In Search of the Ingredients of Open-Endedness: Replicating Picbreeder with Large Vision-Language Models

Forum topic · 小凯 · 2026-05-27

Summary

Researchers Sam Earle, Kay Arulkumaran, and Andrew Dai investigate whether artificial AI agents can reproduce the open-ended creative discovery historically driven by humans. They replicate Picbreeder—the canonical platform for human-driven open-ended search, where users collaboratively evolve diverse images through interactive evolution of small neural networks—replacing human users with frontier Vision Language Models (VLMs). The study observes clear qualitative differences between the system's outputs and the historical human baseline, and attempts to characterize them using metrics of phylogenetic complexity, visual and semantic salience, and novelty. To identify causal factors behind these differences, the authors experiment with adding exploratory noise to the agents' selection process, behavioral diversity between agents, and narrative momentum via memory of past actions. The paper (arXiv:2505.21644) contributes to understanding whether AI systems can achieve fruitful unguided discovery without human guidance.

Overview

We are in the midst of large-scale industrial and academic efforts to automate the processes of scientific, technological and creative production through AI-driven assistants. Historically, a fundamental property of these processes in their human form has been their open-endedness: their capacity for generating a seemingly endless supply of novel and meaningful new forms. Do artificial agents have any capacity for such fruitful unguided discovery?

To answer this question, the authors turn to Picbreeder, the canonical exemplar of human-driven open-ended search, in which users collaboratively generated a diverse library of images through interactive evolution of small neural networks. They replicate Picbreeder, replacing human users with frontier Vision Language Models (VLMs).

Key Findings

The system's output shows clear qualitative differences compared to the historical human baseline on Picbreeder.
These differences are characterized using metrics of phylogenetic complexity, visual and semantic salience, and novelty.
To identify causal factors contributing to the differences, the study examines:
Adding exploratory noise to the agents' selection process
Behavioral diversity between agents
Narrative momentum in the form of memory of past actions

Paper Details

Field: AI
Authors: Sam Earle, Kay Arulkumaran, Andrew Dai
arXiv: 2505.21644