Decision Making

Decision making is the process of choosing the "best" action among several alternatives. In AGI, this must happen under conditions of Uncertainty and Limited Resources.

The Decision Framework

How does an AGI decide between Action A and Action B?

1. Utility Maximization

Assigning a "Value" to each possible outcome.

• Formula: $Value = Probability \times Utility - Cost$
• AGI twist: Utility isn't just about a score; it's about how much the outcome satisfies the agent's current Drives.

2. Multi-Criteria Decisions

Often, goals conflict. Should the agent be Fast or Safe?

• PRIMUS handles this by weighting different drives. If the agent is low on power, "Efficiency" gains weight over "Exploration."

Decision Flow

graph TD
    Options[Identify Possible Actions] --> Evaluate[Evaluate Outcomes]
    
    subgraph Evaluation
    Prob[Probabilistic Success]
    Gain[Goal Progress]
    Risk[Potential Harm]
    end
    
    Prob --> WeightedVal[Weighted Value]
    Gain --> WeightedVal
    Risk --> WeightedVal
    
    WeightedVal --> Pick[Select Best Action]

Bounded Rationality

In the real world, an AGI cannot think forever. It must make a "good enough" decision within a time limit. This is called Satisficing.

• If a high-utility action is found quickly, the agent might stop searching and act immediately rather than seeking the absolute perfect move.

Next: Example Agent