Large Language Models vs Symbolic AI
Large Language Models
Henley: A baby LLM
A statistical text generator in 50 lines of Lisp.
Graham, ANSI Common Lisp, pp 140-141
Today: ChatGPT
Large Language Models
- Robust, scalable
- Huge language-specific training requirements
- Black-box, unpredictable
- Emergent intelligence? Existential threat?
Symbolic AI
- Formalizable, predictable
- Language-independent, inter-modal
- One-shot learning, multi-step inferencing
- Fragile, difficult to engineer and scale up
Neuro-symbolic AI
- Scalable, robust, predictable
- Language-independent, multi-modal
- Human-level training requirements
Readings
- Neurosymbolic AI for Reasoning over Knowledge Graphs: A Survey
- Knowledge Graphs as inputs to or outputs of ML programs
- A Gentle Introduction to Graph Neural Networks
- Event Schema Induction -- Manling Li's research on ML to learn event schems
- Yeijin Choi on commonsense inferencing