

94 | Stuart Russell on Making Artificial Intelligence Compatible with Humans
7 snips Apr 27, 2020
In this thought-provoking conversation, Stuart Russell, a distinguished professor of computer science at UC Berkeley and co-founder of the Center for Human-Compatible Artificial Intelligence, discusses the complexities of artificial intelligence and its alignment with human values. He explores the need for AI to learn from human behavior rather than imposing rigid goals. Russell also addresses the existential risks of superintelligent AI, the challenges of decision-making, and the transformative potential of AI in enhancing civilization, calling for a flexible approach to programming these systems.
AI Snips
Chapters
Books
Transcript
Episode notes
AI's Core Function
- Artificial intelligence builds systems with human-assigned objectives.
- These systems then devise ways to achieve those objectives, like maximizing rewards.
Continuum of Intelligence
- A spectrum of intelligence exists, from simple thermostats to complex humans.
- The complexity arises from the environment and the objective's intricacy.
Go vs. Driving
- Go has explicit rules and a finite state space, unlike driving, where rules and objectives are unclear.
- Driving involves unknown human behavior and undefined 'good driving' standards.