Seeking the Unfamiliar but Memorable: Conceptual Creativity as Meta-Learning

2026-05-15

AI Summary

This work addresses the challenge of generating concepts that are both novel and readily learnable, rather than merely reproducing existing content. It introduces a Creator-Appraiser framework that formalizes creativity as “learnability for an adaptive observer” and leverages meta-learning to drive a frozen generative model. The Creator employs a pretrained diffusion model, while the Appraiser—implemented either as an MNIST autoencoder or a CLIP model with low-rank adapters—simulates the observer’s learning process in an inner optimization loop. Without fine-tuning the generative model or enhancing linguistic conditioning, this approach synthesizes novel concepts and stylistic variants that lie outside the base model’s original generative capacity, yet remain sufficiently structured to be quickly grasped by observers.

Abstract

What does it mean to create a new concept, rather than retrieve a familiar one? Repeatedly sampling a generative model at the same prompt produces variations with similar styles and typical content. We propose that creativity is the production of stimuli that are unfamiliar to an adaptive observer at first sight, but quickly learnable from a few exposures. We formalize this as a Creator-Appraiser pair: a Creator generates a candidate, an Appraiser adapts to it for a few inner-loop learning steps, and the Appraiser's improvement becomes the reward the Creator optimizes through. We instantiate the framework with diffusion as the Creator, an autoencoder Appraiser on MNIST, and a CLIP Appraiser with a low-rank adapter for natural images. The diffusion model remains frozen with no additional language conditioning; the meta-learning gradient is enough to produce both stylistic variations and concept compositions that the base model does not generate on its own.

Problem

• conceptual creativity
• meta-learning
• unfamiliar stimuli
• learnability
• generative models

Innovation

• conceptual creativity
• meta-learning
• Creator-Appraiser framework
• diffusion models
• fast adaptation