Yoshiko Arima
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Yoshiko Arima has a total of 8 co-authors. Here are the top co-authors by number of shared publications:
Mahiro Okada
Department of Psychology, Center for Social and Psychological Research of Metaverse, Kyoto University of Advanced Science
3 shared publications- Susumu Ohnuma
Hokkaido University
3 shared publications Yuki Harada
Department of Psychology, Center for Social and Psychological Research of Metaverse, Kyoto University of Advanced Science
3 shared publications- Mahiro Okada
Center for Social and Psychological Research of Metaverse, Kyoto University of Advanced Science, Kyoto 615-8577, Japan
2 shared publications Ryoji Yukihiro
Department of Psychology, Kyoto Gakuen University, Kameoka, Japan
1 shared publication- Yosuke Hattori
National Hospital Organization Nagoya Medical Center, Orthopaedic Surgery and Rheumatology, Nagoya, Japan
1 shared publication
Latest Publications
- Open Access
- 12 November 2026
- Published by Frontiers Media SA in Frontiers in Virtual Reality
- Vol. 7, 1988267
Correction: Classifying interpersonal interaction in virtual reality: sensor-based analysis of human interaction with pre-recorded avatars
Correction on: Arima Y, Harada Y and Okada M (2025) Classifying interpersonal interaction in virtual reality: sensor-based analysis of human interaction with pre-recorded avatars. Front. Virtual Real. 6:1623764. doi:10.3389/frvir.2025.1623764.Adding/removing text In the published article, the order of the Results and Discussion sections was incorrect. The published structure was: Section 5 Results → Section 5.1 Preliminary experiment → Section 6 Discussion → Section 6.1 Main experiment → Section 6.1.1 Decision tree → Section 6.1.2 Effects of human or bot conditions → Section 6.1.3 Mediation analysis. Consequently, the main-experiment Results were placed under the Discussion section. A correction has been made to the sections 5 Results and 6 Discussion: The section structure has been corrected as follows: Section 5 Results; Section 5.1 Preliminary experiment, with the discussion of the preliminary experiment placed within this subsection; Section 5.2 Main experiment; Section 5.2.1 Decision tree; Section 5.2.2 Effects of human or bot conditions; Section 5.2.3 Mediation analysis; and Section 6 Discussion, which follows all Results sections. In addition, the duplicated interpretative passage in the Results has been removed, while the corresponding text in the Discussion is retained. The original version of this article has been updated.
- 12 November 2026
- Published by Frontiers Media SA in Frontiers in Virtual Reality
- Vol. 7, 1988267
Open AccessCorrection: Classifying interpersonal interaction in virtual reality: sensor-based analysis of human interaction with pre-recorded avatars
Correction on: Arima Y, Harada Y and Okada M (2025) Classifying interpersonal interaction in virtual reality: sensor-based analysis of human interaction with pre-recorded avatars. Front. Virtual Real. 6:1623764. doi:10.3389/frvir.2025.1623764.Adding/removing text In the published article, the order of the Results and Discussion sections was incorrect. The published structure was: Section 5 Results → Section 5.1 Preliminary experiment → Section 6 Discussion → Section 6.1 Main experiment → Section 6.1.1 Decision tree → Section 6.1.2 Effects of human or bot conditions → Section 6.1.3 Mediation analysis. Consequently, the main-experiment Results were placed under the Discussion section. A correction has been made to the sections 5 Results and 6 Discussion: The section structure has been corrected as follows: Section 5 Results; Section 5.1 Preliminary experiment, with the discussion of the preliminary experiment placed within this subsection; Section 5.2 Main experiment; Section 5.2.1 Decision tree; Section 5.2.2 Effects of human or bot conditions; Section 5.2.3 Mediation analysis; and Section 6 Discussion, which follows all Results sections. In addition, the duplicated interpretative passage in the Results has been removed, while the corresponding text in the Discussion is retained. The original version of this article has been updated.
- Open Access
- 23 October 2026
- Published by Elsevier in Computers in Human Behavior: Artificial Humans
- Vol. 10
Semantic network sharing in human versus LLM discussion groups
- 23 October 2026
- Published by Elsevier in Computers in Human Behavior: Artificial Humans
- Vol. 10
Open AccessSemantic network sharing in human versus LLM discussion groups
- Open Access
- 25 September 2026
- Published by Springer Nature in Virtual Reality
Classifying interpersonal synchrony in VR joint actions: human-human versus human-bot interactions
- 25 September 2026
- Published by Springer Nature in Virtual Reality
Open AccessClassifying interpersonal synchrony in VR joint actions: human-human versus human-bot interactions
- Open Access
- 17 August 2026
- Published by Elsevier in Computers in Human Behavior Reports
- Vol. 23
Forming and losing trust with AI agents: A computational account of human–AI interaction in the trust game
- 17 August 2026
- Published by Elsevier in Computers in Human Behavior Reports
- Vol. 23
Open AccessForming and losing trust with AI agents: A computational account of human–AI interaction in the trust game
- Open Access
- 10 August 2026
- Published by Center for Open Science in PsyArXiv
Semantic Network Sharing in Human Versus LLM Discussion Groups
Using a corpus of 51 mock-jury discussion groups, we analyzed the discussion process of human groups and then compared it with that of large language model (LLM) groups in an exploratory analysis. The task was a hidden profile task, solvable only by combining information distributed across members. We examined solution-path attainment and the sharedness of the word-level semantic network (the cross-member edge ratio: the proportion of edges linking words introduced by different members) in relation to the information-sharing process. Human discussion progressed from individual to group-level sharing: an information-sharing phase in which intra-individual semantic networks remained (cross-member edge ratio .34), followed by an information-integration phase in which network links became shared across members (.54), and a decision phase (.66). In LLM groups calibrated to human behavioral statistics, this three-phase structure was radically compressed, lacking the human allocation of time. Under a memory constraint, however, the individual-to-group sharing shift observed in humans emerged. A stepwise prompt ladder reproduced the reported collapse of LLM collective reasoning and identified the constraints that avert it. Moreover, the association between the sharedness of the semantic network and solution-path attainment found in human groups (ρ = +.52, n = 50) was not observed in LLM groups (n = 8 per condition). These differences are interpreted as showing the contrast between human groups, which come to share a semantic network gradually through discussion, and LLM groups, which share a knowledge structure from the outset.
- 10 August 2026
- Published by Center for Open Science in PsyArXiv
Open AccessSemantic Network Sharing in Human Versus LLM Discussion Groups
Using a corpus of 51 mock-jury discussion groups, we analyzed the discussion process of human groups and then compared it with that of large language model (LLM) groups in an exploratory analysis. The task was a hidden profile task, solvable only by combining information distributed across members. We examined solution-path attainment and the sharedness of the word-level semantic network (the cross-member edge ratio: the proportion of edges linking words introduced by different members) in relation to the information-sharing process. Human discussion progressed from individual to group-level sharing: an information-sharing phase in which intra-individual semantic networks remained (cross-member edge ratio .34), followed by an information-integration phase in which network links became shared across members (.54), and a decision phase (.66). In LLM groups calibrated to human behavioral statistics, this three-phase structure was radically compressed, lacking the human allocation of time. Under a memory constraint, however, the individual-to-group sharing shift observed in humans emerged. A stepwise prompt ladder reproduced the reported collapse of LLM collective reasoning and identified the constraints that avert it. Moreover, the association between the sharedness of the semantic network and solution-path attainment found in human groups (ρ = +.52, n = 50) was not observed in LLM groups (n = 8 per condition). These differences are interpreted as showing the contrast between human groups, which come to share a semantic network gradually through discussion, and LLM groups, which share a knowledge structure from the outset.
- Open Access
- 17 June 2026
- Published by Elsevier in SSRN Electronic Journal
Forming and losing trust with AI agents: A computational account of human–AI interaction in the trust game
How do humans form trust with AI agents? We address this question with a computational model of human trust game behaviour grounded in the Active Inference Framework (AIF), tested against two behavioural experiments with bot partners. The model introduces a parameter γ, a trial-by-trial weighting of social versus economic inference in the free energy computation; higher γ indicates an internal model in which both self and other are predicted to prioritize social over economic value. Participants’ γ was estimated from ob- served behaviour via gradient descent. Across both experiments, the model achieved high predictive accuracy for trust/betrayal choices (F1 > .80) under cooperative and betrayal conditions alike. Study 1, in which bots maintained a fixed cooperation or betrayal pattern, replicated the simulated belief trajec- tories with only a small effect of Bot condition on γ. Study 2 useda between- subjects design contrasting a Trust condition (continuous cooperation) with a Betray condition (cooperation–betrayal–cooperation). The simulation pre- dicted that γ would rebound once cooperation resumed; in human data it did not. Beliefs (posterior predictions of the partner’s next cooperation) re- covered partially in the final phase, consistent with precision convergence in Bayesian updating, but investment rate and γ remained suppressed. Thus humans update their predictions about a bot’s behaviour in line with Active Inference accounts, yet their underlying social–economic value weighting car- ries a longer trace of the betrayal. This asymmetry, observed only in humans, points to factors beyond the present model, including individual differences such as sex.
- 17 June 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessForming and losing trust with AI agents: A computational account of human–AI interaction in the trust game
How do humans form trust with AI agents? We address this question with a computational model of human trust game behaviour grounded in the Active Inference Framework (AIF), tested against two behavioural experiments with bot partners. The model introduces a parameter γ, a trial-by-trial weighting of social versus economic inference in the free energy computation; higher γ indicates an internal model in which both self and other are predicted to prioritize social over economic value. Participants’ γ was estimated from ob- served behaviour via gradient descent. Across both experiments, the model achieved high predictive accuracy for trust/betrayal choices (F1 > .80) under cooperative and betrayal conditions alike. Study 1, in which bots maintained a fixed cooperation or betrayal pattern, replicated the simulated belief trajec- tories with only a small effect of Bot condition on γ. Study 2 useda between- subjects design contrasting a Trust condition (continuous cooperation) with a Betray condition (cooperation–betrayal–cooperation). The simulation pre- dicted that γ would rebound once cooperation resumed; in human data it did not. Beliefs (posterior predictions of the partner’s next cooperation) re- covered partially in the final phase, consistent with precision convergence in Bayesian updating, but investment rate and γ remained suppressed. Thus humans update their predictions about a bot’s behaviour in line with Active Inference accounts, yet their underlying social–economic value weighting car- ries a longer trace of the betrayal. This asymmetry, observed only in humans, points to factors beyond the present model, including individual differences such as sex.
- Open Access
- 6 April 2026
- Published by Springer Nature in AI & SOCIETY
- Vol. 41 (7) , 6489-6500
Emergence of class awareness: text analysis of reports from simulated society game participants
- 6 April 2026
- Published by Springer Nature in AI & SOCIETY
- Vol. 41 (7) , 6489-6500
Open AccessEmergence of class awareness: text analysis of reports from simulated society game participants
- Open Access
Active Inference Modeling of Socially Shared Cognition in Virtual Reality
This study proposes a process model for sharing ambiguous category concepts in virtual reality (VR) using an active inference framework. The model executes a dual-layer Bayesian update after observing both self and partner actions and predicts actions that minimize free energy. To incorporate agreement-seeking with others into active inference, we added disagreement in category judgments as a risk term in the free energy, weighted by gaze synchrony measured using Dynamic Time Warping (DTW), which is assumed to reflect joint attention. To validate the model, an object classification task in VR including ambiguous items was created. The experiment was conducted first under a bot avatar condition, in which ambiguous category judgments were always incorrect, and then under a human–human pair condition. This design allowed verification of the collaborative learning process by which human pairs reached agreement from the same degree of ambiguity. Analysis of experimental data from 14 participants showed that the model achieved high prediction accuracy for observed values as learning progressed. Introducing gaze synchrony weighting () further improved prediction accuracy, yielding optimal performance. This approach provides a new framework for modeling socially shared cognition using active inference in human–robot interaction contexts.Open AccessActive Inference Modeling of Socially Shared Cognition in Virtual Reality
This study proposes a process model for sharing ambiguous category concepts in virtual reality (VR) using an active inference framework. The model executes a dual-layer Bayesian update after observing both self and partner actions and predicts actions that minimize free energy. To incorporate agreement-seeking with others into active inference, we added disagreement in category judgments as a risk term in the free energy, weighted by gaze synchrony measured using Dynamic Time Warping (DTW), which is assumed to reflect joint attention. To validate the model, an object classification task in VR including ambiguous items was created. The experiment was conducted first under a bot avatar condition, in which ambiguous category judgments were always incorrect, and then under a human–human pair condition. This design allowed verification of the collaborative learning process by which human pairs reached agreement from the same degree of ambiguity. Analysis of experimental data from 14 participants showed that the model achieved high prediction accuracy for observed values as learning progressed. Introducing gaze synchrony weighting () further improved prediction accuracy, yielding optimal performance. This approach provides a new framework for modeling socially shared cognition using active inference in human–robot interaction contexts. - Open Access
- 14 November 2025
- Published by Springer Nature in Research Square
Active Inference Modeling of Socially Shared Cognition in Virtual Reality
This study proposes a process model for sharing ambiguous category concepts in virtual reality (VR) using an active inference framework. The model executes a dual-layer Bayesian update after observing both self and partner actions and predicts actions that minimize free energy. A disagreement in category judgment was added to the free energy as a risk term (corresponding to expected surprise in active inference). As the weight of this term, gaze synchrony measured by Dynamic Time Warping (DTW), assumed to re- flect joint attention, was used. The hypothesis was that higher weighting of gaze synchrony would improve prediction accuracy. To validate the model, an object classification task in VR including ambiguous items was created. The experiment was conducted first under a bot avatar condition, in which gaze was not synchronized and ambiguous category judgments were always incorrect, and then under a human–human pair condition. This design allowed verification of the collaborative learning process by which human pairs reached agreement. Analysis of experimental data from 14 participants showed that the model achieved high prediction accuracy for observed values as learning progressed. Introducing DTW as a model parameter further improved prediction accuracy, with optimal performance at synchrony weights of γ 0 = 0 . 5 - 0 . 9 . This approach provides a new framework for modeling shared social cognition using active inference.
- 14 November 2025
- Published by Springer Nature in Research Square
Open AccessActive Inference Modeling of Socially Shared Cognition in Virtual Reality
This study proposes a process model for sharing ambiguous category concepts in virtual reality (VR) using an active inference framework. The model executes a dual-layer Bayesian update after observing both self and partner actions and predicts actions that minimize free energy. A disagreement in category judgment was added to the free energy as a risk term (corresponding to expected surprise in active inference). As the weight of this term, gaze synchrony measured by Dynamic Time Warping (DTW), assumed to re- flect joint attention, was used. The hypothesis was that higher weighting of gaze synchrony would improve prediction accuracy. To validate the model, an object classification task in VR including ambiguous items was created. The experiment was conducted first under a bot avatar condition, in which gaze was not synchronized and ambiguous category judgments were always incorrect, and then under a human–human pair condition. This design allowed verification of the collaborative learning process by which human pairs reached agreement. Analysis of experimental data from 14 participants showed that the model achieved high prediction accuracy for observed values as learning progressed. Introducing DTW as a model parameter further improved prediction accuracy, with optimal performance at synchrony weights of γ 0 = 0 . 5 - 0 . 9 . This approach provides a new framework for modeling shared social cognition using active inference.
- Open Access
- 15 October 2025
- Published by Springer Nature in Research Square
Classifying Interpersonal Synchrony in VR Joint Actions: Human-Human versus Human-Bot Interactions
This study collected sensor data from VR goggles in pairs and examined interpersonal synchrony using a within-subject design. Participants positioned opposite each other in VR space judged object categories (kitchen vs. garage utensils) and responded with movements. We created an LSTM classification model using movement data from pairs instructed to interact either competitively (Game session) or synchronously (Collab session). The model achieved 93.2% accuracy (F1 score = 0.93.2) in classifying these interaction modes using only three-axis head acceleration data from the goggles. We then applied this within-subject trained model to Bot interaction sessions, where participants worked with Bot avatars under high or low accuracy conditions. Dynamic Time Warping (DTW) distance was used to quantify synchrony. Results showed that when movements were classified as synchronous, DTW distance was lower, indicating higher synchrony. A significant interaction effect was found between movement classification (Game/Collab) and Bot accuracy condition on DTW distance. Exploratory cross-subject validation showed reduced performance (F1 = 0.72), though head rotation features remained important across participants. These findings demonstrate that the classification model successfully captures interaction dynamics across different task conditions within subjects, and that synchrony with Bots is modulated by behavioral predictability.
- 15 October 2025
- Published by Springer Nature in Research Square
Open AccessClassifying Interpersonal Synchrony in VR Joint Actions: Human-Human versus Human-Bot Interactions
This study collected sensor data from VR goggles in pairs and examined interpersonal synchrony using a within-subject design. Participants positioned opposite each other in VR space judged object categories (kitchen vs. garage utensils) and responded with movements. We created an LSTM classification model using movement data from pairs instructed to interact either competitively (Game session) or synchronously (Collab session). The model achieved 93.2% accuracy (F1 score = 0.93.2) in classifying these interaction modes using only three-axis head acceleration data from the goggles. We then applied this within-subject trained model to Bot interaction sessions, where participants worked with Bot avatars under high or low accuracy conditions. Dynamic Time Warping (DTW) distance was used to quantify synchrony. Results showed that when movements were classified as synchronous, DTW distance was lower, indicating higher synchrony. A significant interaction effect was found between movement classification (Game/Collab) and Bot accuracy condition on DTW distance. Exploratory cross-subject validation showed reduced performance (F1 = 0.72), though head rotation features remained important across participants. These findings demonstrate that the classification model successfully captures interaction dynamics across different task conditions within subjects, and that synchrony with Bots is modulated by behavioral predictability.