# Habermas Machine Dataset: Human Deliberation Opinions, Group Statements and Preference Rankings from AI-Mediated Deliberation (Google DeepMind, Science 2024)

## Facts

- Provider: Google DeepMind (Michael Henry Tessler, Michiel A. Bakker, Christopher Summerfield et al.)
- License: CC-BY-4.0 (https://github.com/google-deepmind/habermas_machine)
- Format: Parquet (in sampled files)
- Stored at: Google Cloud Storage, gs://habermas_machine/datasets (us)
- Access: public, readable without an account
- Category: Machine Learning & AI
- Source: https://www.science.org/doi/10.1126/science.adq2852
- Page: https://scigantic.com/archive/ef006585-83bc-5d97-8926-54d5254ce0e3
- Keywords: Machine Learning & AI, Parquet, Google Cloud Storage

## About

Data from experiments in which an AI mediator, the Habermas Machine, helped small groups of people find common ground on social and political questions, reported in Science 2024. Participants (N = 5,734 in the study) wrote personal opinions on a question, the system generated candidate group statements, participants ranked and critiqued them, and the statement was refined over rounds; AI-generated statements were compared with ones written by human mediators, and the findings were replicated in a virtual citizens' assembly with a demographically representative UK sample. The datasets/ folder holds five Parquet tables (about 450 MB): hm_all_candidate_comparisons (pairwise comparisons of candidate statements), hm_all_final_preference_rankings, hm_all_position_statement_ratings (participants' agreement with statements), hm_all_question_importances_ratings and hm_all_round_survey_responses (per-round survey answers). Read with pandas.read_parquet. Used to study opinion change, consensus formation and deliberative democracy, and to train and evaluate reward models and statement generators for collective decision making. License: CC BY 4.0 for the materials (Google DeepMind). Cite Tessler, Bakker et al., AI can help humans find common ground in democratic deliberation, Science 2024 (doi 10.1126/science.adq2852).

## Load the data

```python
import gcsfs
fs = gcsfs.GCSFileSystem(token='anon')
fs.ls('habermas_machine/datasets/', detail=False)[:10]
```

Lists the files with anonymous access. Runs on any machine, with no cloud account.

## Data structure

File types in sampled files: Parquet

Example file: `datasets/`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/ef006585-83bc-5d97-8926-54d5254ce0e3
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/ef006585-83bc-5d97-8926-54d5254ce0e3
- Over MCP: `get_schema_card` with id `ef006585-83bc-5d97-8926-54d5254ce0e3` at https://api.scigantic.com/mcp

## Related topics

- [Parquet datasets](https://scigantic.com/topics/parquet)
- [Machine Learning & AI datasets](https://scigantic.com/topics/machine-learning-and-ai)
