Inference, not prediction — Prof. Michael I. Jordan on what modern AI is still missing
Summary
Professor Michael Jordan, a highly influential computer scientist, critiques the current trajectory of AI development, particularly the anthropomorphizing of intelligence and the pervasive "AI" and "AGI" buzzwords. He argues that these terms are often misleading, distracting from the practical, engineering-focused tradition of machine learning that has driven industrial success for decades. Jordan emphasizes the need for a "collectivist economic perspective" on AI, advocating for systems that integrate economic thinking, game theory, and robust uncertainty quantification. He highlights the limitations of current large language models (LLMs) in handling real-world uncertainty and their potential to create misaligned incentives, citing examples like Spotify's artist compensation model and the need for data markets that value user contributions and privacy. He also dismisses alarmist superintelligence and extinction narratives as demoralizing science fiction, urging a focus on human-AI collaboration to improve societal systems.
Key takeaway
For AI scientists and directors of AI/ML evaluating strategic directions, you should prioritize integrating economic principles and robust uncertainty quantification into your AI system designs. Avoid the distracting hype of "AGI" and "superintelligence" narratives, instead focusing on building practical, human-centric systems that address real-world problems, foster fair data markets, and enhance human capabilities through thoughtful collaboration, rather than aiming for full automation or replacement.
Key insights
Anthropomorphizing AI and AGI buzzwords distract from practical machine learning and economic system design.
Principles
- AI systems must integrate economic thinking and uncertainty quantification.
- Data markets should incentivize truthful data and respect privacy.
- Human-AI collaboration is more effective than full automation.
Method
Design AI systems using a three-pillar approach: computational thinking, inferential thinking (statistics), and economic thinking (game theory, mechanism design) to address real-world problems and incentives.
In practice
- Implement prediction-powered inference to robustify foundation models.
- Develop data markets with tunable privacy levels and fair compensation.
- Focus on hybrid human-AI systems for complex tasks like air traffic control.
Topics
- AI Anthropomorphism Critique
- Collectivist Economic AI
- Uncertainty Quantification
- Game Theory & Mechanism Design
- Data Market Models
Best for: Research Scientist, AI Scientist, Director of AI/ML, AI Ethicist
Related on AIssential
Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning Street Talk.