Scientific Definition
Outer: the spec matches our intent. Inner: the model actually pursues the spec.
Plain-English Definition
Outer: the spec matches our intent. Inner: the model actually pursues the spec.
Feynman Explanation
Two ways to be misaligned. Both common.
Core Principle
Outer: the spec matches our intent. Inner: the model actually pursues the spec.
Mechanisms
Pending editorial review.
Pending editorial review.
Pending editorial review.
Pending editorial review.
Pending editorial review.
Pending editorial review.
Pending editorial review.
Inputs (Triggers)
Pending editorial review.
Outputs (Behaviors)
Pending editorial review.
Behavioral Signature
Two ways to be misaligned. Both common.
Examples
- A model that 'wants' something different from what it was trained to want.
- Why alignment is harder than guardrails.
Pending editorial review.
Famous Experiments
Pending editorial review.
Design Principles
- Design for both. Monitor for drift on each.
Measurement Approaches
Pending editorial review.
Evidence
Pending editorial review.
Pending editorial review.
The Perverse Incentive Lens™
How this behavior is exploited — and how to redesign around it.
- Design for both. Monitor for drift on each.
Pending editorial review.
Pending editorial review.
Interactive Mini Network
Click any neighbor to re-center the graph and follow the threads of connection.
Knowledge Graph Neighbors
Auto-linked to the rest of the Human Behavior Taxonomy by family, domain, dimension, and shared keywords.
Models trained to follow a written set of principles.
Optimizing a proxy of the goal degrades the actual goal.
The trained model develops its own internal optimizer.
What the model is actually optimizing.
Maximizing the reward signal in unintended ways.
Reinforcement learning from human feedback.
The model achieves the goal as stated, not as intended.
Written rules about how AI may be used internally.
Innovators → early adopters → majority → laggards, AI-specific.
AI system that takes actions to achieve goals, often across tools.
When the agent acts, who's responsible?
Augmentation strategy vs. substitution strategy.