How does AlphaGo work?
AlphaGo relies on two neural networks:
- An evaluation neural network to assess the quality of moves
- Monte Carlo tree search for random simulation
These networks are trained through self-play and optimization.
What is the essence of learning?
The essence of learning can be summarized as:
- Pattern recognition
- Pattern transfer
Learning is about discovering patterns and transferring them to new contexts.
What are some core problems with current AI?
Some core problems with current AI:
- Composability problem: AI systems cannot decompose and recompose problems like humans based on relationships between components.
- Long tail problem: Too many edge cases in the real world not seen by AI systems during training.
- Uninterpretability: AI systems cannot explain reasoning like humans, leading to lack of trust.
- Insufficient language understanding: Language requires common sense to truly understand, still lacking in AI.
- Data-driven mechanics: AI relies on training data, more like evolution than active learning.
- Clever Hans problem: AI may just be picking up on surface patterns in data, not true understanding.
What are key differences between human learning and machine learning?
Human learning vs machine learning:
- Humans actively ask questions, machines passively ingest data
- Humans abstract concepts and attributes, machines learn concrete examples
- Humans have open categories, machines only see fixed categories from training data
What is the essence of language?
The essence of language is symbolized thought. Language understanding requires common sense and abstract reasoning, not just statistical patterns.