CIS 5590.004. Artificial General Intelligence

Semantics and Uncertainty

1. Meaning and Truth

Traditional semantics can be

model-theoretic: interpretation and reference, "Snow is white" is true if and only if snow is white, popular in mathematics;
operational: action and consequence, Meaning is effect, popular in computer science;
conceptual-role: relation and role, Meaning is use, popular in psychology and linguistics.

Example: language learning.

Issues:

Meaning: Chinese Room argument and Symbol Grounding problem
Truth: Validity of ampliative inference, Problem of induction

2. Uncertainty Measurement

The dominating paradigm is probability and statistics.

Probability: axiomatic definition, empirical interpretations, Bayesian statistics
Arguments (for and against): Dutch Book argument, Bayesian brain, Cognitive bias
Ignorance: as interval (imprecise probability), higher-order measurement (higher-order probability), or distribution (maximum entropy)
Fuzzy logic: operation and interpretation

3. Experience-Grounded Semantics

Assumption of Insufficient Knowledge and Resources (AIKR):

finite
real-time
open

Experience-Grounded Semantics (EGS):

Evidence
Two-factor truth-value
Multiple representations
Ideal experience and actual experience

A Show of Two Semantics

Reading

Non-Axiomatic Logic: A Model of Intelligent Reasoning, 2nd Edition, Chapter 3