Simple autonomous agents can enhance creative semantic discovery by human groups

June 20, 2024   Matt Jones

  • Group Decision Making
  • Academic Articles


Innovation is challenging, and theory and experiments indicate that groups may be better able to identify and preserve innovations than individuals. But innovation within groups faces its own challenges, including groupthink and truncated diffusion. We performed experiments involving a game in which people search for ideas in various conditions: alone, in networked social groups, or in networked groups featuring autonomous agents (bots). The objective was to search a semantic space of 20,000 nouns with defined similarities for an arbitrary noun with the highest point value. Participants (N = 1875) were embedded in networks (n = 125) of 15 nodes to which we sometimes added 2 bots. The bots had 3 possible strategies: they shared a random noun generated by their immediate neighbors, or a noun most similar from among those identified, or a noun least similar. We first confirm that groups are better able to explore a semantic space than isolated individuals. Then we show that when bots that share the most similar noun operate in groups facing a semantic space that is relatively easy to navigate, group performance is superior. Simple autonomous agents with interpretable behavior can affect the capacity for creative discovery of human groups.


The discovery of innovative ideas can enhance the immediate welfare of a population and even modify the course of evolution1,2,3. However, finding such valuable ideas often involves exploring a large pool of possibilities – which can be a challenging process for both individuals and groups. The primary roadblock to finding good ideas is normally not that innovations are hard to evaluate, but that coming up with an original, paradigm-shifting idea that pushes the boundary of the space of available ideas is difficult. Ironically, this is a challenge that being in groups can both mitigate and amplify. Moreover, since simple autonomous agents can alter group behavior in a variety of ways4,5,6,7,8, such agents might also affect the creative capacity of groups.

For the emergence of collective intelligence, prior work has highlighted the importance of both independence and inter-dependence among group members9,10,11,12,13. The presence of too much interdependence within a group can result in a quick convergence on an inferior idea (e.g., groupthink14). Such social herding has been shown to have negative effects on collective intelligence15,16. On the other hand, if there is not a focused group whose members draw inspiration from each other, the lack of inter-dependence can lead to uncoordinated and inefficient exploration of ideas and a failure to exploit any beneficial innovations once they are discovered.

Prior work on social learning within human groups has focused on critical factors including, for example, network structure17,18,19,20, learning strategy21,22,23,24, and group size25,26,27. However, prior experimental studies of networked collective decision-making have generally neglected the critical issue of relationships among candidate ideas – for instance, semantic similarity between ideas in an idea space. In daily life, similar ideas tend to have similar value and also tend to be easier to discover via marginal improvements to existing ideas. Groups can follow a strategy whereby members use the ideas proposed by their neighbors to help guide their next attempt. It is, therefore, important to understand the strategies that groups can adopt in order to enhance collective creativity in such a situation.

The interplay between independence and interdependence in idea sharing can also inform the development of intervention strategies. For instance, a group producing overly similar ideas could benefit from an intervention that promotes independence in idea generation, thus reducing idea similarity and facilitating the discovery of novel ideas, while a group that is already effectively exploring solutions might benefit from additional sharing of ideas to promote exploitation of high-value regions of the semantic space.

Here, we first develop a word search game mimicking such challenges. Then, we test it in groups of isolated individuals and in groups that can share information in a social network; and we show that social information helps groups explore the idea space. Finally, we demonstrate how the use of simple autonomous agents (bots) can affect collective idea exploration. We test several different potential group-level interventions involving such simple bots. We also explore the impact of making the problem harder to solve by adding a variety of decoys to the idea landscape.

Despite the ongoing transformation of social and computational science research by large language models28, here we focus on simple autonomous agents that work with classic natural language processing techniques and that are thus relatively transparent in nature4,29. Doing so allows us to have full control over how our AI-bots intervene in human groups; to obtain more interpretability in what the bots are doing; and to focus on human creativity rather than AI capability per se. Nevertheless, this methodology also sheds light on how more complex forms of AI might shape the behavior of human groups.

In total, we show that adding simple bots to networked human groups has a notable impact on the ability of groups to find rewarding regions of semantic space, particularly when sharing similar ideas in less challenging landscapes.

Read the Full Report >>