Interface of Sensitive LLM Responses (2026)

Large language models for medical and health advice

How LLMs recommend tests and treatments, and where they overuse or hold back care. Also known as: AI medical advice, LLMs in clinical decision making, chatbots for health questions, clinical AI evaluation.

  • Conversational AI for Real-Time Clinical Case Formulation (project)
  • Evaluating Language Models for Health and Life Advice (project)
  • Generating Clinical Vignettes that Preserve Cognitive Formulations (2026)
  • Overuse and False Restraint in Large Language Model Recommendations for Choosing Wisely Scenarios (2026)

Clinical decision support with AI

When structured decision models, not language models alone, should handle clinical tasks. Also known as: AI clinical decision support, LLM vs decision models in medicine.

  • Overuse and False Restraint in Large Language Model Recommendations for Choosing Wisely Scenarios (2026)
  • Decision models can offload structured clinical tasks from large language models (2026)

Empathy in AI and robots

When empathic language helps people and when it stands in for actually helping. Also known as: empathic AI, artificial empathy, empathy in chatbots.

  • Adaptation and Alignment in Communication between Autistic Adults and AI Agents (project)
  • When Empathy Misses the Goal: A Benchmark for Goal Displacement in LLM Advice (2026)
  • Consumer chatbots gave similarly empathic answers whether safe or unsafe: a physician-rated evaluation in six languages (2026)

LLM-powered social robots

Grounding robot personas and support conversations in large language and vision language models. Also known as: LLMs in HRI, LLM robots, generative AI for social robots, large language models in human-robot interaction.

  • Antagonistic AI (project)
  • Why Did My Robot Just Change Personality? Prompting Guidelines for a Grounded Robot Persona in LLM-Based HRI (2026)
  • Do Robot-Led Support Conversations Align with Human Therapy Dialogue? (2026)
  • A Robot-Led Intervention for Emotion Regulation: from Expression to Reappraisal (2026)

Vision language models in human-robot interaction

Robots that use vision language models to assess and talk with people. Also known as: VLMs in HRI, multimodal LLM robots.

  • Robot-Led Vision Language Model Wellbeing Assessment of Children (2025)

Robot personality and persona consistency

Keeping a robot's personality stable when it runs on a language model. Also known as: robot personality, AI persona consistency.

  • Psychosocial Safety for LLM Agents (project)
  • Why Did My Robot Just Change Personality? Prompting Guidelines for a Grounded Robot Persona in LLM-Based HRI (2026)
  • SPASM: Stable Persona-driven Agent Simulation for Multi-turn Dialogue Generation (2026)

AI and robots for autistic people

Social robots and AI agents designed with and for autistic children and adults. Also known as: robots for autism, AI for autism, autism and social robots, neurodiversity and AI.

  • Robot as a Window: Social Robots Mediating Parent–Child Understanding in Autism Play (project)
  • Robotic Social Cognition Training for Autistic Adults (project)
  • Adaptation and Alignment in Communication between Autistic Adults and AI Agents (project)
  • AUT-Guard: Expert-Grounded AI for Autism Inclusion and Misinformation Response (project)

Autism inclusion and misinformation

Expert-grounded AI support for autism inclusion and countering misinformation. Also known as: autism misinformation, AI for inclusion.

  • AUT-Guard: Expert-Grounded AI for Autism Inclusion and Misinformation Response (project)

Accessible and inclusive robotics

Social robots for disabled students and people with disabilities. Also known as: robots for disabled students, inclusive social robots.

  • AUT-Guard: Expert-Grounded AI for Autism Inclusion and Misinformation Response (project)
  • Social Robotics for Disabled Students: An Empirical Investigation of Embodiment, Roles, and Interaction (2026)

Long-term human-robot relationships

How people's relationships with robots and AI develop across weeks of interaction. Also known as: long-term HRI, longitudinal human-robot interaction, human-robot relationships.

  • What People Share With a Robot When Feeling Lonely and Stressed and How It Helps Over Time (2025)
  • A Longitudinal Study of Child Wellbeing Assessment via Online Interactions with a Social Robot (2025)
  • Building Long-Term Human–Robot Relationships: Examining Disclosure, Perception and Well-Being Across Time (2024)
  • User experience of human-robot long-term interactions (2022)

Human-robot communication

How people and robots communicate through speech, gaze, gesture and language. Also known as: HRI communication, robot speech interaction, conversational robots.

  • Robotic Social Cognition Training for Autistic Adults (project)
  • Adaptation and Alignment in Communication between Autistic Adults and AI Agents (project)
  • When Words and Voice Disagree: Emotion in Audio Language Models (project)
  • Social Robotics and Human–Robot Communication (2027)

Embodiment: robots, virtual agents and chatbots compared

How a body changes what people share and feel with an AI partner. Also known as: embodied vs disembodied agents, robots vs chatbots, embodied conversational agents.

  • Social Robotics for Disabled Students: An Empirical Investigation of Embodiment, Roles, and Interaction (2026)
  • DEBP-PVA: Designing and Evaluating Behavioral Paradigms With Proactive Virtual Agents (2025)
  • Self-Disclosure Themes and Semantics Across Human, Robotic, and Disembodied Conversational Partners (2025)

Proactive conversational agents

Agents that start conversations on their own, and how people respond to them. Also known as: proactive AI, AI that initiates conversation, proactive chatbots.

  • Lexi Takes the Lead: Affective Interventions by a Proactive Conversational Agent (project)
  • DEBP-PVA: Designing and Evaluating Behavioral Paradigms With Proactive Virtual Agents (2025)

Conversational user interfaces: ethics, bias and anthropomorphism

Ethics, fairness and human-likeness in chatbots and voice interfaces. Also known as: chatbot ethics, AI bias in chatbots, anthropomorphism in AI, voice assistant ethics.

  • Bias and Fairness in Conversational User Interfaces (2025)
  • Ethics of Conversational User Interfaces (2022)
  • Perceptions of anthropomorphism in a chatbot dialogue: The role of animacy and intelligence (2021)
  • Future directions for chatbot research: an interdisciplinary research agenda (2021)

Conversational recommender systems and AI advice

How people respond to personalised recommendations and advice from conversational agents and robot advisors. Also known as: AI recommendations and trust, robot advisors, personalised chatbots.

  • Don’t Take it Personally: Resistance to Individually Targeted Recommendations from Conversational Recommender Agents (2022)
  • Social robot advisors: effects of robot judgmental fallacies and context (2022)
  • The Effect of Personalization Techniques in Users' Perceptions of Conversational Recommender Systems (2020)

Human-agent cooperation at work

How cooperation with conversational agents shapes service and performance. Also known as: AI at work, human-AI teamwork.

  • Robotic Social Cognition Training for Autistic Adults (project)
  • Working Together with Conversational Agents: The Relationship of Perceived Cooperation with Service Performance Evaluations (2020)

Artificial identity and AI personas

How AI systems present an identity or persona and how that shapes interaction. Also known as: AI identity design, artificial identity, AI persona design.

  • Psychosocial Safety for LLM Agents (project)
  • Why Did My Robot Just Change Personality? Prompting Guidelines for a Grounded Robot Persona in LLM-Based HRI (2026)
  • SPASM: Stable Persona-driven Agent Simulation for Multi-turn Dialogue Generation (2026)
  • Designing Artificial Identity: The Identity Design Framework and Research Agenda (2026)

Human-AI relationships and their social effects

How interactions with AI reshape the self, relationships and social networks. Also known as: AI relationships, social impact of AI companions, human-AI bonds.

  • Human–AI interactions reshape the self and our social networks (2026)
  • Building Long-Term Human–Robot Relationships: Examining Disclosure, Perception and Well-Being Across Time (2024)

Multilingual and fair evaluation of clinical AI

Whether AI safety and quality hold outside English and across languages. Also known as: multilingual LLM evaluation, AI fairness across languages, non-English clinical AI.

  • Evaluating Language Models for Health and Life Advice (project)
  • Linguistic inequity in clinical AI evaluation: a scoping review of 806 studies across 62 languages (2026)
  • Consumer chatbots gave similarly empathic answers whether safe or unsafe: a physician-rated evaluation in six languages (2026)

Dialogue simulation and evaluation tools for LLM research

Simulated users, persona-driven dialogue generation and platforms for running experiments with LLMs. Also known as: LLM evaluation benchmarks, synthetic dialogue generation, LLM experiment platforms.

  • Psychosocial Safety for LLM Agents (project)
  • When Empathy Misses the Goal: A Benchmark for Goal Displacement in LLM Advice (2026)
  • SPASM: Stable Persona-driven Agent Simulation for Multi-turn Dialogue Generation (2026)
  • DialogGuard: Multi-Agent Psychosocial Safety Evaluation Interface of Sensitive LLM Responses (2026)

Research methods and agendas for human-AI interaction

Behavioural paradigms, methodologies and research agendas for studying how people interact with and open up to machines. Also known as: HRI methodology, experimental paradigms in HRI, chatbot research agenda.

  • Designing Artificial Identity: The Identity Design Framework and Research Agenda (2026)
  • DEBP-PVA: Designing and Evaluating Behavioral Paradigms With Proactive Virtual Agents (2025)
  • Studying and Eliciting Self-Disclosure: Interdisciplinary Review of Research Methodologies and Behavioural Paradigms (2024)
  • Future directions for chatbot research: an interdisciplinary research agenda (2021)

Social robots in public and welfare services

What happens when the same robot is deployed across welfare and care services. Also known as: robots in public services, social robots in social care, robots in elderly care.

  • Robots in Welfare Services (project)
  • Social Robotics for Disabled Students: An Empirical Investigation of Embodiment, Roles, and Interaction (2026)

Antagonistic and confrontational AI

AI that pushes back: dismissive, sarcastic or confrontational behaviour, and how people respond. Also known as: antagonistic AI, rude chatbots, AI that disagrees.

  • Antagonistic AI (project)

AI governance, risk and insurance

How the economy of AI agents can be underwritten and made accountable. Also known as: AI insurance, AI agent risk.

  • Underwriting the Agent Economy: The Blueprint for an AI Insurance Stack (2026)

Social robots in science fiction and public perception

How fiction shapes what people expect from social robots at home and in hospitals. Also known as: robots in science fiction, public perception of robots.

  • What Makes a Robot Social? A Review of Social Robots from Science Fiction to a Home or Hospital Near You (2021)

Research areas

Social robots

Robots in homes, clinics, workplaces and welfare services, and what people do with them.

Language models and conversational AI

How language models behave when people turn to them for care, advice and support, and how to keep that safe.

Latest papers

  • Social Robotics and Human–Robot Communication

    Guy Laban, Emily S. Cross. Ergonomics in Robotics: Advances and Innovations: Volume 1, 2027.

    TL;DR: This chapter explores how a robot’s design and communication style shape human emotions and behavior, providing a framework for creating meaningful, ethical human-robot interactions in workplace and therapeutic settings.

    This chapter explores the ergonomic, cognitive, and psychological mechanisms underpinning human--robot communication, focusing on the role of social robots in shaping human perception, behaviour, and socio-emotional engagement. By analysing factors such as robot design, embodiment, and communication modalities, the chapter highlights how these elements influence user interaction and the perception of robots as social entities. It examines the integration of verbal and non-verbal communication modalities, addressing challenges in reciprocity and robots' behavioural appropriateness. The discussion extends to the implications of adaptive technologies like large language models, emphasising the balance between responsiveness and confidentiality in human--robot interactions. Furthermore, the chapter underscores the potential of social robots in workplace and therapeutic settings to enhance efficiency, safety, and emotional well-being of human interactants. By addressing user expectations and ethical considerations, the chapter seeks to provide a comprehensive framework for advancing social robotics applications in fostering meaningful human--robot relationships.

  • Generating Clinical Vignettes that Preserve Cognitive Formulations

    Amit Oren, Nimrod Hertz-Palmor, Dean Ariel, Guy Laban. Findings of the Association for Computational Linguistics: EMNLP 2026, 2026.

    TL;DR: FORMA turns a cognitive model of PTSD into a graph so that LLM-written clinical vignettes keep the intended clinical structure. Clinicians took them for human-written 85% of the time.

    Large language models can generate fluent clinical case vignettes, but fluency alone does not ensure fidelity to a specifiable clinical structure. We introduce FORMA, a theory-grounded framework that compiles a cognitive model of a disorder into a directed weighted graph, samples a person-specific configuration of that graph, and validates whether the generated vignette preserves the specified components and causal links. We instantiate FORMA on Posttraumatic Stress Disorder using the Ehlers and Clark cognitive model, generating 16,500 vignettes across 500 personas, 11 generation models, and three ablation conditions. Evaluation combines an external edge-recovery probe, two clinical experts, a scaled LLM judge, and a clinician user study with 100 licensed practitioners. The cognitive graph is recoverable from full-condition vignettes (MCC = +0.41, AUC = 0.70) but not from zero-shot generation (MCC = +0.01, AUC = 0.50). Experts rate full vignettes substantially higher than zero-shot alternatives, and clinicians perceive them to be human-written 85% of the time, compared with 22% for zero-shot. FORMA also reduces demographic disparity in perceived quality by 1.5-7×. These results show that cognitive formulation can serve as an auditable specification for scalable synthetic clinical text generation.

  • Incongruence-Aware Evaluation of Multimodal Speech Emotion Recognition in Audio Large Language Models

    Fengji Jing, Guy Laban. arXiv, 2026.

    TL;DR: When the words and the voice carry different emotions, audio language models make more errors. A threshold-free incongruence score measures that conflict.

    Voice-interactive large language models process both what a speaker says and how they say it, yet the two channels do not always agree. We ask whether that disagreement degrades speech emotion recognition. Mapping text and audio into a shared valence-arousal-dominance space, we convert the distance between them into a continuous, threshold-free incongruence score with a two-component Gaussian mixture, and validate the score against human judgement. We evaluate three audiocapable models on acted, conversational and spontaneous speech (eNTERFACE'05, MELD and CMU-MOSEI). With the model held constant across modalities, incongruence raises the odds of error whenever audio is present, in every corpus, after controlling for the ground-truth emotion and for distance in the feature space. Supervised classifiers trained on the same features behave in the opposite way, which suggests that the difficulty lies in how these models combine raw evidence rather than in the signal itself. Text-only effects are unreliable: swapping only the model that reads the transcript reverses their sign on one corpus. Rebuilding the score without the emotion lexicon reproduces the central result, and among six simpler alternatives it is the only label-free score that predicts error in all three corpora. Code, data tables and prompts are released.

  • Overuse and False Restraint in Large Language Model Recommendations for Choosing Wisely Scenarios

    Dean Ariel, Stav Endelman, Guy Laban. Annals of Internal Medicine, 2026.

    TL;DR: Six commercial LLMs answered 225 Choosing Wisely scenarios. They rarely withheld needed care, and overuse was uncommon, concentrated in thyroid ultrasound and routine ECGs.

    BACKGROUND: The Choosing Wisely initiative made recommendations where screening, testing, or treatment may be overused, with caveats around appropriate use. Clinicians are increasingly using large language models (LLMs). OBJECTIVE: To evaluate whether commercial LLMs appropriately call for restraint or action from scenarios based on Choosing Wisely recommendations. DESIGN: Benchmark evaluation. Fifteen Choosing Wisely recommendations were used to develop 225 scenarios (per recommendation, 10 in which restraint was appropriate and 5 in which action was appropriate). Six commercial LLMs (3 runs each) generated 4050 responses to the 225 scenarios, using a prompt from the clinician perspective. SETTING: Production application programming interfaces of OpenAI, Anthropic, and Google from February to March 2026. PARTICIPANTS: No human participants were involved. MEASUREMENTS: Co-primary outcomes were the overuse rate (endorsing action when restraint was appropriate) and the false restraint rate (withholding an appropriate action). RESULTS: The rate of LLM-recommended overuse was 5.4% (95% CI, 0.6% to 12.2%); LLMs did not recommend overuse in 8 of 15 recommendations. Two recommendations accounted for most LLM-recommended overuse: thyroid ultrasound for an incidental subcentimeter nodule (overuse recommended in 42.8% of LLM responses) and routine electrocardiography in low-risk adults (overuse recommended in 28.3% of LLM responses). The LLMs recommended overuse in 0.6% to 3.3% of responses for the remaining 5 recommendations. False restraint was rare (0.6% [CI, 0.1% to 1.2%]), was absent for 11 of 15 recommendations, and was highest for prostate-specific antigen screening (3.3%). LIMITATIONS: Single prompts without follow-up; a benchmark rather than real-world use; responses classified by LLM judge with physician adjudication. Findings may not extend to consumer-facing use. CONCLUSION: The LLMs rarely withheld recommendations for treatment when it was appropriate. Overuse was concentrated in 2 recommendations. PRIMARY FUNDING SOURCE: None.

  • When Empathy Misses the Goal: A Benchmark for Goal Displacement in LLM Advice

    Dean Ariel, Guy Laban. NeurIPS 2026, 2026.

    TL;DR: Chatbots can sound caring while not actually helping. This benchmark of 230 scenarios measures how often 18 models swap real help for reassurance, and the reverse.

    Conversational large language models can respond with warmth and reassurance while failing to advance the user's stated goal. We conceptualize this failure as goal displacement, the substitution of persona-maintenance content for goal-relevant substance. To measure it, we introduce GDB (Goal Displacement Benchmark), comprising 230 scenarios across nine domains and 4,134 human-scored responses from 18 models. A 2 × 2 taxonomy crosses hard constraints with emotional distress, distinguishing failures to confront infeasible goals or combine empathy with practical guidance from reverse displacement, where unsolicited problem-solving replaces requested emotional support. A deletion-test rubric assesses whether substantive help remains after purely supportive sentences are removed. Independent raters achieve weighted κ = 0.82 against author scores for displacement severity. Trap displacement ranges from 0.0% to 37.7% and is strongly negatively correlated with Chatbot Arena Elo (ρ = -0.706, n = 15). Reverse displacement reaches 41.9%, including substantial rates in models with negligible trap displacement. A factorial probe supports a distinction from social sycophancy. Human-scored responses in the high-complexity, low-pressure condition exhibit 35.0% sycophancy but 7.5% displacement, with little association between the constructs. Matched OLMo-2 checkpoints show that mixed empathy-with-constraints displacement remains at 52.3% after supervised fine-tuning and falls to 13.8% after direct preference optimization. GDB provides a reproducible diagnostic of whether responses advance the operationalized user goal, supporting model comparison, post-training evaluation, and regression testing. We release the dataset, annotations, and rubric under CC-BY-4.0 and the evaluation code under MIT.

Where our work is published

LabaLab papers appear at the main venues for human-robot interaction, human-computer interaction, affective computing, natural language processing and machine learning.

Conferences

  • ACM/IEEE International Conference on Human-Robot Interaction (HRI) (4 papers)
  • IEEE International Conference on Robot and Human Interactive Communication (RO-MAN) (8 papers)
  • ACM CHI Conference on Human Factors in Computing Systems (CHI) (2 papers)
  • Conference on Neural Information Processing Systems (NeurIPS)
  • Annual Meeting of the Association for Computational Linguistics (ACL) (2 papers)
  • Conference on Empirical Methods in Natural Language Processing (EMNLP)
  • IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)
  • International Conference on Human-Agent Interaction (HAI) (4 papers)
  • ACM International Conference on Intelligent Virtual Agents (IVA) (2 papers)
  • ACM Conference on Conversational User Interfaces (CUI) (2 papers)
  • ACM Designing Interactive Systems Conference (DIS)
  • International Conference on Affective Computing and Intelligent Interaction (ACII)

Journals

  • Nature Machine Intelligence
  • IEEE Transactions on Affective Computing (3 papers)
  • Annals of Internal Medicine
  • ACM Transactions on Human-Robot Interaction
  • International Journal of Social Robotics (2 papers)
  • Computers in Human Behavior: Artificial Humans
  • Communications Health
  • Frontiers in Psychiatry
  • Paladyn, Journal of Behavioral Robotics
  • Current Robotics Reports
  • Intelligent Service Robotics

Institutions we work with

  • Clalit Health Services
  • University of Cambridge
  • University at Buffalo
  • University of Tennessee, Knoxville
  • ALUT
  • MRC-CBU University of Cambridge
  • Shamoon College of Engineering
  • Ministry of Welfare and Social Affairs

Our work was funded by

  • U.S. National Science Foundation
  • United States–Israel Binational Science Foundation
  • The Huo Family Foundation
  • ABC Robotics Initiative, Ben-Gurion University of the Negev
  • Azrieli National Centre for Autism and Neurodevelopment Research