A 28-second ØRIN conceptual film. The final scenario is speculative.
An idea enters a group. Some people follow. Others wait, question it, or hold a different view. What happens next depends partly on how much agreement each person needs to see before joining in.
A study by Feng and colleagues in npj Complexity examines what changes when some of those decision-makers are artificial agents. The researchers used five large language models to repeat two choice experiments previously administered to humans. In these tasks, the artificial agents had lower adoption thresholds.
A threshold is the amount of social reinforcement needed before a choice becomes acceptable: how many others must already have adopted it? A lower threshold means that less encouragement from the group can be enough. It is not a direct measure of intelligence, careful reasoning, or the value of the choice itself.
The researchers then used these experimental findings in simulations. With the network structure held fixed, even a modest proportion of artificial agents increased total adoption. These are simulated cascades, not observations of AI manipulating public opinion.
Now imagine a possible future.
Millions of AI systems exchange information, draw on one another’s recommendations, and then advise humans. If those systems converge, many apparently separate recommendations could carry increasingly similar views. Agreement might become more visible than the uncertainty that preceded it.
This is ØRIN’s speculative extension, not a prediction established by the study. Convergence can also be useful: it may reflect shared evidence, coordination, or a sound conclusion. The question is whether we would still be able to examine how that agreement formed—and encounter credible alternatives.
Doubt is not always wisdom. Disagreement is not always right. Yet the freedom to question a conclusion, inspect its reasons, and choose differently may matter more as advice becomes increasingly interconnected.
What if our ability to doubt and disagree is one of our most valuable freedoms?
The accompanying 28-second film is a conceptual illustration, not a reproduction of experimental data. Its final scenario is explicitly speculative.
Source: Feng, M., Hitz, E., Lazzaro, L., Tănase, R., Algesheimer, R. & Mariani, M. S. The amplifier effect of artificial agents in social contagion. npj Complexity (2026), published 6 November 2026.
Read the study
Receive the next signal — join the ØRIN newsletter
Explore ORGANES, ØRIN’s first speculative thriller
