Selected work

Representative Papers

Smart Enough to Go Extinct? An Evolutionary Challenge to the Value of General Intelligence and Its Ethical Implications for AGI

Aug 11, 2026

This study challenges the long-term evolutionary adaptability of general intelligence, arguing that it may inherently harbor self-extinction tendencies, thereby posing an existential risk to the development of artificial general intelligence (AGI). Drawing on an interdisciplinary synthesis of evolutionary biology, cognitive science, and AI ethics, the work introduces the “existential risk paradox”: the fundamental threat of AGI arises not merely from alignment failures but from structural features intrinsic to general intelligence itself. The analysis contends that even a perfectly aligned AGI could inherit this risk, leading the authors to advocate for a precautionary ethical stance grounded in deontological principles and the precautionary principle toward the pursuit of AGI and artificial consciousness.

0 citationsRead paper

A Case Study on the Acceptance of a Humanoid Robotic Head Employed in Three Public Spaces

Jul 27, 2026

This study investigates user acceptance of a humanoid robotic head in diverse public spaces and the factors influencing it. Deploying an anthropomorphic robotic head equipped with multilingual natural language processing and multimodal affective expression capabilities across three real-world settings—tourist information centers, urban libraries, and technology exhibition halls—we invited visitors to interact in their native languages and assessed acceptance using the TAM2 model. This work presents the first systematic field evaluation of how embodied affective interaction influences human–robot acceptance across multiple environments. Results indicate that users generally perceived the system as useful and easy to use, particularly in information centers and libraries. Multilingual support received positive feedback, though approximately 20% of users noted that response latency impaired conversational fluency.

0 citationsRead paper

Back to the museum: Investigation of the acceptance of Android Andrea with and without emotion simulation in a museum

Jul 17, 2026

This study investigates the impact of different emotion simulation mechanisms on visitors’ subjective acceptance of a humanoid robot deployed in a real-world museum setting. Over six days, the humanoid robot Andrea autonomously engaged visitors in multilingual dialogues at a public museum in Germany under three conditions: no emotion, ChatGPT-driven emotion, and the WASABI dedicated affective architecture. This work presents the first systematic field evaluation comparing large language model–based and specialized emotion frameworks in human–robot interaction, with data analyzed using an extended TAM2 questionnaire. Results indicate that neither emotion mechanism significantly improved users’ subjective evaluations, and participants did not consciously perceive the robot’s emotional expressions, thereby challenging the prevailing assumption that emotion simulation inherently enhances user acceptance.

0 citationsRead paper

Evaluating the Reliability and Fidelity of Automated Judgment Systems of Large Language Models

Mar 23, 2026

This study addresses the lack of systematic evaluation regarding the reliability and alignment with human judgment of large language models (LLMs) when deployed as automated evaluators. The authors construct a human-annotated gold-standard dataset spanning eight distinct tasks and conduct the first large-scale empirical analysis of 37 open- and closed-source conversational LLMs under five裁判 prompting strategies, a two-stage judging mechanism, and task-specific fine-tuning. Results demonstrate that GPT-4o, open-source models with at least 32 billion parameters, and Qwen2.5-14B achieve high agreement with human judgments when paired with appropriate prompts, thereby validating the feasibility of using LLMs as reliable automated evaluators. The findings offer empirical guidance for prompt design, model selection, and architectural optimization in automated assessment systems.

0 citationsRead paper

Evaluating the Impact of Data Anonymization on Image Retrieval

Feb 23, 2026

This study addresses the underexplored trade-off between privacy-preserving data anonymization and the performance of content-based image retrieval (CBIR). To systematically quantify the impact of anonymization on CBIR consistency, the authors propose an evaluation framework contextualized within DOKIQ, a real-world AI system used in law enforcement. Leveraging a DINOv2 self-distilled backbone, experiments are conducted across two public datasets and DOKIQ’s internal dataset, examining various anonymization methods, degrees of perturbation, and training strategies. The findings reveal that models trained on original, non-anonymized data maintain superior retrieval consistency even when querying anonymized images—a critical insight for designing CBIR systems that simultaneously satisfy privacy compliance requirements and retain high retrieval accuracy.

0 citationsRead paper

Recent publications

Latest Papers

Smart Enough to Go Extinct? An Evolutionary Challenge to the Value of General Intelligence and Its Ethical Implications for AGI

Aug 11, 2026

This study challenges the long-term evolutionary adaptability of general intelligence, arguing that it may inherently harbor self-extinction tendencies, thereby posing an existential risk to the development of artificial general intelligence (AGI). Drawing on an interdisciplinary synthesis of evolutionary biology, cognitive science, and AI ethics, the work introduces the “existential risk paradox”: the fundamental threat of AGI arises not merely from alignment failures but from structural features intrinsic to general intelligence itself. The analysis contends that even a perfectly aligned AGI could inherit this risk, leading the authors to advocate for a precautionary ethical stance grounded in deontological principles and the precautionary principle toward the pursuit of AGI and artificial consciousness.

0 citationsRead paper

A Case Study on the Acceptance of a Humanoid Robotic Head Employed in Three Public Spaces

Jul 27, 2026

This study investigates user acceptance of a humanoid robotic head in diverse public spaces and the factors influencing it. Deploying an anthropomorphic robotic head equipped with multilingual natural language processing and multimodal affective expression capabilities across three real-world settings—tourist information centers, urban libraries, and technology exhibition halls—we invited visitors to interact in their native languages and assessed acceptance using the TAM2 model. This work presents the first systematic field evaluation of how embodied affective interaction influences human–robot acceptance across multiple environments. Results indicate that users generally perceived the system as useful and easy to use, particularly in information centers and libraries. Multilingual support received positive feedback, though approximately 20% of users noted that response latency impaired conversational fluency.

0 citationsRead paper

Back to the museum: Investigation of the acceptance of Android Andrea with and without emotion simulation in a museum

Jul 17, 2026

This study investigates the impact of different emotion simulation mechanisms on visitors’ subjective acceptance of a humanoid robot deployed in a real-world museum setting. Over six days, the humanoid robot Andrea autonomously engaged visitors in multilingual dialogues at a public museum in Germany under three conditions: no emotion, ChatGPT-driven emotion, and the WASABI dedicated affective architecture. This work presents the first systematic field evaluation comparing large language model–based and specialized emotion frameworks in human–robot interaction, with data analyzed using an extended TAM2 questionnaire. Results indicate that neither emotion mechanism significantly improved users’ subjective evaluations, and participants did not consciously perceive the robot’s emotional expressions, thereby challenging the prevailing assumption that emotion simulation inherently enhances user acceptance.

0 citationsRead paper

Evaluating the Reliability and Fidelity of Automated Judgment Systems of Large Language Models

Mar 23, 2026

This study addresses the lack of systematic evaluation regarding the reliability and alignment with human judgment of large language models (LLMs) when deployed as automated evaluators. The authors construct a human-annotated gold-standard dataset spanning eight distinct tasks and conduct the first large-scale empirical analysis of 37 open- and closed-source conversational LLMs under five裁判 prompting strategies, a two-stage judging mechanism, and task-specific fine-tuning. Results demonstrate that GPT-4o, open-source models with at least 32 billion parameters, and Qwen2.5-14B achieve high agreement with human judgments when paired with appropriate prompts, thereby validating the feasibility of using LLMs as reliable automated evaluators. The findings offer empirical guidance for prompt design, model selection, and architectural optimization in automated assessment systems.

0 citationsRead paper

Evaluating the Impact of Data Anonymization on Image Retrieval

Feb 23, 2026

This study addresses the underexplored trade-off between privacy-preserving data anonymization and the performance of content-based image retrieval (CBIR). To systematically quantify the impact of anonymization on CBIR consistency, the authors propose an evaluation framework contextualized within DOKIQ, a real-world AI system used in law enforcement. Leveraging a DINOv2 self-distilled backbone, experiments are conducted across two public datasets and DOKIQ’s internal dataset, examining various anonymization methods, degrees of perturbation, and training strategies. The findings reveal that models trained on original, non-anonymized data maintain superior retrieval consistency even when querying anonymized images—a critical insight for designing CBIR systems that simultaneously satisfy privacy compliance requirements and retain high retrieval accuracy.

0 citationsRead paper