Definition
Conceptual Breakthroughs are the fundamental scientific and mathematical advances required to achieve human-level (or superintelligent) AI. Stuart Russell argues that simply scaling up computing power or data will not suffice; we must first solve several specific “bottlenecks” in AI capability.
Why It Matters
It refocuses AI research on solving core logic and knowledge representation problems rather than just scaling up raw compute.
Core Concepts
- Language and Common Sense: The ability for a machine to truly understand the content and context of language, allowing it to read and acquire the vast store of human knowledge.
- Cumulative Learning of Theories: The ability for a machine to generate new concepts and theories (e.g., mass, acceleration, charge) from data, rather than just fitting curves to existing variables (feature engineering).
- Discovering Actions: The ability for a machine to discover and name its own high-level, abstract actions (e.g., “standing up,” “typing”), allowing it to plan over long time scales by hierarchical search.
- Managing Mental Activity: “Cognitive efficiency”—the ability for a machine to reason about its own thinking process, choosing which computations are most likely to improve its decision quality.