HCAI TutorialHCAI 学习指南
A FIELD GUIDE TO HUMAN-CENTERED AI以人为中心的人工智能学习指南
Learn AI.
Design for people.
A project-driven path from foundations to independent HCAI research.
不只学会 AI，
更要学会为人设计。
从基础能力到独立研究，以项目产出驱动的 HCAI 学习路线。
Build an HCAI mindset
Separate three questions: can the model work, is the system usable, and does it improve people's situation? Treat the person–AI–task–context relationship as the unit of design.
Core ideas
- Define success through human goals
- Balance automation and control
- Plan for trust, failure, and recovery
Exercise
- Map stakeholders
- List benefits, misuse, and exclusion
- Rewrite model metrics as human outcomes
Deliverable: a one-page HCAI problem brief.
建立 HCAI 视角
先区分三个问题：模型能不能做、系统是否好用、它是否真正改善人的处境。把“人—AI—任务—情境”的关系作为设计与研究单元。
核心概念
- 以人的目标定义成功
- 平衡自动化与人的控制权
- 考虑信任、失败与恢复
思考练习
- 绘制利益相关者地图
- 列出收益、误用与边缘用户
- 将模型指标改写成人的结果指标
必须产出：一页 HCAI 问题简报。
Tools and research foundations
Learn Python, notebooks, data analysis, statistics, Git, literature search, academic reading, visualization, and reproducible workflows.
Deliverable: a reproducible notebook with a question, cleaning process, analysis, figures, and limitations.
基础工具与研究能力
掌握 Python、Notebook、数据分析、统计、Git、文献检索、学术阅读、可视化与可复现工作流。
必须产出：一个包含问题、数据清洗、分析、图表、结论边界的可复现 Notebook。
Understand people
Study perception, attention, memory, decision-making, emotion, and mental models. Practice interviews, observation, thematic analysis, experiments, surveys, behavioral logs, and research ethics.
Deliverable: 3–5 semi-structured interviews, a coding scheme, themes, and a user task flow.
理解人：认知与研究方法
学习知觉、注意、记忆、决策、情绪与心智模型；练习访谈、观察、主题分析、实验、问卷、行为日志及研究伦理。
必须产出：完成 3–5 次半结构化访谈，形成编码表、需求主题和用户任务流程。
Understand AI capabilities and limits
Learn supervised and unsupervised learning, representation learning, transformers, LLMs, multimodal models, retrieval, agents, and evaluation. Examine where data comes from, who it represents, uncertainty, bias, and failure cost.
Deliverable: an AI baseline plus a model behavior report.
理解 AI：能力、数据与边界
学习监督与无监督学习、表征学习、Transformer、LLM、多模态模型、检索、智能体与评估；分析数据来源、代表性、不确定性、偏差与失败成本。
必须产出：一个小型 AI 基线和一份模型行为报告。
Design human–AI collaboration
Put the model inside an interaction loop: help people express intent, understand output and uncertainty, correct errors, adjust automation, and recover after failure. Prototype before over-engineering.
Deliverable: an interactive prototype, a five-person usability study, and a prioritized issue list.
设计人机协作系统
把模型放进交互闭环：支持用户表达意图、理解输出与不确定性、纠正错误、调整自动化程度，并在失败后恢复。先验证交互价值，再增加工程复杂度。
必须产出：一个交互原型、一次 5 人可用性测试和按严重程度排序的问题清单。
Move into HCAI research
Build an evidence matrix from 20–40 papers. Distinguish technical, design, and knowledge gaps. Align research questions, methods, evidence, ethics, and expected contributions.
Deliverable: a four-page research proposal and a 12-week capstone plan.
进入 HCAI 研究
系统阅读 20–40 篇论文并建立证据矩阵，区分技术缺口、设计缺口与知识缺口，让研究问题、方法、证据、伦理风险和预期贡献严格对应。
必须产出：一份四页研究提案和一个 12 周综合项目计划。
Start with these resources
Google People + AI Guidebook · Microsoft Human–AI Interaction Guidelines · MIT Missing Semester · Dive into Deep Learning · NIST AI RMF