让 AI 照见人心讓 AI 照見人心Recovering Human Decision Logic from Behavior

扶光心舍(Cormira)致力于研究人与 AI 的行为及决策,
以预测、模拟、可解释建模为核心,用科研驱动产业智能。
扶光心舍(Cormira)致力於研究人與 AI 的行為及決策,
以預測、模擬、可解釋建模為核心,用科研驅動產業智能。
Cormira builds AI that elicits, formalizes, and reconstructs the decision logic behind human behavior —
through predictive modeling, behavioral simulation, and interpretable inference.

从理解人的行为,到评估 AI 的行为從理解人的行為,到評估 AI 的行為From modeling human behavior to evaluating artificial agents

我们围绕行为的模拟、测量与解释开展研究,并将相关方法应用于心理健康服务和智能体行为治理我們圍繞行為的模擬、測量與解釋開展研究,並將相關方法應用於心理健康服務和智能體行為治理We develop methods for the simulation, measurement, and interpretation of behavior,
extending them to mental health and the governance of autonomous agents.

当前聚焦的研究當前聚焦的研究Current research focus 项目已落地項目已落地Deployed

高保真行为模拟高保真行為模擬High-Fidelity Behavioral Simulation

消费者研究消費者研究Consumer research概念测试概念測試Concept testing

团队基于自研算法,构建可控制、可复现的人类行为模拟环境,用于消费者决策研究、产品概念测试与群体推演。團隊基於自研算法,構建可控制、可復現的人類行為模擬環境,用於消費者決策研究、產品概念測試與羣體推演。Using proprietary algorithms, we construct controlled, reproducible environments for simulating human behavior, supporting consumer-decision research, concept testing, and population-level simulation.

在适用场景下,行为模拟可作为真人实验的补充,降低样本成本、提升研究效率,并增强结果的可复现性。在適用場景下,行為模擬可作為真人實驗的補充,降低樣本成本、提升研究效率,並增強結果的可復現性。Where appropriate, behavioral simulation complements human-subject studies — reducing sample costs, improving efficiency, and strengthening reproducibility.

进展 已与一家亚太区头部保险公司完成项目落地,并持续迭代优化。進展 已與一家亞太區頭部保險公司完成項目落地,並持續迭代優化。Progress Deployed with a leading Asia-Pacific insurer, with ongoing refinement.
正在构建的产品正在構建的產品Product in development 开发验证中開發驗證中In validation

行为智能与心理行為智能與心理Behavioral Intelligence × Mental Health

批量筛查与分级批量篩查與分級Screening & triage医疗级辅助干预醫療級輔助干預Clinical-grade intervention

Cormira 与医疗机构及高校研究团队合作,开发面向心理健康服务与研究流程的 AI 辅助系统。Cormira 與醫療機構及高校研究團隊合作,開發面向心理健康服務與研究流程的 AI 輔助系統。In collaboration with clinical institutions and academic research teams, Cormira develops AI-assisted systems for mental health services and research workflows.

工作覆盖筛查分级、辅助干预、专业督导与数据分析,重点关注专业监督、安全边界和结果审查,并同步探索能力延伸至普通消费者的可能性。工作覆蓋篩查分級、輔助干預、專業督導與數據分析,重點關注專業監督、安全邊界和結果審查,並同步探索能力延伸至普通消費者的可能性。This work spans screening and triage, assisted intervention, clinical supervision, and data analysis, with emphasis on professional oversight, safety boundaries, and outcome review — while examining how these capabilities may extend to consumer-facing applications.

合作机构 浙大一院、湘雅二医院、浙江省精神卫生中心、西湖大学、香港科技大学等。合作機構 浙大一院、湘雅二醫院、浙江省精神衞生中心、西湖大學、香港科技大學等。Partners First Affiliated Hospital of Zhejiang University School of Medicine, Second Xiangya Hospital of Central South University, Zhejiang Mental Health Center, Westlake University, HKUST, and others.
长期研究方向長期研究方向Long-term research direction 持续研究中持續研究中Ongoing research

Agent 治理Agent 治理Agent Governance

智能体行为评测智能體行為評測Behavior evaluation风险预判和治理風險預判和治理Risk anticipation & governance

高保真行为模拟只是起点。随着 Agent 更多参与分析、判断与决策,我们将持续评估其行为表现,解释决策逻辑并识别潜在风险。高保真行為模擬只是起點。隨着 Agent 更多參與分析、判斷與決策,我們將持續評估其行為表現,解釋決策邏輯並識別潛在風險。High-fidelity behavioral simulation is only a starting point. As agents assume greater roles in analysis, judgment, and decision-making, we systematically evaluate their behavior, explain their decision logic, and identify emerging risks.

基于此,我们持续探索可供人类审查、校正与约束的评测方法与治理框架,推动智能体向更安全、可控、可信的方向发展。基於此,我們持續探索可供人類審查、校正與約束的評測方法與治理框架,推動智能體向更安全、可控、可信的方向發展。Building on this, we develop evaluation methods and governance frameworks that enable human review, correction, and constraint of agent behavior — toward agents that are safer, more controllable, and more trustworthy.

目标 建立面向 Agent 经济时代的智能体行为解释与可信度评测体系,并构建相关标准。目標 建立面向 Agent 經濟時代的智能體行為解釋與可信度評測體系,並構建相關標準。Goal Establish a framework for explaining agent behavior and evaluating trustworthiness in the emerging agent economy, and contribute to the standards that govern it.

团队与联系團隊與聯繫Team & Contact

团队背景團隊背景Team background

Cormira 是一支融合计算机科学、心理学与行为经济学的跨学科团队。核心成员曾就读或任职于 CMU、香港科技大学、清华大学、浙江大学、四川大学、西湖大学,以及 Google AI、字节跳动、华为、大疆等高校、研究机构与科技企业。
诚邀更多人才加入。
Cormira 是一支融合計算機科學、心理學與行為經濟學的跨學科團隊。核心成員曾就讀或任職於 CMU、香港科技大學、清華大學、浙江大學、四川大學、西湖大學,以及 Google AI、字節跳動、華為、大疆等高校、研究機構與科技企業。
誠邀更多人才加入。
Cormira is an interdisciplinary research team spanning computer science, psychology, and behavioral economics, with members who have studied or worked at Carnegie Mellon University, HKUST, Tsinghua University, Zhejiang University, Sichuan University, and Westlake University, as well as Google AI, ByteDance, Huawei, and DJI, among other institutions.
We welcome new talent to join us.

团队聚焦因果推断、大模型对齐、多智能体系统与 AI 辅助心理干预,相关成果发表于 Marketing Science、JMR、ISR 等学术期刊及 ACL、NeurIPS 等人工智能会议,并与国内多家三甲医院开展联合研究与实验验证。
欢迎合作洽谈。
團隊聚焦因果推斷、大模型對齊、多智能體系統與 AI 輔助心理干預,相關成果發表於 Marketing Science、JMR、ISR 等學術期刊及 ACL、NeurIPS 等人工智能會議,並與國內多家三甲醫院開展聯合研究與實驗驗證。
歡迎合作洽談。
Our research spans causal inference, LLM alignment, multi-agent systems, and AI-assisted psychological intervention. This work has appeared in journals including Marketing Science, JMR, and ISR, and at conferences such as ACL and NeurIPS, alongside clinical research and validation studies with leading hospitals in China.
We're always open to new collaborations.

我们坚信,研究的真正价值不止于提出新的解释,也在于回答真实的问题。
这些问题横跨行为、技术与应用,仍有许多尚未被解答。
我們堅信,研究的真正價值不止於提出新的解釋,也在於回答真實的問題。
這些問題橫跨行為、技術與應用,仍有許多尚未被解答。
We hold that the value of research lies not only in offering new explanations, but in addressing real questions —
questions that span behavior, technology, and practice, many of which remain open.

Cormira 期待与全球的优秀伙伴一起,找到更好的答案。Cormira 期待與全球的優秀夥伴一起,找到更好的答案。Cormira looks forward to working with great partners worldwide to find better answers.