China's Perspective on Global Development Initiative

Introduction

• Speaker: Ming Zhang
• Affiliation: Computer Science and AI Institute
• Topic: Progress and Concepts in Artificial Intelligence (AI)

Structure of the Lecture

1. Fundamental Concepts of AI
2. Current Challenging Topics
3. Recent AI Research and Applications in China

Fundamental Concepts of AI

Types of Knowledge:

1. Precise Knowledge:
   • Unique answers (e.g., 1 + 1 = 2)
   • Limited in universe
2. Common Sense Knowledge:
   • General knowledge with exceptions (e.g., birds can fly, but not ostriches)
   • Difficult for AI to manage exceptions
3. Uncertain Knowledge:
   • Probabilistic (e.g., weather forecasts)
   • Probability-based approaches are successful

Knowledge Acquisition

• Learning by practice and examples (e.g., riding a bicycle, identifying cats)
• Machine learning involves learning from features
• Neural networks were proposed over 50 years ago but gained success with modern computation resources

Key AI Concepts

1. Input: Data provided to AI
2. Representation: How data is described in the system
3. Output: The intended results
4. Algorithm: The model used for learning or problem-solving
5. Performance: Evaluation of results

Current AI Challenges

1. Confidence: Trust in AI results and systems
2. Ethics: Ethical principles governing AI actions
3. Fluency and Accuracy: Balancing between generating fluent but accurate responses
4. Explainability and Interpretation: Explaining AI decisions to users and understanding system processes
5. Fairness: Ensuring equal and unbiased AI results
6. Privacy: Protecting user data and information
7. Effectiveness and Efficiency: Optimizing resources in AI development
8. Philosophy: Underlying principles and methodologies in AI

Recent AI Progress in China

Tsinghua University AI Institute

• Focus on eight research directions
• Achievements in language processing, music generation, robot planning, autonomous driving, machine learning, etc.

Fundamental AI and Machine Learning

• Bayesian methods, deep learning, efficient machine learning, adversarial learning, reinforcement learning, brain-inspired AI
• Developed resources such as Zhu Sun GPU library
• Advances in diffusion models and adversarial attack/defense

Interdisciplinary Research

• Cognitive science applied to user modeling for video recommendations
• Immersion defined as a state of flow with characteristics such as lack of time awareness
• Research shows mixing personalized and random recommendations can enhance immersion

Large-Language Models

• Usage and evaluation among students
• Frequent use for task assistance, problem-solving, idea generation
• Satisfaction varies by task type; technical issues like hallucinations noted

AI in Legal and Scientific Fields

• Legal foundation models developed with legal knowledge integration
• AI in chemistry and drug discovery, outpacing human experts in some areas

Robotics and AI Applications

• Robots for music playing, service assistance, autonomous vehicles
• Rehabilitation robots aiding in therapy

Conclusion

• Open question: Will AI rule the world and defeat human beings?
• Encouragement to think about the future of AI and its ethical implications

These notes summarize the key points and structure of the lecture on AI progress, challenges, and applications from a Chinese perspective, particularly focusing on Tsinghua University's contributions.