A recent study from the World Bank reveals that the dissemination of scientific ideas within large organizations, particularly concerning generative artificial intelligence, is far from uniform. The research indicates that the initial recipient of information can significantly influence how widely it is shared.
Conducted by the World Bank’s Development Research Group in collaboration with the University of Warwick and Warwick Business School, the study explored not just whether AI enhances productivity and creativity, but also how organizations learn from new information. This investigation took place through a large-scale field experiment involving over 6,000 employees across 511 divisions at the World Bank’s headquarters in Washington, D.C.
In the experiment, one employee from each division was randomly chosen to act as a “seed” and received summaries of leading academic studies on the implications of ChatGPT in the workplace. These summaries highlighted findings that indicated ChatGPT could substantially improve writing productivity, quality, and even enhance creativity in storytelling tasks. Following the receipt of the research summaries, the seeds completed a survey and were provided with an infographic and a link to the full research paper. Researchers later monitored the intention of these seeds to share the findings, colleagues’ engagement with the materials, and the retention of information weeks later.
The Influence of Hierarchy
A key finding of the research was the pronounced impact of organizational hierarchy on information dissemination. When the seed was a senior employee, such as a Senior Economist or Director, the research findings were more likely to be shared and discussed within their divisions. Senior seeds reported a higher intent to share the information, and divisions with such seeds showed greater engagement with the study materials. Colleagues of senior seeds also exhibited a better recall of at least one detail from the research weeks later.
Despite these findings, overall recall levels within the organization remained low, with only a small percentage of colleagues able to remember even one notable aspect of the studies. Nevertheless, the difference in the spread of information between senior and junior seeds was significant, indicating that ideas are more likely to gain traction when they are initiated by higher-ranking staff.
The researchers further examined whether social norms or credibility signals might influence the sharing of evidence. Some seeds were informed that most of their colleagues actively used ChatGPT, while others were told that its adoption was relatively low. Although these messages impacted perceptions of peer behavior, they did not translate into increased sharing of the evidence or improved recall among colleagues.
Similarly, emphasizing the credibility of the research—whether through publication in prestigious journals or affiliation with well-known institutions—had limited effect on disseminating the findings. While some recipients rated the studies as more credible, this perception did not correlate with broader sharing of information. In essence, the study concluded that neither peer approval nor academic prestige propelled the spread of ideas; rather, formal rank within the organization was the decisive factor.
This research arrives at a time when generative AI is already widely utilized within the World Bank. A majority of surveyed employees reported having a ChatGPT account, and many have incorporated it into their work processes. Despite robust support for these tools, most employees struggled to retain information regarding rigorous studies shared just weeks prior.
For organizations navigating rapid technological advancements, the implications are clear: evidence alone does not ensure effective information diffusion. The choice of messenger is as crucial as the content of the message itself. Senior employees are not only more inclined to disseminate information but also command greater attention when they do.
These findings also reveal a potential bottleneck in knowledge sharing, indicating that while ideas may flow more freely from the top, innovations from junior employees could remain obscured. As artificial intelligence continues to transform workplace dynamics, understanding the mechanisms of knowledge transfer within organizations is likely to become as vital as grasping the technology itself.
