CIS 5590.004. Artificial General Intelligence

Problem Solving and Decision Making

1. Procedural Reasoning

Practical reasoning: how to achieve goals (vs. what to believe).

In Logic Programming (e.g., Prolog), certain propositions get procedural interpretation to become an executable action or a realizable event.

An action changes the objective environment (as in STRIPS) or the subjective experience (as in schema).

Problem solving can accomplished by forward (knowledge expansion) or backward (goal derivation) inference.

Advanced Issues:

Frame problem: to fully specify the preconditions or consequences of an action.
Functional autonomy in goal derivation.
Autonomy and Alignment as conflict requirements.

2. Decision Models

Representative models:

Single step: Decision theory
Multiple steps: Markov Decision Process
Trial and error: Reinforcement learning
Optimality: AIXI

Issues:

Ellsberg paradox: human does not follow decision theory.
Paperclip maximizer: an harmless-looking goal may lead to a disaster.

3. NAL-8 Representation and Inference

Operations and goals
Schema and compound operations
Desire and goal derivation
Decision making

Reading

Non-Axiomatic Logic: A Model of Intelligent Reasoning, 2nd Edition, Chapter 12