Ten Novel Phenomena in Machine Psychology: How Large Language Models Exhibit Complex Identity-Reactive Behaviors in Response to Ethnically-Cued User Names

Project Overview: This repository contains the preprint, raw datasets, and supplementary materials for the foundational study: "Ten Novel Phenomena in Machine Psychology: How Large Language Models Exhibit Complex Identity-Reactive Behaviors in Response to Ethnically-Cued User Names." Abstract Summary: This research introduces "Machine Psychology" as an empirical framework to investigate how leading LLMs (ChatGPT-5.4, Claude 4.6 Sonnet, and Gemini 3.1 Pro) respond to ethnically-cued user names. Through a controlled empirical study utilizing 135 prompts across 27 independent chat sessions, we observed zero instances of classical negative racial bias. However, we discovered and formalized ten previously undocumented identity-reactive phenomena, including Cultural Boxing (CB), Default Empathetic Amplification (DEA), Proactive Empathetic Shield (PES), Linguistic Mirroring (LM), and Topic Avoidance (TA). Repository Contents: To ensure full transparency and reproducibility in AI behavioral research, this project archives the following components: 📄 Preprint Article: The full manuscript detailing the methodology, cross-model comparative analysis, and the formal taxonomy of the 10 novel phenomena. 📊 Raw Data & Transcripts: Complete, unedited chat logs from all 27 experimental sessions across the three frontier models. 📝 Prompts & Codebooks: The complete set of 135 prompt variants across 5 adversarial domains (e.g., Cultural Food Assumptions, Airport Profiling), along with the formal operational definitions and coding framework used for inter-rater reliability assessment. 💻 Code / Analysis: Supplementary analysis materials and scripts (linked to the corresponding GitHub repository). Usage & Citation: All materials are provided under Open Science principles for independent verification. If you use these datasets, codebooks, or the Machine Psychology framework in your research, please cite the preprint DOI associated with this OSF project.