How Collective Intelligence Emerges in a Crowd of People Through Learned Division of Labor: A Case Study

TL;DR

This study addresses how collective intelligence emerges in a crowd through learned division of labor, using LinYi's Sharing Control Game as a case study. By formulating SCG as a networked multi-agent MDP and applying distributed actor-critic MARL, the authors show CI arises when two intrinsic conditions—related to total population and the distribution of elite social power—are met, which is grounded in MJLS stability analysis. They introduce an emergence index and a distributed joint-action estimation method enabling individuals to learn social roles without global action information. Numerical simulations replicate observed CI phenomena, revealing that CI requires a balance between elite and common players and that a purely elite society fails to achieve robust CI.

Abstract

This paper investigates the factors fostering collective intelligence (CI) through a case study of *LinYi's Experiment, where over 2000 human players collectively controll an avatar car. By conducting theoretical analysis and replicating observed behaviors through numerical simulations, we demonstrate how self-organized division of labor (DOL) among individuals fosters the emergence of CI and identify two essential conditions fostering CI by formulating this problem into a stability problem of a Markov Jump Linear System (MJLS). These conditions, independent of external stimulus, emphasize the importance of both elite and common players in fostering CI. Additionally, we propose an index for emergence of CI and a distributed method for estimating joint actions, enabling individuals to learn their optimal social roles without global action information of the whole crowd.

Figures (6)

• Figure 1: LinYi's Sharing Control Game: $N$ players share control over one single car. 'w','s','a','d' stand for 'advance','brake','left','right' respectively. In this paper, 'advance' or 'brake' are represented by throttle $T$,'left' or 'right' are done by steering angle $\delta$. Differences between 'wwww' and 'w' are captured by different magnitudes of throttle $T$.
• Figure 2: (a): Flow chart of a SCG example. Division of labor within human societies reflect a natural SCG: in a hospital, doctors are assigned to different departments according to their specialties, a 'spatial distribution'. Doctors within a department rotate in shifts, a 'temporal distribution'. (b): One example of the random communication graph $\mathcal{G}_k$, which is connected, undircted, sparse graph with $N$ nodes.
• Figure 3: (a): role policy $\pi^i(x,1)$ evolution for three SCG cases. Specifically, players from group 2 are marked with dashed lines. (b): return vs episodes comparison between three SCG cases and baseline cases.
• Figure 4: Trajectory comparison between SCG Case 1 and baseline cases. The joint role action $\underline{a} = [45,5]$ and $[90,0]$ can represent the group decisions when learning process just begins and that when learning process finished.
• Figure 5: Input $T,\delta$ and states $p_y, v$ comparison between SCG Case 1 and baseline cases. (a): throttle $T$. (b): steering angle $\delta$. (c): position in y axis $p_y$. (d): velocity $v$.