A Turing Test for Crowds

TL;DR

This study investigates whether humans can distinguish between real and simulated crowds, revealing that while people can tell the difference, they cannot reliably identify real crowds, highlighting gaps in simulation realism and believability.

Contribution

It introduces a Turing Test for crowds to evaluate the realism and believability of crowd simulations from a human perception perspective.

Findings

Humans can reliably distinguish real from simulated crowds.

Humans are unable to identify real crowds.

Realistic simulations are not perceived as believable.

Abstract

The realism and believability of crowd simulations underpins computational studies of human collective behaviour, with implications for urban design, policing, security and many other areas. Realism concerns the closeness of the fit between a simulation and observed data, and believability concerns the human perception of plausibility. In this paper, we ask two questions, via a so-called "Turing Test" for crowds: (1) Can human observers distinguish between real and simulated crowds, and (2) Can human observers identify real crowds versus simulated crowds? In a study with student volunteers (n=384), we find convincing evidence that non-specialist individuals are able to reliably distinguish between real and simulated crowds. A rather more surprising result is that such individuals are overwhelmingly unable to identify real crowds. That is, they can tell real from simulated crowds, but are unable to say which is which. Our main conclusion is that (to the lay-person, at least) realistic crowds are not believable (and vice versa).