Publications
Search
1.
Wang, Yunzhe; Lucas, Gale M.; Becerik-Gerber, Burcin; Ustun, Volkan
Implicit Behavioral Alignment of Language Agents in High-Stakes Crowd Simulations Miscellaneous
2025, (arXiv:2509.16457 [cs]).
@misc{wang_implicit_2025,
title = {Implicit Behavioral Alignment of Language Agents in High-Stakes Crowd Simulations},
author = {Yunzhe Wang and Gale M. Lucas and Burcin Becerik-Gerber and Volkan Ustun},
url = {http://arxiv.org/abs/2509.16457},
doi = {10.48550/arXiv.2509.16457},
year = {2025},
date = {2025-09-01},
urldate = {2025-09-25},
publisher = {arXiv},
abstract = {Language-driven generative agents have enabled large-scale social simulations with transformative uses, from interpersonal training to aiding global policy-making. However, recent studies indicate that generative agent behaviors often deviate from expert expectations and real-world data–a phenomenon we term the Behavior-Realism Gap. To address this, we introduce a theoretical framework called Persona-Environment Behavioral Alignment (PEBA), formulated as a distribution matching problem grounded in Lewin's behavior equation stating that behavior is a function of the person and their environment. Leveraging PEBA, we propose PersonaEvolve (PEvo), an LLM-based optimization algorithm that iteratively refines agent personas, implicitly aligning their collective behaviors with realistic expert benchmarks within a specified environmental context. We validate PEvo in an active shooter incident simulation we developed, achieving an 84% average reduction in distributional divergence compared to no steering and a 34% improvement over explicit instruction baselines. Results also show PEvo-refined personas generalize to novel, related simulation scenarios. Our method greatly enhances behavioral realism and reliability in high-stakes social simulations. More broadly, the PEBA-PEvo framework provides a principled approach to developing trustworthy LLM-driven social simulations.},
note = {arXiv:2509.16457 [cs]},
keywords = {},
pubstate = {published},
tppubtype = {misc}
}
Language-driven generative agents have enabled large-scale social simulations with transformative uses, from interpersonal training to aiding global policy-making. However, recent studies indicate that generative agent behaviors often deviate from expert expectations and real-world data–a phenomenon we term the Behavior-Realism Gap. To address this, we introduce a theoretical framework called Persona-Environment Behavioral Alignment (PEBA), formulated as a distribution matching problem grounded in Lewin's behavior equation stating that behavior is a function of the person and their environment. Leveraging PEBA, we propose PersonaEvolve (PEvo), an LLM-based optimization algorithm that iteratively refines agent personas, implicitly aligning their collective behaviors with realistic expert benchmarks within a specified environmental context. We validate PEvo in an active shooter incident simulation we developed, achieving an 84% average reduction in distributional divergence compared to no steering and a 34% improvement over explicit instruction baselines. Results also show PEvo-refined personas generalize to novel, related simulation scenarios. Our method greatly enhances behavioral realism and reliability in high-stakes social simulations. More broadly, the PEBA-PEvo framework provides a principled approach to developing trustworthy LLM-driven social simulations.
2.
Hale, James; Kim, Peter H.; Gratch, Jonathan
“Provably fair” algorithms may perpetuate racial and gender bias: a study of salary dispute resolution Journal Article
In: Auton Agent Multi-Agent Syst, vol. 39, no. 1, pp. 20, 2025, ISSN: 1387-2532, 1573-7454.
@article{hale_provably_2025,
title = {“Provably fair” algorithms may perpetuate racial and gender bias: a study of salary dispute resolution},
author = {James Hale and Peter H. Kim and Jonathan Gratch},
url = {https://link.springer.com/10.1007/s10458-025-09703-x},
doi = {10.1007/s10458-025-09703-x},
issn = {1387-2532, 1573-7454},
year = {2025},
date = {2025-06-01},
urldate = {2025-03-18},
journal = {Auton Agent Multi-Agent Syst},
volume = {39},
number = {1},
pages = {20},
abstract = {Abstract
Prior work suggests automated dispute resolution tools using “provably fair” algorithms can address disparities between demographic groups. These methods use multi-criteria elicited preferences from all disputants and satisfy constraints to generate “fair” solutions. However, we analyze the potential for inequity to permeate proposals through the preference elicitation stage. This possibility arises if differences in dispositional attitudes differ between demographics, and those dispositions affect elicited preferences. Specifically, risk aversion plays a prominent role in predicting preferences. Risk aversion predicts a weaker relative preference for
salary
and a softer within-issue utility for each issue; this leads to worse compensation packages for risk-averse groups. These results raise important questions in AI-value alignment about whether an AI mediator should take explicit preferences at face value.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Abstract
Prior work suggests automated dispute resolution tools using “provably fair” algorithms can address disparities between demographic groups. These methods use multi-criteria elicited preferences from all disputants and satisfy constraints to generate “fair” solutions. However, we analyze the potential for inequity to permeate proposals through the preference elicitation stage. This possibility arises if differences in dispositional attitudes differ between demographics, and those dispositions affect elicited preferences. Specifically, risk aversion plays a prominent role in predicting preferences. Risk aversion predicts a weaker relative preference for
salary
and a softer within-issue utility for each issue; this leads to worse compensation packages for risk-averse groups. These results raise important questions in AI-value alignment about whether an AI mediator should take explicit preferences at face value.
Prior work suggests automated dispute resolution tools using “provably fair” algorithms can address disparities between demographic groups. These methods use multi-criteria elicited preferences from all disputants and satisfy constraints to generate “fair” solutions. However, we analyze the potential for inequity to permeate proposals through the preference elicitation stage. This possibility arises if differences in dispositional attitudes differ between demographics, and those dispositions affect elicited preferences. Specifically, risk aversion plays a prominent role in predicting preferences. Risk aversion predicts a weaker relative preference for
salary
and a softer within-issue utility for each issue; this leads to worse compensation packages for risk-averse groups. These results raise important questions in AI-value alignment about whether an AI mediator should take explicit preferences at face value.
3.
Klumpe, Stella; Mitchell, Kelsey C.; Cox, Emma; Katz, Jeffrey S.; Lazarowski, Lucia; Deshpande, Gopikrishna; Gratch, Jonathan; Visser, Ewart J. De; Ayaz, Hasan; Li, Xingnan; Franke, Adrian A.; Krueger, Frank
Social bonding between humans, animals, and robots: Dogs outperform AIBOs, their robotic replicas, as social companions Journal Article
In: PLoS One, vol. 20, no. 6, pp. e0324312, 2025, ISSN: 1932-6203.
@article{klumpe_social_2025,
title = {Social bonding between humans, animals, and robots: Dogs outperform AIBOs, their robotic replicas, as social companions},
author = {Stella Klumpe and Kelsey C. Mitchell and Emma Cox and Jeffrey S. Katz and Lucia Lazarowski and Gopikrishna Deshpande and Jonathan Gratch and Ewart J. De Visser and Hasan Ayaz and Xingnan Li and Adrian A. Franke and Frank Krueger},
editor = {Casey R. Lynch},
url = {https://dx.plos.org/10.1371/journal.pone.0324312},
doi = {10.1371/journal.pone.0324312},
issn = {1932-6203},
year = {2025},
date = {2025-06-01},
urldate = {2025-06-12},
journal = {PLoS One},
volume = {20},
number = {6},
pages = {e0324312},
abstract = {In the evolving landscape of technology, robots have emerged as social companions, prompting an investigation into social bonding between humans and robots. While human-animal interactions are well-studied, human-robot interactions (HRI) remain comparatively underexplored. Ethorobotics, a field of social robotic engineering based on ecology and ethology, suggests designing companion robots modeled on animal companions, which are simpler to emulate than humans. However, it is unclear whether these robots can match the social companionship provided by their original models. This study examined social bonding between humans and AIBOs, dog-inspired companion robots, compared to real dogs. Nineteen female participants engaged in 12 affiliative interactions with dogs and AIBOs across two counter-balanced, one-month bonding phases. Social bonding was assessed through urinary oxytocin (OXT) level change over an interaction, self-reported attachment using an adapted version of the Lexington Attachment to Pets Scale, and social companionship evaluations administering the Robot-Dog Questionnaire. To examine OXT level changes and self-reported attachment by comparing the two social companions, we conducted mixed-effects model analyses and planned follow-up comparisons. Frequency comparison, binary logistic regression, and thematic analysis were performed to analyze social companionship evaluations. Results revealed significant differences between dogs and AIBOs in fostering social bonds. OXT level change increased during interactions with dogs but decreased with AIBOs. Participants reported stronger attachment to dogs and rated them as better social companions. These findings highlight the current limitations of AIBOs in fostering social bonding immediately compared to dogs. Our study contributes to the growing HRI research by demonstrating an existing gap between AIBOs and dogs as social companions. It highlights the need for further investigation to understand the complexities of social bonding with companion robots, which is essential to implement successful applications for social robots in diverse domains such as the elderly and health care, education, and entertainment.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
In the evolving landscape of technology, robots have emerged as social companions, prompting an investigation into social bonding between humans and robots. While human-animal interactions are well-studied, human-robot interactions (HRI) remain comparatively underexplored. Ethorobotics, a field of social robotic engineering based on ecology and ethology, suggests designing companion robots modeled on animal companions, which are simpler to emulate than humans. However, it is unclear whether these robots can match the social companionship provided by their original models. This study examined social bonding between humans and AIBOs, dog-inspired companion robots, compared to real dogs. Nineteen female participants engaged in 12 affiliative interactions with dogs and AIBOs across two counter-balanced, one-month bonding phases. Social bonding was assessed through urinary oxytocin (OXT) level change over an interaction, self-reported attachment using an adapted version of the Lexington Attachment to Pets Scale, and social companionship evaluations administering the Robot-Dog Questionnaire. To examine OXT level changes and self-reported attachment by comparing the two social companions, we conducted mixed-effects model analyses and planned follow-up comparisons. Frequency comparison, binary logistic regression, and thematic analysis were performed to analyze social companionship evaluations. Results revealed significant differences between dogs and AIBOs in fostering social bonds. OXT level change increased during interactions with dogs but decreased with AIBOs. Participants reported stronger attachment to dogs and rated them as better social companions. These findings highlight the current limitations of AIBOs in fostering social bonding immediately compared to dogs. Our study contributes to the growing HRI research by demonstrating an existing gap between AIBOs and dogs as social companions. It highlights the need for further investigation to understand the complexities of social bonding with companion robots, which is essential to implement successful applications for social robots in diverse domains such as the elderly and health care, education, and entertainment.
4.
Kwon, Deuksin; Hae, Jiwon; Clift, Emma; Shamsoddini, Daniel; Gratch, Jonathan; Lucas, Gale M.
ASTRA: A Negotiation Agent with Adaptive and Strategic Reasoning through Action in Dynamic Offer Optimization Miscellaneous
2025, (arXiv:2503.07129 [cs]).
@misc{kwon_astra_2025,
title = {ASTRA: A Negotiation Agent with Adaptive and Strategic Reasoning through Action in Dynamic Offer Optimization},
author = {Deuksin Kwon and Jiwon Hae and Emma Clift and Daniel Shamsoddini and Jonathan Gratch and Gale M. Lucas},
url = {http://arxiv.org/abs/2503.07129},
doi = {10.48550/arXiv.2503.07129},
year = {2025},
date = {2025-03-01},
urldate = {2025-03-18},
publisher = {arXiv},
abstract = {Negotiation requires dynamically balancing self-interest and cooperation to maximize one's own utility. Yet, existing agents struggle due to bounded rationality in human data, low adaptability to counterpart behavior, and limited strategic reasoning. To address this, we introduce principle-driven negotiation agents, powered by ASTRA, a novel framework for turn-level offer optimization grounded in two core principles: opponent modeling and Tit-for-Tat reciprocity. ASTRA operates in three stages: (1) interpreting counterpart behavior, (2) optimizing counteroffers via a linear programming (LP) solver, and (3) selecting offers based on negotiation tactics and the partner's acceptance probability. Through simulations and human evaluations, our agent effectively adapts to an opponent's shifting stance and achieves favorable outcomes through enhanced adaptability and strategic reasoning. Beyond improving negotiation performance, it also serves as a powerful coaching tool, offering interpretable strategic feedback and optimal offer recommendations.},
note = {arXiv:2503.07129 [cs]},
keywords = {},
pubstate = {published},
tppubtype = {misc}
}
Negotiation requires dynamically balancing self-interest and cooperation to maximize one's own utility. Yet, existing agents struggle due to bounded rationality in human data, low adaptability to counterpart behavior, and limited strategic reasoning. To address this, we introduce principle-driven negotiation agents, powered by ASTRA, a novel framework for turn-level offer optimization grounded in two core principles: opponent modeling and Tit-for-Tat reciprocity. ASTRA operates in three stages: (1) interpreting counterpart behavior, (2) optimizing counteroffers via a linear programming (LP) solver, and (3) selecting offers based on negotiation tactics and the partner's acceptance probability. Through simulations and human evaluations, our agent effectively adapts to an opponent's shifting stance and achieves favorable outcomes through enhanced adaptability and strategic reasoning. Beyond improving negotiation performance, it also serves as a powerful coaching tool, offering interpretable strategic feedback and optimal offer recommendations.
5.
Gurney, Nikolos; Pynadath, David V.; Miller, John H.
Willingness to work as a predictor of human-agent team success Journal Article
In: Front. Comput. Sci., vol. 7, pp. 1405436, 2025, ISSN: 2624-9898.
@article{gurney_willingness_2025,
title = {Willingness to work as a predictor of human-agent team success},
author = {Nikolos Gurney and David V. Pynadath and John H. Miller},
url = {https://www.frontiersin.org/articles/10.3389/fcomp.2025.1405436/full},
doi = {10.3389/fcomp.2025.1405436},
issn = {2624-9898},
year = {2025},
date = {2025-03-01},
urldate = {2025-04-15},
journal = {Front. Comput. Sci.},
volume = {7},
pages = {1405436},
abstract = {Research shows that the effectiveness of human-agent teams depends heavily on human team members' prior experiences, whether from direct teaming activities or relevant domain knowledge. While researchers have proposed various mechanisms to explain this relationship, we present a simpler alternative explanation: experience serves primarily as an indicator of a person's fundamental willingness to engage in teaming tasks. We introduce a measure called “willingness to work” that quantifies this underlying disposition. Our empirical analysis demonstrates that this straightforward metric robustly predicts human-agent team performance. Beyond its practical value as a predictive tool, this reconceptualization of the experience-performance relationship necessitates a fresh examination of existing findings in the field. The results suggest that a team member's basic willingness to invest effort may be more fundamental to success than previously recognized mechanisms.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Research shows that the effectiveness of human-agent teams depends heavily on human team members' prior experiences, whether from direct teaming activities or relevant domain knowledge. While researchers have proposed various mechanisms to explain this relationship, we present a simpler alternative explanation: experience serves primarily as an indicator of a person's fundamental willingness to engage in teaming tasks. We introduce a measure called “willingness to work” that quantifies this underlying disposition. Our empirical analysis demonstrates that this straightforward metric robustly predicts human-agent team performance. Beyond its practical value as a predictive tool, this reconceptualization of the experience-performance relationship necessitates a fresh examination of existing findings in the field. The results suggest that a team member's basic willingness to invest effort may be more fundamental to success than previously recognized mechanisms.
6.
Lin, Spencer; Rizk, Basem; Jun, Miru; Artze, Andy; Sullivan, Caitlin; Mozgai, Sharon; Fisher, Scott
Estuary: A Framework For Building Multimodal Low-Latency Real-Time Socially Interactive Agents Miscellaneous
2024, (arXiv:2410.20116 [cs]).
@misc{lin_estuary_2024,
title = {Estuary: A Framework For Building Multimodal Low-Latency Real-Time Socially Interactive Agents},
author = {Spencer Lin and Basem Rizk and Miru Jun and Andy Artze and Caitlin Sullivan and Sharon Mozgai and Scott Fisher},
url = {http://arxiv.org/abs/2410.20116},
doi = {10.1145/3652988.3696198},
year = {2024},
date = {2024-10-01},
urldate = {2024-12-06},
abstract = {The rise in capability and ubiquity of generative artificial intelligence (AI) technologies has enabled its application to the field of Socially Interactive Agents (SIAs). Despite rising interest in modern AI-powered components used for real-time SIA research, substantial friction remains due to the absence of a standardized and universal SIA framework. To target this absence, we developed Estuary: a multimodal (text, audio, and soon video) framework which facilitates the development of low-latency, real-time SIAs. Estuary seeks to reduce repeat work between studies and to provide a flexible platform that can be run entirely off-cloud to maximize configurability, controllability, reproducibility of studies, and speed of agent response times. We are able to do this by constructing a robust multimodal framework which incorporates current and future components seamlessly into a modular and interoperable architecture.},
note = {arXiv:2410.20116 [cs]},
keywords = {},
pubstate = {published},
tppubtype = {misc}
}
The rise in capability and ubiquity of generative artificial intelligence (AI) technologies has enabled its application to the field of Socially Interactive Agents (SIAs). Despite rising interest in modern AI-powered components used for real-time SIA research, substantial friction remains due to the absence of a standardized and universal SIA framework. To target this absence, we developed Estuary: a multimodal (text, audio, and soon video) framework which facilitates the development of low-latency, real-time SIAs. Estuary seeks to reduce repeat work between studies and to provide a flexible platform that can be run entirely off-cloud to maximize configurability, controllability, reproducibility of studies, and speed of agent response times. We are able to do this by constructing a robust multimodal framework which incorporates current and future components seamlessly into a modular and interoperable architecture.
7.
Owayyed, Mohammed Al; Tielman, Myrthe; Hartholt, Arno; Specht, Marcus; Brinkman, Willem-Paul
Agent-based social skills training systems: the ARTES architecture, interaction characteristics, learning theories and future outlooks Journal Article
In: Behaviour & Information Technology, pp. 1–28, 2024, ISSN: 0144-929X, 1362-3001.
@article{al_owayyed_agent-based_2024,
title = {Agent-based social skills training systems: the ARTES architecture, interaction characteristics, learning theories and future outlooks},
author = {Mohammed Al Owayyed and Myrthe Tielman and Arno Hartholt and Marcus Specht and Willem-Paul Brinkman},
url = {https://www.tandfonline.com/doi/full/10.1080/0144929X.2024.2374891},
doi = {10.1080/0144929X.2024.2374891},
issn = {0144-929X, 1362-3001},
year = {2024},
date = {2024-07-01},
urldate = {2024-08-15},
journal = {Behaviour & Information Technology},
pages = {1–28},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
8.
Nye, Benjamin D.; Okado, Yuko; Shiel, Aaron; Carr, Kayla; Rosenberg, Milton; Rice, Enora; Ostrander, Luke; Ju, Megan; Gutierrez, Cassandra; Ramirez, Dilan; Auerbach, Daniel; Aguirre, Angelica; Swartout, William
MentorStudio: Amplifying diverse voices through rapid, self-authorable virtual mentors Proceedings Article
In: Zenodo, 2023, (Publisher: Zenodo).
@inproceedings{nye_mentorstudio_2023,
title = {MentorStudio: Amplifying diverse voices through rapid, self-authorable virtual mentors},
author = {Benjamin D. Nye and Yuko Okado and Aaron Shiel and Kayla Carr and Milton Rosenberg and Enora Rice and Luke Ostrander and Megan Ju and Cassandra Gutierrez and Dilan Ramirez and Daniel Auerbach and Angelica Aguirre and William Swartout},
url = {https://zenodo.org/record/8226275},
doi = {10.5281/ZENODO.8226275},
year = {2023},
date = {2023-07-01},
urldate = {2024-01-11},
publisher = {Zenodo},
abstract = {Mentoring promotes underserved students' STEM persistence but it is difficult to scale up. Virtual agents can amplify mentors' experiences to larger audiences, which is particularly important for mentors from under-represented backgrounds and for underserved students with less access to mentors. This paper introduces MentorStudio, an online platform that allows real-life mentors to self-record and publish video-based conversational virtual agents. MentorStudio's goals are to increase speed, scheduling flexibility, and autonomy in creating intelligent virtual mentors. MentorStudio platform components are introduced, along with initial feedback regarding usability and acceptance collected from 20 STEM mentors who recorded virtual mentors. Overall, the MentorStudio platform has good ease-of-use and acceptance among mentors and offers a platform capable of recording large number of mentors to expand their reach to an unlimited number of students.},
note = {Publisher: Zenodo},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Mentoring promotes underserved students' STEM persistence but it is difficult to scale up. Virtual agents can amplify mentors' experiences to larger audiences, which is particularly important for mentors from under-represented backgrounds and for underserved students with less access to mentors. This paper introduces MentorStudio, an online platform that allows real-life mentors to self-record and publish video-based conversational virtual agents. MentorStudio's goals are to increase speed, scheduling flexibility, and autonomy in creating intelligent virtual mentors. MentorStudio platform components are introduced, along with initial feedback regarding usability and acceptance collected from 20 STEM mentors who recorded virtual mentors. Overall, the MentorStudio platform has good ease-of-use and acceptance among mentors and offers a platform capable of recording large number of mentors to expand their reach to an unlimited number of students.
Filter
2025
Wang, Yunzhe; Lucas, Gale M.; Becerik-Gerber, Burcin; Ustun, Volkan
Implicit Behavioral Alignment of Language Agents in High-Stakes Crowd Simulations Miscellaneous
2025, (arXiv:2509.16457 [cs]).
Abstract | Links | BibTeX | Tags: AI, DTIC, Virtual Agents
@misc{wang_implicit_2025,
title = {Implicit Behavioral Alignment of Language Agents in High-Stakes Crowd Simulations},
author = {Yunzhe Wang and Gale M. Lucas and Burcin Becerik-Gerber and Volkan Ustun},
url = {http://arxiv.org/abs/2509.16457},
doi = {10.48550/arXiv.2509.16457},
year = {2025},
date = {2025-09-01},
urldate = {2025-09-25},
publisher = {arXiv},
abstract = {Language-driven generative agents have enabled large-scale social simulations with transformative uses, from interpersonal training to aiding global policy-making. However, recent studies indicate that generative agent behaviors often deviate from expert expectations and real-world data–a phenomenon we term the Behavior-Realism Gap. To address this, we introduce a theoretical framework called Persona-Environment Behavioral Alignment (PEBA), formulated as a distribution matching problem grounded in Lewin's behavior equation stating that behavior is a function of the person and their environment. Leveraging PEBA, we propose PersonaEvolve (PEvo), an LLM-based optimization algorithm that iteratively refines agent personas, implicitly aligning their collective behaviors with realistic expert benchmarks within a specified environmental context. We validate PEvo in an active shooter incident simulation we developed, achieving an 84% average reduction in distributional divergence compared to no steering and a 34% improvement over explicit instruction baselines. Results also show PEvo-refined personas generalize to novel, related simulation scenarios. Our method greatly enhances behavioral realism and reliability in high-stakes social simulations. More broadly, the PEBA-PEvo framework provides a principled approach to developing trustworthy LLM-driven social simulations.},
note = {arXiv:2509.16457 [cs]},
keywords = {AI, DTIC, Virtual Agents},
pubstate = {published},
tppubtype = {misc}
}
Language-driven generative agents have enabled large-scale social simulations with transformative uses, from interpersonal training to aiding global policy-making. However, recent studies indicate that generative agent behaviors often deviate from expert expectations and real-world data–a phenomenon we term the Behavior-Realism Gap. To address this, we introduce a theoretical framework called Persona-Environment Behavioral Alignment (PEBA), formulated as a distribution matching problem grounded in Lewin's behavior equation stating that behavior is a function of the person and their environment. Leveraging PEBA, we propose PersonaEvolve (PEvo), an LLM-based optimization algorithm that iteratively refines agent personas, implicitly aligning their collective behaviors with realistic expert benchmarks within a specified environmental context. We validate PEvo in an active shooter incident simulation we developed, achieving an 84% average reduction in distributional divergence compared to no steering and a 34% improvement over explicit instruction baselines. Results also show PEvo-refined personas generalize to novel, related simulation scenarios. Our method greatly enhances behavioral realism and reliability in high-stakes social simulations. More broadly, the PEBA-PEvo framework provides a principled approach to developing trustworthy LLM-driven social simulations.
Hale, James; Kim, Peter H.; Gratch, Jonathan
“Provably fair” algorithms may perpetuate racial and gender bias: a study of salary dispute resolution Journal Article
In: Auton Agent Multi-Agent Syst, vol. 39, no. 1, pp. 20, 2025, ISSN: 1387-2532, 1573-7454.
Abstract | Links | BibTeX | Tags: Virtual Agents
@article{hale_provably_2025,
title = {“Provably fair” algorithms may perpetuate racial and gender bias: a study of salary dispute resolution},
author = {James Hale and Peter H. Kim and Jonathan Gratch},
url = {https://link.springer.com/10.1007/s10458-025-09703-x},
doi = {10.1007/s10458-025-09703-x},
issn = {1387-2532, 1573-7454},
year = {2025},
date = {2025-06-01},
urldate = {2025-03-18},
journal = {Auton Agent Multi-Agent Syst},
volume = {39},
number = {1},
pages = {20},
abstract = {Abstract
Prior work suggests automated dispute resolution tools using “provably fair” algorithms can address disparities between demographic groups. These methods use multi-criteria elicited preferences from all disputants and satisfy constraints to generate “fair” solutions. However, we analyze the potential for inequity to permeate proposals through the preference elicitation stage. This possibility arises if differences in dispositional attitudes differ between demographics, and those dispositions affect elicited preferences. Specifically, risk aversion plays a prominent role in predicting preferences. Risk aversion predicts a weaker relative preference for
salary
and a softer within-issue utility for each issue; this leads to worse compensation packages for risk-averse groups. These results raise important questions in AI-value alignment about whether an AI mediator should take explicit preferences at face value.},
keywords = {Virtual Agents},
pubstate = {published},
tppubtype = {article}
}
Abstract
Prior work suggests automated dispute resolution tools using “provably fair” algorithms can address disparities between demographic groups. These methods use multi-criteria elicited preferences from all disputants and satisfy constraints to generate “fair” solutions. However, we analyze the potential for inequity to permeate proposals through the preference elicitation stage. This possibility arises if differences in dispositional attitudes differ between demographics, and those dispositions affect elicited preferences. Specifically, risk aversion plays a prominent role in predicting preferences. Risk aversion predicts a weaker relative preference for
salary
and a softer within-issue utility for each issue; this leads to worse compensation packages for risk-averse groups. These results raise important questions in AI-value alignment about whether an AI mediator should take explicit preferences at face value.
Prior work suggests automated dispute resolution tools using “provably fair” algorithms can address disparities between demographic groups. These methods use multi-criteria elicited preferences from all disputants and satisfy constraints to generate “fair” solutions. However, we analyze the potential for inequity to permeate proposals through the preference elicitation stage. This possibility arises if differences in dispositional attitudes differ between demographics, and those dispositions affect elicited preferences. Specifically, risk aversion plays a prominent role in predicting preferences. Risk aversion predicts a weaker relative preference for
salary
and a softer within-issue utility for each issue; this leads to worse compensation packages for risk-averse groups. These results raise important questions in AI-value alignment about whether an AI mediator should take explicit preferences at face value.
Klumpe, Stella; Mitchell, Kelsey C.; Cox, Emma; Katz, Jeffrey S.; Lazarowski, Lucia; Deshpande, Gopikrishna; Gratch, Jonathan; Visser, Ewart J. De; Ayaz, Hasan; Li, Xingnan; Franke, Adrian A.; Krueger, Frank
Social bonding between humans, animals, and robots: Dogs outperform AIBOs, their robotic replicas, as social companions Journal Article
In: PLoS One, vol. 20, no. 6, pp. e0324312, 2025, ISSN: 1932-6203.
Abstract | Links | BibTeX | Tags: DTIC, Virtual Agents, Virtual Humans
@article{klumpe_social_2025,
title = {Social bonding between humans, animals, and robots: Dogs outperform AIBOs, their robotic replicas, as social companions},
author = {Stella Klumpe and Kelsey C. Mitchell and Emma Cox and Jeffrey S. Katz and Lucia Lazarowski and Gopikrishna Deshpande and Jonathan Gratch and Ewart J. De Visser and Hasan Ayaz and Xingnan Li and Adrian A. Franke and Frank Krueger},
editor = {Casey R. Lynch},
url = {https://dx.plos.org/10.1371/journal.pone.0324312},
doi = {10.1371/journal.pone.0324312},
issn = {1932-6203},
year = {2025},
date = {2025-06-01},
urldate = {2025-06-12},
journal = {PLoS One},
volume = {20},
number = {6},
pages = {e0324312},
abstract = {In the evolving landscape of technology, robots have emerged as social companions, prompting an investigation into social bonding between humans and robots. While human-animal interactions are well-studied, human-robot interactions (HRI) remain comparatively underexplored. Ethorobotics, a field of social robotic engineering based on ecology and ethology, suggests designing companion robots modeled on animal companions, which are simpler to emulate than humans. However, it is unclear whether these robots can match the social companionship provided by their original models. This study examined social bonding between humans and AIBOs, dog-inspired companion robots, compared to real dogs. Nineteen female participants engaged in 12 affiliative interactions with dogs and AIBOs across two counter-balanced, one-month bonding phases. Social bonding was assessed through urinary oxytocin (OXT) level change over an interaction, self-reported attachment using an adapted version of the Lexington Attachment to Pets Scale, and social companionship evaluations administering the Robot-Dog Questionnaire. To examine OXT level changes and self-reported attachment by comparing the two social companions, we conducted mixed-effects model analyses and planned follow-up comparisons. Frequency comparison, binary logistic regression, and thematic analysis were performed to analyze social companionship evaluations. Results revealed significant differences between dogs and AIBOs in fostering social bonds. OXT level change increased during interactions with dogs but decreased with AIBOs. Participants reported stronger attachment to dogs and rated them as better social companions. These findings highlight the current limitations of AIBOs in fostering social bonding immediately compared to dogs. Our study contributes to the growing HRI research by demonstrating an existing gap between AIBOs and dogs as social companions. It highlights the need for further investigation to understand the complexities of social bonding with companion robots, which is essential to implement successful applications for social robots in diverse domains such as the elderly and health care, education, and entertainment.},
keywords = {DTIC, Virtual Agents, Virtual Humans},
pubstate = {published},
tppubtype = {article}
}
In the evolving landscape of technology, robots have emerged as social companions, prompting an investigation into social bonding between humans and robots. While human-animal interactions are well-studied, human-robot interactions (HRI) remain comparatively underexplored. Ethorobotics, a field of social robotic engineering based on ecology and ethology, suggests designing companion robots modeled on animal companions, which are simpler to emulate than humans. However, it is unclear whether these robots can match the social companionship provided by their original models. This study examined social bonding between humans and AIBOs, dog-inspired companion robots, compared to real dogs. Nineteen female participants engaged in 12 affiliative interactions with dogs and AIBOs across two counter-balanced, one-month bonding phases. Social bonding was assessed through urinary oxytocin (OXT) level change over an interaction, self-reported attachment using an adapted version of the Lexington Attachment to Pets Scale, and social companionship evaluations administering the Robot-Dog Questionnaire. To examine OXT level changes and self-reported attachment by comparing the two social companions, we conducted mixed-effects model analyses and planned follow-up comparisons. Frequency comparison, binary logistic regression, and thematic analysis were performed to analyze social companionship evaluations. Results revealed significant differences between dogs and AIBOs in fostering social bonds. OXT level change increased during interactions with dogs but decreased with AIBOs. Participants reported stronger attachment to dogs and rated them as better social companions. These findings highlight the current limitations of AIBOs in fostering social bonding immediately compared to dogs. Our study contributes to the growing HRI research by demonstrating an existing gap between AIBOs and dogs as social companions. It highlights the need for further investigation to understand the complexities of social bonding with companion robots, which is essential to implement successful applications for social robots in diverse domains such as the elderly and health care, education, and entertainment.
Kwon, Deuksin; Hae, Jiwon; Clift, Emma; Shamsoddini, Daniel; Gratch, Jonathan; Lucas, Gale M.
ASTRA: A Negotiation Agent with Adaptive and Strategic Reasoning through Action in Dynamic Offer Optimization Miscellaneous
2025, (arXiv:2503.07129 [cs]).
Abstract | Links | BibTeX | Tags: DTIC, Virtual Agents
@misc{kwon_astra_2025,
title = {ASTRA: A Negotiation Agent with Adaptive and Strategic Reasoning through Action in Dynamic Offer Optimization},
author = {Deuksin Kwon and Jiwon Hae and Emma Clift and Daniel Shamsoddini and Jonathan Gratch and Gale M. Lucas},
url = {http://arxiv.org/abs/2503.07129},
doi = {10.48550/arXiv.2503.07129},
year = {2025},
date = {2025-03-01},
urldate = {2025-03-18},
publisher = {arXiv},
abstract = {Negotiation requires dynamically balancing self-interest and cooperation to maximize one's own utility. Yet, existing agents struggle due to bounded rationality in human data, low adaptability to counterpart behavior, and limited strategic reasoning. To address this, we introduce principle-driven negotiation agents, powered by ASTRA, a novel framework for turn-level offer optimization grounded in two core principles: opponent modeling and Tit-for-Tat reciprocity. ASTRA operates in three stages: (1) interpreting counterpart behavior, (2) optimizing counteroffers via a linear programming (LP) solver, and (3) selecting offers based on negotiation tactics and the partner's acceptance probability. Through simulations and human evaluations, our agent effectively adapts to an opponent's shifting stance and achieves favorable outcomes through enhanced adaptability and strategic reasoning. Beyond improving negotiation performance, it also serves as a powerful coaching tool, offering interpretable strategic feedback and optimal offer recommendations.},
note = {arXiv:2503.07129 [cs]},
keywords = {DTIC, Virtual Agents},
pubstate = {published},
tppubtype = {misc}
}
Negotiation requires dynamically balancing self-interest and cooperation to maximize one's own utility. Yet, existing agents struggle due to bounded rationality in human data, low adaptability to counterpart behavior, and limited strategic reasoning. To address this, we introduce principle-driven negotiation agents, powered by ASTRA, a novel framework for turn-level offer optimization grounded in two core principles: opponent modeling and Tit-for-Tat reciprocity. ASTRA operates in three stages: (1) interpreting counterpart behavior, (2) optimizing counteroffers via a linear programming (LP) solver, and (3) selecting offers based on negotiation tactics and the partner's acceptance probability. Through simulations and human evaluations, our agent effectively adapts to an opponent's shifting stance and achieves favorable outcomes through enhanced adaptability and strategic reasoning. Beyond improving negotiation performance, it also serves as a powerful coaching tool, offering interpretable strategic feedback and optimal offer recommendations.
Gurney, Nikolos; Pynadath, David V.; Miller, John H.
Willingness to work as a predictor of human-agent team success Journal Article
In: Front. Comput. Sci., vol. 7, pp. 1405436, 2025, ISSN: 2624-9898.
Abstract | Links | BibTeX | Tags: DTIC, Virtual Agents
@article{gurney_willingness_2025,
title = {Willingness to work as a predictor of human-agent team success},
author = {Nikolos Gurney and David V. Pynadath and John H. Miller},
url = {https://www.frontiersin.org/articles/10.3389/fcomp.2025.1405436/full},
doi = {10.3389/fcomp.2025.1405436},
issn = {2624-9898},
year = {2025},
date = {2025-03-01},
urldate = {2025-04-15},
journal = {Front. Comput. Sci.},
volume = {7},
pages = {1405436},
abstract = {Research shows that the effectiveness of human-agent teams depends heavily on human team members' prior experiences, whether from direct teaming activities or relevant domain knowledge. While researchers have proposed various mechanisms to explain this relationship, we present a simpler alternative explanation: experience serves primarily as an indicator of a person's fundamental willingness to engage in teaming tasks. We introduce a measure called “willingness to work” that quantifies this underlying disposition. Our empirical analysis demonstrates that this straightforward metric robustly predicts human-agent team performance. Beyond its practical value as a predictive tool, this reconceptualization of the experience-performance relationship necessitates a fresh examination of existing findings in the field. The results suggest that a team member's basic willingness to invest effort may be more fundamental to success than previously recognized mechanisms.},
keywords = {DTIC, Virtual Agents},
pubstate = {published},
tppubtype = {article}
}
Research shows that the effectiveness of human-agent teams depends heavily on human team members' prior experiences, whether from direct teaming activities or relevant domain knowledge. While researchers have proposed various mechanisms to explain this relationship, we present a simpler alternative explanation: experience serves primarily as an indicator of a person's fundamental willingness to engage in teaming tasks. We introduce a measure called “willingness to work” that quantifies this underlying disposition. Our empirical analysis demonstrates that this straightforward metric robustly predicts human-agent team performance. Beyond its practical value as a predictive tool, this reconceptualization of the experience-performance relationship necessitates a fresh examination of existing findings in the field. The results suggest that a team member's basic willingness to invest effort may be more fundamental to success than previously recognized mechanisms.
2024
Lin, Spencer; Rizk, Basem; Jun, Miru; Artze, Andy; Sullivan, Caitlin; Mozgai, Sharon; Fisher, Scott
Estuary: A Framework For Building Multimodal Low-Latency Real-Time Socially Interactive Agents Miscellaneous
2024, (arXiv:2410.20116 [cs]).
Abstract | Links | BibTeX | Tags: Virtual Agents
@misc{lin_estuary_2024,
title = {Estuary: A Framework For Building Multimodal Low-Latency Real-Time Socially Interactive Agents},
author = {Spencer Lin and Basem Rizk and Miru Jun and Andy Artze and Caitlin Sullivan and Sharon Mozgai and Scott Fisher},
url = {http://arxiv.org/abs/2410.20116},
doi = {10.1145/3652988.3696198},
year = {2024},
date = {2024-10-01},
urldate = {2024-12-06},
abstract = {The rise in capability and ubiquity of generative artificial intelligence (AI) technologies has enabled its application to the field of Socially Interactive Agents (SIAs). Despite rising interest in modern AI-powered components used for real-time SIA research, substantial friction remains due to the absence of a standardized and universal SIA framework. To target this absence, we developed Estuary: a multimodal (text, audio, and soon video) framework which facilitates the development of low-latency, real-time SIAs. Estuary seeks to reduce repeat work between studies and to provide a flexible platform that can be run entirely off-cloud to maximize configurability, controllability, reproducibility of studies, and speed of agent response times. We are able to do this by constructing a robust multimodal framework which incorporates current and future components seamlessly into a modular and interoperable architecture.},
note = {arXiv:2410.20116 [cs]},
keywords = {Virtual Agents},
pubstate = {published},
tppubtype = {misc}
}
The rise in capability and ubiquity of generative artificial intelligence (AI) technologies has enabled its application to the field of Socially Interactive Agents (SIAs). Despite rising interest in modern AI-powered components used for real-time SIA research, substantial friction remains due to the absence of a standardized and universal SIA framework. To target this absence, we developed Estuary: a multimodal (text, audio, and soon video) framework which facilitates the development of low-latency, real-time SIAs. Estuary seeks to reduce repeat work between studies and to provide a flexible platform that can be run entirely off-cloud to maximize configurability, controllability, reproducibility of studies, and speed of agent response times. We are able to do this by constructing a robust multimodal framework which incorporates current and future components seamlessly into a modular and interoperable architecture.
Owayyed, Mohammed Al; Tielman, Myrthe; Hartholt, Arno; Specht, Marcus; Brinkman, Willem-Paul
Agent-based social skills training systems: the ARTES architecture, interaction characteristics, learning theories and future outlooks Journal Article
In: Behaviour & Information Technology, pp. 1–28, 2024, ISSN: 0144-929X, 1362-3001.
Links | BibTeX | Tags: Social Simulation, Virtual Agents, Virtual Humans
@article{al_owayyed_agent-based_2024,
title = {Agent-based social skills training systems: the ARTES architecture, interaction characteristics, learning theories and future outlooks},
author = {Mohammed Al Owayyed and Myrthe Tielman and Arno Hartholt and Marcus Specht and Willem-Paul Brinkman},
url = {https://www.tandfonline.com/doi/full/10.1080/0144929X.2024.2374891},
doi = {10.1080/0144929X.2024.2374891},
issn = {0144-929X, 1362-3001},
year = {2024},
date = {2024-07-01},
urldate = {2024-08-15},
journal = {Behaviour & Information Technology},
pages = {1–28},
keywords = {Social Simulation, Virtual Agents, Virtual Humans},
pubstate = {published},
tppubtype = {article}
}
2023
Nye, Benjamin D.; Okado, Yuko; Shiel, Aaron; Carr, Kayla; Rosenberg, Milton; Rice, Enora; Ostrander, Luke; Ju, Megan; Gutierrez, Cassandra; Ramirez, Dilan; Auerbach, Daniel; Aguirre, Angelica; Swartout, William
MentorStudio: Amplifying diverse voices through rapid, self-authorable virtual mentors Proceedings Article
In: Zenodo, 2023, (Publisher: Zenodo).
Abstract | Links | BibTeX | Tags: DTIC, Learning Sciences, UARC, Virtual Agents
@inproceedings{nye_mentorstudio_2023,
title = {MentorStudio: Amplifying diverse voices through rapid, self-authorable virtual mentors},
author = {Benjamin D. Nye and Yuko Okado and Aaron Shiel and Kayla Carr and Milton Rosenberg and Enora Rice and Luke Ostrander and Megan Ju and Cassandra Gutierrez and Dilan Ramirez and Daniel Auerbach and Angelica Aguirre and William Swartout},
url = {https://zenodo.org/record/8226275},
doi = {10.5281/ZENODO.8226275},
year = {2023},
date = {2023-07-01},
urldate = {2024-01-11},
publisher = {Zenodo},
abstract = {Mentoring promotes underserved students' STEM persistence but it is difficult to scale up. Virtual agents can amplify mentors' experiences to larger audiences, which is particularly important for mentors from under-represented backgrounds and for underserved students with less access to mentors. This paper introduces MentorStudio, an online platform that allows real-life mentors to self-record and publish video-based conversational virtual agents. MentorStudio's goals are to increase speed, scheduling flexibility, and autonomy in creating intelligent virtual mentors. MentorStudio platform components are introduced, along with initial feedback regarding usability and acceptance collected from 20 STEM mentors who recorded virtual mentors. Overall, the MentorStudio platform has good ease-of-use and acceptance among mentors and offers a platform capable of recording large number of mentors to expand their reach to an unlimited number of students.},
note = {Publisher: Zenodo},
keywords = {DTIC, Learning Sciences, UARC, Virtual Agents},
pubstate = {published},
tppubtype = {inproceedings}
}
Mentoring promotes underserved students' STEM persistence but it is difficult to scale up. Virtual agents can amplify mentors' experiences to larger audiences, which is particularly important for mentors from under-represented backgrounds and for underserved students with less access to mentors. This paper introduces MentorStudio, an online platform that allows real-life mentors to self-record and publish video-based conversational virtual agents. MentorStudio's goals are to increase speed, scheduling flexibility, and autonomy in creating intelligent virtual mentors. MentorStudio platform components are introduced, along with initial feedback regarding usability and acceptance collected from 20 STEM mentors who recorded virtual mentors. Overall, the MentorStudio platform has good ease-of-use and acceptance among mentors and offers a platform capable of recording large number of mentors to expand their reach to an unlimited number of students.