Artificial intelligence agents are capable of making collective decisions through voting to select a specific option, and they tend to follow the majority choice, which resembles the 'herd behavior' observed in animals and humans.
This was demonstrated in an experiment conducted by researchers from the University of Konstanz in Germany. The study involved forming groups consisting of approximately 1000 bots that utilized algorithms such as GPT from OpenAI, Claude from Anthropic, and Llama from Meta.
The AI agents were asked to choose between 'opinions' (which were meaningless and consisted only of letters) or, in another test, to choose between the words 'yes' or 'no' (also without any context).
The robots exhibited two curious behavioral patterns: they showed an unusual tendency to vote 'yes,' and they also displayed a propensity to follow the majority. When placed in a situation where a certain 'opinion' already had a large number of votes, they became more inclined to choose it, potentially changing their initial vote.