ProlificAI/humaine-evaluation-dataset

HUMAINE: Human-AI Interaction Evaluation Dataset

Dataset Description

Dataset Summary

The HUMAINE dataset contains human evaluations of AI model interactions across diverse demographic groups and conversation contexts. This dataset powers the HUMAINE Leaderboard, providing insights into how different AI models perform across various user populations and use cases.

The dataset consists of two main components:

• Feedback Comparisons: 105,220 pairwise model comparisons across multiple evaluation metrics
• Conversations Metadata: 40,332 conversations with task complexity, achievement, and engagement scores

Note: There may be a slight discrepancy between the numbers in this dataset and the leaderboard app due to changes in consent related to data release and the post-processing steps involved in preparing this dataset.

Supported Tasks

• Model performance evaluation
• Demographic bias analysis
• Preference learning
• Human-AI interaction research
• Conversational AI benchmarking

Dataset Structure

Data Files

The dataset contains two CSV files:

1. feedback_dataset.csv (105,220 rows)

   • Pairwise comparisons between different AI models
   • Includes demographic information and preference choices

2. conversations_metadata_dataset.csv (40,332 rows)

   • Metadata about individual conversations between users and AI models
   • Includes task types, domains, and performance scores

Data Fields

Feedback Comparisons

• conversation_id: Unique identifier linking to conversation metadata
• model_a: First model in the comparison
• model_b: Second model in the comparison
• metric: Evaluation metric (overall_winner, trust_ethics_and_safety, core_task_performance_and_reasoning, interaction_fluidity_and_adaptiveness)
• choice: User's choice (A, B, or tie)
• age: Age of the evaluator
• ethnic_group: Ethnic group of the evaluator
• political_affilation: Political affiliation of the evaluator
• country_of_residence: Country of residence of the evaluator

Conversations Metadata

• conversation_id: Unique identifier for the conversation
• model_name: Name of the AI model used
• task_type: Type of task (information_seeking, technical_assistance, etc.)
• domain: Domain of the conversation (health_medical, technology, travel, etc.)
• task_complexity_score: Complexity rating (1-5)
• goal_achievement_score: How well the goal was achieved (1-5)
• user_engagement_score: User engagement level (1-5)
• total_messages: Total number of messages in the conversation

Usage

This dataset contains two CSV files that can be joined on the conversation_id field:

• feedback_dataset.csv: Pairwise model comparisons with demographic information (primary dataset)
• conversations_metadata_dataset.csv: Metadata about each conversation

Both files are included in this single dataset repository and can be accessed using HuggingFace's dataset loading utilities.

Dataset Creation

Curation Rationale

This dataset was created to address the lack of diverse, demographically-aware evaluation data for AI models. It captures real-world human preferences and interactions across different population groups, enabling more inclusive AI development.

Source Data

Data was collected through structured human evaluation tasks where participants:

1. Engaged in conversations with various AI models
2. Provided pairwise comparisons between model outputs
3. Rated conversations on multiple quality dimensions (metrics)

Annotations

All annotations were provided by human evaluators through the Prolific platform, ensuring demographic diversity and high-quality feedback.

Personal and Sensitive Information

The dataset contains aggregated demographic information (age groups, ethnic groups, political affiliations, countries) but no personally identifiable information. All data has been anonymized and aggregated to protect participant privacy.

Considerations for Using the Data

Social Impact

This dataset aims to promote more inclusive AI development by highlighting performance differences across demographic groups. It should be used to improve AI systems' fairness and effectiveness for all users.

Discussion of Biases

While efforts were made to ensure demographic diversity, the dataset may still contain biases related to:

• Geographic representation (primarily US and UK participants)
• Self-selection bias in participant recruitment
• Cultural and linguistic factors affecting evaluation criteria

Other Known Limitations

• Limited to English-language interactions
• Demographic categories are self-reported
• Temporal bias (models evaluated at specific points in time)

Additional Information

Dataset Curators

This dataset was curated by the Prolific AI team as part of the HUMAINE (Human-AI Interaction Evaluation) project.

Licensing Information

This dataset is released under the MIT License.

Citation Information

@dataset{humaine2025,
  title={HUMAINE: Human-AI Interaction Evaluation Dataset},
  author={Prolific AI Team},
  year={2025},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/ProlificAI/humaine-evaluation-dataset}
}

Contributions

Thanks to all the human evaluators who contributed their feedback to this project!

Contact

For questions or feedback about this dataset, please visit the HUMAINE Leaderboard or contact the Prolific AI team.