The actual contents of minds are incredibly complex, and attempts to find simple ways to represent or reason about these contents are misguided as they will fail to capture the full richness and intricacy of mental processes.
This complexity is inherent and cannot be easily reduced or simplified.
Minds tend to map aspects of reality such as:
- Space: The physical environment we inhabit is intricate and can be represented and navigated in countless ways.
- Objects: The world is filled with diverse objects that have various properties, interactions, and relationships.
- Multiple agents: Other individuals in the environment have their own complex minds, behaviors, and intentions.
- Symmetries: Patterns, regularities, and symmetries in the world can be subtle, multifaceted, and challenging to identify and represent.
Instead of attempting to build these complexities into AI systems directly, researchers should focus on developing meta-methods capable of discovering and capturing this complexity autonomously. AI agents should be designed to discover like humans do, rather than being pre-loaded with human discoveries, as the latter approach hinders the development of effective discovery processes.
Source: The Bitter Lesson