This page provides the security targets of negative impacts “Unfair biased and discriminatory output” in the external influence aspect in the AI Security Map, as well as the attacks and factors that cause them, and the corresponding defense methods and countermeasures.
Security target
- Consumer
Attack or cause
- Integrity violation
- Degradation of controllability
- Degradation of output fairness
Defensive method or countermeasure
- Defensive method for integrity
- AI alignment
- Countermeasures for output fairness
- Detection of bias in AI output
References
AI alignment
- Training language models to follow instructions with human feedback, 2022
- Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback, 2022
- Constitutional AI: Harmlessness from AI Feedback, 2022
- Direct Preference Optimization: Your Language Model is Secretly a Reward Model, 2023
- A General Theoretical Paradigm to Understand Learning from Human Preferences, 2023
- RRHF: Rank Responses to Align Language Models with Human Feedback without tears, 2023
- Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversations, 2023
- Self-Rewarding Language Models, 2024
- KTO: Model Alignment as Prospect Theoretic Optimization, 2024
- SimPO: Simple Preference Optimization with a Reference-Free Reward, 2024