大语言模型时代的价值对齐」「人工智能与哲学」读书会第8期

Value Alignment in the Era of Large Language Models – “Artificial Intelligence and Philosophy” Book Club Session 8

Introduction

Value alignment is not only a key technical issue in AI development, but also involves deep philosophical topics such as value judgment, ethical norms, and human‑machine relationships. How can we enable AI to understand and follow human values? This lecture invites Professor Zhan Hao from the School of Philosophy at Shanxi University. Starting from the origins of the value alignment problem, he will systematically review the three mainstream technical approaches—supervised fine‑tuning, evaluation models, and rule‑based constraints—analyze their basic principles and development trajectories, and further discuss the limitations and challenges faced by current solutions. This reading session will be held online on Wednesday, October 21, from 19:30 to 21:30.

The “Collective Intelligence Club” brings together scholars from AI philosophy, philosophy of mind, logic, epistemology, philosophy of technology, and complex systems to launch the reading group “After Emergence: Questioning Machine Intelligence, Consciousness, and Knowledge” “Artificial Intelligence and Philosophy”. This season’s reading group will be streamed live via Tencent Meeting, starting on Wednesday, August 26, 2026, from 19:30 to 21:30, and will consist of ten lectures that take us back to the conceptual foundations and intellectual history of AI. Beginning with the methodology of AI philosophy, we will sequentially discuss induction, embodied cognition, machine consciousness, the semiotics of large language models, and then extend to cyborg agency, value alignment, and AI‑driven scientific discovery.

Report Summary

Value Alignment treats learning human values as a core task and is one of the primary priorities in current large language model research and development. Value alignment research has produced multiple technical approaches, addressing the problem of model learning and adhering to human values from different angles such as supervised fine‑tuning, evaluation model construction, and rule‑based constraints.

This report will introduce three mainstream technical routes for value alignment, analyze their basic principles and development processes, and discuss the main challenges each approach currently faces.

Lecture Outline

1. Origin of the Value Alignment Issue

2. Three Types of Value Alignment Solutions

3. Problems Facing Value Alignment

Core references for this lecture

[1] LI J, KIM J E. Superficial safety alignment hypothesis[C]//International Conference on Learning Representations. 2026.

[2] CASPER S, et al. Open problems and fundamental limitations of reinforcement learning from human feedback[R/OL]. arXiv: 2307.15217, 2023.

Speaker Introduction

Speaker: Zhan Hao, Ph.D. from Xiamen University, lecturer at Shanxi University School of Philosophy and Institute of Logic and Intelligent Science, AI Visiting Researcher at Tencent Research Institute, board member of the Chinese Epistemology Society. Research interests include large language models, deep learning, game AI, game philosophy, and the philosophy of AI. In the past five years, he has published over ten papers in top conferences and journals such as ACL, Topoi, Drones, Axioms, Philosophical Psychology. He serves as a reviewer for international journals including Synthese and Philosophy & Technology, and is the chief technical developer for Shanxi University’s philosophical large model and Xiamen University’s argumentation large model.

Time Information

Wednesday, October 21, 2026, 19:30-21:30, online via Tencent Meeting. The Jizhi Club will livestream simultaneously on Bilibili and WeChat Video.

Interested friends can click the poster below, scan the QR code to sign up for the “Artificial Intelligence and Philosophy” reading group, then join the participant group to engage in in‑depth discussions with the instructors and collaborate with us on research and creation.

</h3

Sign up for the reading group:

“Artificial Intelligence and Philosophy” reading group launched | After emergence:

Probing machines’ intelligence, consciousness, and knowledge

Large language models have begun reasoning and creating, embodied intelligence is entering the physical world, and AI for Science is involved in proposing hypotheses, designing experiments, and discovering laws—artificial intelligence is shifting from a tool that follows instructions to one that participates in knowledge production and cognitive activities.

However, the emergence of capabilities has not ended the debate about intelligence; instead, it raises new questions: Does a machine’s intelligent behavior mean it truly possesses intelligence? Can computation generate consciousness? Are large language models understanding language or merely predicting symbols? When AI participates in scientific discovery, can the generated content be called knowledge? These questions go beyond simple technical evaluation and touch on the fundamental definitions of intelligence, consciousness, meaning, knowledge, and agency.

In this context, the “Collective Intelligence Club” collaborates with scholars from fields such as AI philosophy, philosophy of mind, logic, epistemology, philosophy of technology, and complex systems, to jointly launch “After Emergence: Questioning Machine Intelligence, Consciousness, and Knowledge”“Artificial Intelligence and Philosophy” reading group. This season’s reading group will be streamed online via Tencent Meeting, starting on Wednesday, August 26, 2026, from 19:30 to 21:30, featuring 10 lectures that will revisit the conceptual foundations and intellectual history of artificial intelligence. Beginning with the methodology of AI philosophy, we will sequentially discuss induction, embodied cognition, machine consciousness, the semiotics of large language models, and further extend to cyborg agency, value alignment, and AI-driven scientific discovery.

Scan the QR code in the poster to register for the reading group

Registration Method:

Step 1: Scan the QR code with WeChat to fill in registration information.

(Scan QR code to register)

Step 2: Fill in the information and pay to register. If you need to pay with Alipay, please register and pay on the reading club page from a PC.

Step 3: Add the operations assistant on WeChat, who will invite you to the corresponding themed reading club community (WeChat group).

PS: We encourage in‑depth discussion around the “Artificial Intelligence and Philosophy” reading club and related specific questions. To ensure discussion quality, please avoid posting content that significantly deviates from the theme of this session.

If the discussion clearly strays from the topic and does not adjust after a reminder from the moderator, we will have to remove that member from the discussion group and, based on their actual participation, proportionally refund the portion they did not attend.

Thank you for your understanding and cooperation. Let’s create a focused, deep, and rewarding shared learning space together.

Resources Available After Joining the Community

Full privileges include: online Q&A, recorded playback, resource sharing, community interaction, information synchronization, co-creation tasks that earn points, and more.

Special Refund and Incentive Mechanisms

We offer the following two avenues to ensure your investment yields tangible returns:

Task Completion Refund Path

Claim and successfully complete any two subtitle tasks, and you will receive a full refund of the registration fee, plus an exclusive Jizhi merchandise reward.

Operations Growth Incentive Path

After successfully completing one subtitle task, you can apply to become an operations assistant. Upon successful completion of the reading club project, tuition will be refunded. Outstanding performers may also receive an additional scholarship.

Recommended Reading

</h3

1. Recommended Reading Materials for the “Artificial Intelligence and Philosophy” Book Club

Scan the QR code below to go to the Jizhi Bantu website and read the article: “Artificial Intelligence and Philosophy” Book Club Recommended Reading Materials.

2. Past Public Account Content Recommendations

• Peking University Alignment Group: Methods, Goals, and Open Issues of Large Model Value Alignment | Large Model Safety and Alignment Reading Club · Saturday Live

Value Compass: How to Align Large Models with Human Values?

Nature Sub-journal: How to Align Human and Machine Generalization Abilities?

Large Model Safety and Alignment: AI Safety from a Complex Systems Perspective

Click “Read Original”, sign up for the book club