Related Experiment Video
Updated: Sep 28, 2025

Characterization of the Sense of Agency over the Actions of Neural-machine Interface-operated Prostheses
Published on: January 7, 2019
What values should an agent align with?: An empirical comparison of general and context-specific values
Enrico Liscio1, Michiel van der Meer1,2, Luciano C Siebert1
1Delft University of Technology: Technische Universiteit Delft, Delft, The Netherlands.
Identifying context-specific values is crucial for understanding human decisions and aligning AI. The Axies methodology, combining human input and AI, effectively extracts these values from text, proving more specific and suitable than general value systems.
Area of Science:
- Social Sciences
- Computer Science
- Artificial Intelligence
Background:
- Human behavior and cooperation are driven by values.
- General value systems (e.g., Schwartz) lack context specificity.
- Context-specific values are needed for understanding decisions and aligning AI.
Purpose of the Study:
- To introduce Axies, a hybrid human-AI methodology for identifying context-specific values.
- To simplify value identification through guided annotation of value-laden text.
- To leverage Natural Language Processing (NLP) for systematic value extraction.
Main Methods:
- Axies methodology: a hybrid approach using human annotators and AI.
- Utilized value-laden text corpora and NLP tools.
- User study with 80 subjects evaluating context-specific values for Covid-19 measures and sustainable energy.
Main Results:
- Axies generated value lists that were more context-specific than general values.
- The identified values were more suitable for annotation tasks.
- The methodology proved independent of the individuals applying it.
Conclusions:
- Axies is an effective method for identifying context-specific human values.
- This approach enhances the ability to engineer AI systems that align with human values.
- Context-specific values are essential for nuanced understanding and AI alignment.
More Related Videos
07:59Author Spotlight: Alignment of Synchronized Time-Series Data Using the Characterizing Loss of Cell Cycle Synchrony Model for Cross-Experiment Comparisons
Published on: June 9, 2023
07:43Author Spotlight: A Novel Setup to Conduct Naturalistic Laboratory Experiments with Real Human Actors in Scenarios
Published on: August 4, 2023
Related Concept Videos
Impression Management Techniques III: Aligning Actions
The Anchoring-and-Adjustment Heuristic
Causes of Similarity-Dissimilarity Effect
Cultural Influences on Personality
Social Proof
Self Within Cultural Contexts