Watch opinions spread
What happens when people and AI pass opinions around? In these experiments, run with Arlequin AI, statements about Does red meat cause cancer and cardiovascular disease? travel across a 5×5 grid. Each round, a new person or AI agent takes each spot, reads the statements at and around it from the round before, picks the one the group would most likely agree with, and rewrites it. A language model then labels every statement positive, negative or neutral. The grid starts split, with supporters on top and opponents below.
- + positive
- − negative
- 0 neutral
- polarization
- neighbour agreement (NCI)
The replay jumps between the iterations shown in the paper: 0, 1, 3, 5, 7 and 8.
Two numbers for a network
Polarization asks how split the network is. Count supporters as +1, opponents as −1 and neutral as 0, then take the variance: it’s highest when the grid is split down the middle and zero when everyone agrees. Neighbour agreement (the Neighbors Correlation Index, or NCI) asks whether people hold the same opinion as the people next to them. A network can be very polarized and still full of agreement, if each side keeps to its own camp. Try it: click a spot to change its opinion, or use the arrow keys and Enter.
Two camps: high polarization, high agreement. Checkerboard: about the same polarization, with every neighbour disagreeing. Everyone agrees: no polarization, and agreement is undefined because there is nothing to correlate.
Where this comes from
Related work
- Hybrid human-AI societies balance creativity and diversity Mixed networks of people and AI keep more variety than either alone.
- Characterizing the interaction of cultural evolution mechanisms in experimental social networks How copying, selection and network structure combine in experimental networks.
- Using LLMs to advance the cognitive science of collectives The case for studying groups of people and language models together.