CompanionX moves AI beyond narrow task solving toward human-like collaboration. While modern models excel at mathematics and coding, qualities such as social intelligence, empathy, and benevolence remain underexplored. CompanionX develops agents that understand, respond to, and cooperate with people authentically and proactively, forming personalized minds and communicating with emotional competence. These digital companions will support social interaction, customer service, digital coworkers, personal coaching, AI NPCs, and large-scale societal simulations—deepening trust, engagement, and real-world value.
To realize this vision, CompanionX introduces a systematic tech stack for evaluating and enhancing human-like (anthropomorphic) intelligence: foundational social-cognitive capabilities and rigorous evaluations; an agentic framework for human-like task execution and collaboration; social simulation for self-play, safe experimentation, and continual improvement; and Companion Guard, a safety and ethics layer ensuring prosocial behavior, transparency, privacy, and regulatory readiness. Through interdisciplinary research and comprehensive evaluation, CompanionX advances trustworthy, collaborative, and socially adept AI that bridges technology and genuine human experience.
Representative Publications:
- Pretraining context compressor for large language models with embedding-based memory. ACL 2025. (opens in new tab)
- SocialCC: Interactive Evaluation for Cultural Competence in Language Agents. ACL 2025. (opens in new tab)
- MotiveBench: How Far Are We From Human-Like Motivational Reasoning in Large Language Models? ACL 2025 Findings. (opens in new tab)
- Unveiling the Learning Mind of Language Models: A Cognitive Framework and Empirical Study. NeurIPS 2025. (opens in new tab)
- CharacterBox: Evaluating the Role-Playing Capabilities of LLMs in Text-based Virtual Worlds. NAACL 2025. (opens in new tab)
- HumanLLM: Towards Personalized Understanding and Simulation of Human Nature. KDD 2026. To appear.
- Population-Aligned Persona Generation for LLM-based Social Simulation. Preprint. (opens in new tab)
Open-source contributions:
https://github.com/microsoft/AnthropomorphicIntelligence (opens in new tab)
Other achievements:
Our v-team (Yitian Huang, Yuxuan Lei, Jianxun Lian) won one first place, one second place, and two third place awards in CPDC Challenge 2025 (opens in new tab), a competition focused on AI NPC. Interactive AI NPCs Powered by LLMs: Technical Report for the CPDC Challenge 2025 (opens in new tab).