AI agents can help groups agree—or shape what they agree on, a study summary says
A user summarizing research says human-led groups used more concrete, real-world image descriptions, while AI-led groups used more abstract, geometric language.
TLDR
A user's summary describes a study in which 24-person groups of humans and AI agents repeatedly tried to agree on descriptions of the same image. It says consensus improved by 8.0% over the all-human group when AI participation was 12.5%, but agreement worsened at 33.3% and 50%. At 75%, the summary says, strong consensus returned—but humans were moving toward the agents’ language.
Combined views
818
1 Source, first seen 25d ago