Computing & AI · the experiment
Tip the convention over
Thousands of AI agents agreeing on a word, a network that decides who talks to whom, a handful of stubborn agents. Four questions, from beginner to expert, on what the shape of the network changes.
1 · Choose where to place the stubborn agents
Beginner
Where should stubborn agents go to overturn a convention with as few of them as possible?
2 · Count the agents saved
Heterogeneous network of 10,000 agents. Median critical mass: 3% stubborn agents on the hubs, 16% placed at random.
Intermediate
How many stubborn agents do you save by targeting the hubs rather than placing them at random?
3 · Explain the rescued consensus
With Llama, the pair {young, old} never settles in a large well-mixed crowd, but always does on the most homogeneous networks. The authors keep everyone's number of contacts and reshuffle the links before every encounter: the rescue vanishes.
Expert
What was rescuing the consensus?
4 · Reread the old figures
Earlier studies measured the critical mass in crowds where anyone can talk to anyone.
Expert
What are their figures worth for a real deployment, where each agent has its own contacts?
Convention overturned
—