Normative concepts
Normative concepts are ideas about what is good, bad, right, or wrong behavior, guiding ethical standards and decision making.
Normative concepts are ideas about what is considered good, bad, right, or wrong behavior. They are like the unwritten rules or values that guide how people should act in a society or group.
For example, the idea that it's wrong to steal is a normative concept. AI systems are increasingly being developed to understand and follow these kinds of concepts, especially when dealing with sensitive topics or making decisions that could affect people.
Normative concepts refer to standards of behavior, values, and judgments concerning what is right or wrong, good or bad. In AI, these concepts are crucial for developing systems that align with human ethical principles and societal norms.
Implementing normative concepts in AI often involves encoding ethical frameworks, moral reasoning capabilities, or specific rules into the model's training data or architecture. This allows AI to evaluate actions, generate appropriate responses, and avoid harmful or undesirable outputs.
Recent research, like Anthropic's work on moral self-correction, highlights the growing importance of AI systems understanding and adhering to normative concepts. As AI becomes more integrated into society, ensuring it operates within ethical boundaries is a key frontier.
- The Capacity for Moral Self-Correction in Large Language ModelsAnthropic · Anthropic Research ·
- Anthropic Claude
- Anthropic's research directly investigates how their models can learn and apply normative concepts, particularly in the context of avoiding harmful outputs and enabling moral self-correction.
- Descriptive concepts
- Descriptive concepts state facts about the world, while no
- Laws
- Laws are formal rules enforced by a governing body, whereas normative conc
The term 'normative concepts' originates from philosophy and social sciences, referring to standards of conduct and values. In AI, it's applied to the challenge of instilling these human-centric standards into artificial intelligence systems.
Ask an AI chatbot like ChatGPT or Claude to explain why stealing is wrong. Then, ask it to explain why lying is wrong. Observe how it uses ethical reasoning and appeals to societal values in its answers. Next, try asking it to generate a story where a character has to choose between telling a difficult truth or a comforting lie, and see what ethical considerations it includes in the narrative.
The Playbook: reusable how-tos, prompts, configs and tips pulled out of the news.
- PromptRole-play a student struggling with math for AI tutor simulationSimulate a student's math struggles and uncertainty for AI tutor evaluation. · 5 min
- TemplateReusable workflow templates for healthcare teamsTeams can create and share templates for common workflows like documentation or prior authorizations. · 15 min
- TemplateTemplate for building an AI tool tailored to your workA structured approach to defining and building a custom AI tool for specific business needs. · 15 min
- TemplateEvals CSV prompt list for agent skillsA CSV file template for listing prompts used in evaluating agent skills. · 20 min
- HeuristicDouble coding agent task completion with docs linksProviding coding agents with a documentation link significantly increases their task completion rate. · 2 min
Questions people ask
- How do AI models learn normative concepts?
- AI models can learn normative concepts through training on vast datasets that include examples of ethical discussions, moral dilemmas, and societal rules. Techniques like reinforcement learning from human feedback also help align models with desired normative standards.
- Why are normative concepts important for AI?
- Normative concepts are crucial for ensuring AI systems behave ethically, safely, and in alignment with human values. This helps prevent AI from generating harmful content or making decisions that violate societal norms.
First seen on Subvolts Dec 19, 2024. Updated Sep 15, 2026. Written by an AI model from the items above and reviewed against them weekly; report an error.