This site is a work in progress and has not been widely shared. Content may contain errors. Feedback is welcome.
This site is undergoing review. Some annotations were human-generated, some AI-generated — all are being verified.
Back to papers

Simple Autonomous Agents Can Enhance Creative Semantic Discovery by Human Groups

Ueshima, Jones, Christakis

2024Nature Communications4 citations
Experimental evidenceInterdisciplinaryCausal
Machine Learning (pre-LLM)Human-AI collaborationCollective intelligence / teamsDecision-making
View RepositoryDOI: 10.1038/s41467-024-49528-y
Summary

Ueshima, Jones, and Christakis conduct online lab experiments with 1,875 participants in networked groups (with and without simple autonomous bots) searching a semantic space of 20,000 nouns to study how bot strategies affect collective creativity and idea discovery.

Main Finding

Groups outperform isolated individuals in semantic search, and adding bots that share the most similar noun among neighbors improves group performance by 0.56 standard deviations in easier landscapes (no-decoy and narrow conditions), but not in landscapes with many artificially boosted decoy words.

Primary Datasets

Experiment-generated data from online lab experiment via Amazon Mechanical Turk; word2vec semantic embeddings

Secondary Datasets

None

Key Methods
Lab experiment with 125 groups of 15 participants embedded in Erdos-Rényi networks, playing 5 sequential word-search games across different bot conditions (most-similar, least-similar, random, no-bot, solo) and decoy landscapes; Bayesian multilevel regression analysis.
Sample Period
2023
Geographic Coverage
US (Amazon Mechanical Turk)
Sample Size
1,875 participants in 125 groups of 15 people; 625 group-game observations
Level of Analysis
Individual, Firm
Occupation Classification
None
Industry Classification
None
Replication Package
Yes
Notes
Nature Communications, vol. 15, no. 1, p. 5212 [Claude classification]: This is a lab experiment studying collective creativity and idea discovery in semantic space, not a traditional labor market study. The bots use classical NLP (word2vec) rather than LLMs. The AI technology studied is simple autonomous agents using word embeddings, coded as Machine Learning (pre-LLM) since word2vec is a classic NLP technique from 2013. The paper is interdisciplinary (sociology, network science, psychology). Level of analysis is both Individual (participant-level outcomes) and Firm/Group (group-level performance). Published in Nature Communications, vol. 15, no. 1, p. 5212, June 2024. [Claude classification]: This is a lab experiment studying collective creativity and idea discovery in semantic space, not a traditional labor market study. The bots use classical NLP (word2vec) rather than LLMs. The AI technology studied is simple autonomous agents using word embeddings, coded as Machine Learning (pre-LLM) since word2vec is a classic NLP technique from 2013. The paper is interdisciplinary (sociology, network science, psychology). Level of analysis is both Individual (participant-level outcomes) and Firm/Group (group-level performance). Published in Nature Communications, vol. 15, no. 1, p. 5212, June 2024. [Claude classification]: This is a lab experiment studying collective creativity and idea discovery in semantic space, not a traditional labor market study. The bots use classical NLP (word2vec) rather than LLMs. The AI technology studied is simple autonomous agents using word embeddings, coded as Machine Learning (pre-LLM) since word2vec is a classic NLP technique from 2013. The paper is interdisciplinary (sociology, network science, psychology). Level of analysis is both Individual (participant-level outcomes) and Firm/Group (group-level performance). Published in Nature Communications, vol. 15, no. 1, p. 5212, June 2024. [Claude classification]: This is a lab experiment studying collective creativity and idea discovery in semantic space, not a traditional labor market study. The bots use classical NLP (word2vec) rather than LLMs. The AI technology studied is simple autonomous agents using word embeddings, coded as Machine Learning (pre-LLM) since word2vec is a classic NLP technique from 2013. The paper is interdisciplinary (sociology, network science, psychology). Level of analysis is both Individual (participant-level outcomes) and Firm/Group (group-level performance). Published in Nature Communications, vol. 15, no. 1, p. 5212, June 2024. [Claude classification]: This is a lab experiment studying collective creativity and idea discovery in semantic space, not a traditional labor market study. The bots use classical NLP (word2vec) rather than LLMs. The AI technology studied is simple autonomous agents using word embeddings, coded as Machine Learning (pre-LLM) since word2vec is a classic NLP technique from 2013. The paper is interdisciplinary (sociology, network science, psychology). Level of analysis is both Individual (participant-level outcomes) and Firm/Group (group-level performance). Published in Nature Communications, vol. 15, no. 1, p. 5212, June 2024. [Claude classification]: This is a lab experiment studying collective creativity and idea discovery in semantic space, not a traditional labor market study. The bots use classical NLP (word2vec) rather than LLMs. The AI technology studied is simple autonomous agents using word embeddings, coded as Machine Learning (pre-LLM) since word2vec is a classic NLP technique from 2013. The paper is interdisciplinary (sociology, network science, psychology). Level of analysis is both Individual (participant-level outcomes) and Firm/Group (group-level performance). Published in Nature Communications, vol. 15, no. 1, p. 5212, June 2024. [Claude classification]: This is a lab experiment studying collective creativity and idea discovery in semantic space, not a traditional labor market study. The bots use classical NLP (word2vec) rather than LLMs. The AI technology studied is simple autonomous agents using word embeddings, coded as Machine Learning (pre-LLM) since word2vec is a classic NLP technique from 2013. The paper is interdisciplinary (sociology, network science, psychology). Level of analysis is both Individual (participant-level outcomes) and Firm/Group (group-level performance). Published in Nature Communications, vol. 15, no. 1, p. 5212, June 2024. [Claude classification]: This is a lab experiment studying collective creativity and idea discovery in semantic space, not a traditional labor market study. The bots use classical NLP (word2vec) rather than LLMs. The AI technology studied is simple autonomous agents using word embeddings, coded as Machine Learning (pre-LLM) since word2vec is a classic NLP technique from 2013. The paper is interdisciplinary (sociology, network science, psychology). Level of analysis is both Individual (participant-level outcomes) and Firm/Group (group-level performance). Published in Nature Communications, vol. 15, no. 1, p. 5212, June 2024. [Claude classification]: This is a lab experiment studying collective creativity and idea discovery in semantic space, not a traditional labor market study. The bots use classical NLP (word2vec) rather than LLMs. The AI technology studied is simple autonomous agents using word embeddings, coded as Machine Learning (pre-LLM) since word2vec is a classic NLP technique from 2013. The paper is interdisciplinary (sociology, network science, psychology). Level of analysis is both Individual (participant-level outcomes) and Firm/Group (group-level performance). Published in Nature Communications, vol. 15, no. 1, p. 5212, June 2024. [Claude classification]: This is a lab experiment studying collective creativity and idea discovery in semantic space, not a traditional labor market study. The bots use classical NLP (word2vec) rather than LLMs. The AI technology studied is simple autonomous agents using word embeddings, coded as Machine Learning (pre-LLM) since word2vec is a classic NLP technique from 2013. The paper is interdisciplinary (sociology, network science, psychology). Level of analysis is both Individual (participant-level outcomes) and Firm/Group (group-level performance). Published in Nature Communications, vol. 15, no. 1, p. 5212, June 2024. [Claude classification]: This is a lab experiment studying collective creativity and idea discovery in semantic space, not a traditional labor market study. The bots use classical NLP (word2vec) rather than LLMs. The AI technology studied is simple autonomous agents using word embeddings, coded as Machine Learning (pre-LLM) since word2vec is a classic NLP technique from 2013. The paper is interdisciplinary (sociology, network science, psychology). Level of analysis is both Individual (participant-level outcomes) and Firm/Group (group-level performance). Published in Nature Communications, vol. 15, no. 1, p. 5212, June 2024.