Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A social path to human-like artificial intelligence

Edgar A. Duéñez-Guzmán, Suzanne Sadedin, Jane X. Wang, Kevin R. McKee, Joel Z. Leibo

TL;DR

The paper argues that progress in artificial intelligence has largely hinged on large, exogenous datasets, and progress stalls when agents must rely on self-generated data in static environments. It proposes a framework where compounding innovation arises from multi-scale social interactions—collective living, social relationships, and major evolutionary transitions—that generate novel data through population pressures, arms races, social learning, and cumulative culture. It surveys evidence from AI, including multi-agent games (e.g., Capture-The-Flag, StarCraft II) and cooperative strategies (Diplomacy), to illustrate how interactions and communication shape learning signals and data streams. It advocates a hybrid approach that blends emergent social dynamics with engineered modules (curiosity, social learning) and emphasizes language as a driver for rapid, flexible data transmission and cumulative culture. The goal is to enable ongoing data generation across scales, potentially yielding human-like AI through structured, self-reinforcing data ecosystems while cautioning about the risks of self-reinforcing, tail-depleted data when models train on their own outputs.

Abstract

Traditionally, cognitive and computer scientists have viewed intelligence solipsistically, as a property of unitary agents devoid of social context. Given the success of contemporary learning algorithms, we argue that the bottleneck in artificial intelligence (AI) progress is shifting from data assimilation to novel data generation. We bring together evidence showing that natural intelligence emerges at multiple scales in networks of interacting agents via collective living, social relationships and major evolutionary transitions, which contribute to novel data generation through mechanisms such as population pressures, arms races, Machiavellian selection, social learning and cumulative culture. Many breakthroughs in AI exploit some of these processes, from multi-agent structures enabling algorithms to master complex games like Capture-The-Flag and StarCraft II, to strategic communication in Diplomacy and the shaping of AI data streams by other AIs. Moving beyond a solipsistic view of agency to integrate these mechanisms suggests a path to human-like compounding innovation through ongoing novel data generation.

A social path to human-like artificial intelligence

TL;DR

The paper argues that progress in artificial intelligence has largely hinged on large, exogenous datasets, and progress stalls when agents must rely on self-generated data in static environments. It proposes a framework where compounding innovation arises from multi-scale social interactions—collective living, social relationships, and major evolutionary transitions—that generate novel data through population pressures, arms races, social learning, and cumulative culture. It surveys evidence from AI, including multi-agent games (e.g., Capture-The-Flag, StarCraft II) and cooperative strategies (Diplomacy), to illustrate how interactions and communication shape learning signals and data streams. It advocates a hybrid approach that blends emergent social dynamics with engineered modules (curiosity, social learning) and emphasizes language as a driver for rapid, flexible data transmission and cumulative culture. The goal is to enable ongoing data generation across scales, potentially yielding human-like AI through structured, self-reinforcing data ecosystems while cautioning about the risks of self-reinforcing, tail-depleted data when models train on their own outputs.

Abstract

Traditionally, cognitive and computer scientists have viewed intelligence solipsistically, as a property of unitary agents devoid of social context. Given the success of contemporary learning algorithms, we argue that the bottleneck in artificial intelligence (AI) progress is shifting from data assimilation to novel data generation. We bring together evidence showing that natural intelligence emerges at multiple scales in networks of interacting agents via collective living, social relationships and major evolutionary transitions, which contribute to novel data generation through mechanisms such as population pressures, arms races, Machiavellian selection, social learning and cumulative culture. Many breakthroughs in AI exploit some of these processes, from multi-agent structures enabling algorithms to master complex games like Capture-The-Flag and StarCraft II, to strategic communication in Diplomacy and the shaping of AI data streams by other AIs. Moving beyond a solipsistic view of agency to integrate these mechanisms suggests a path to human-like compounding innovation through ongoing novel data generation.
Paper Structure (1 section, 2 figures)

This paper contains 1 section, 2 figures.

Table of Contents

  1. Author contributions

Figures (2)

  • Figure 1: Learning quality depends on the richness and size of the dataset. Learning agents improve as a function of the richness and size of their dataset. a, A single agent learns from a large but static dataset. b, Interaction with the world enables the agent to generate its own data, but can trap the agent in local minima, limiting the dataset richness. c, Other agents enrich the dataset by changing the world with which an agent interacts.
  • Figure 2: Social interactions drive compounding innovation. Interactions in the three drivers of compounding innovation have qualitatively different properties. a, Agent interactions in Collective living are anonymous, mediated by proximity. b, In Social relationships, the identity of individuals and their relationships matter during interactions, creating networks that facilitate cooperation and social learning. c, Major transitions lead to the evolution of multi-scale agents, where larger-scale agents regulate the environments of smaller-scale ones.