Fig. 3: The most-similar bot helped groups achieve better results in landscapes with fewer artificially boosted nouns.
From: Simple autonomous agents can enhance creative semantic discovery by human groups
a Mean cosine similarity between participants’ answer and the target noun across 25 rounds in 5 decoy landscapes. The horizontal line indicates the mean cosine similarity between each of the 18 target nouns and the 20,000 nouns, across conditions. It is apparent that the solo condition (yellow) has the worst performance and the no-bot social condition (black) is an improvement across all landscapes. The bot conditions involving the most-similar bots (red) are helpful, especially so in the narrow landscapes. Error bars indicate standard errors. b Posterior distributions of regression coefficients with the computed highest density intervals. For the dependent variable of the regression analysis, we averaged the cosine similarity between answers and the target for each game. The study incorporated 125 unique groups, each completing 5 games, resulting in 625 data points. The regression model’s independent variables included fixed effects of bot conditions, landscape variables, and their interactions, with the reference variables being the no-bot condition and short/wide landscape.
