Skip to main content Skip to navigation

Imitation versus Innovation: What children can do that large language and language-and-vision models cannot (yet)?

Project Overview

The document explores the impact of generative AI, particularly large language models (LLMs) and language-and-vision models, in the educational landscape, highlighting their potential as cultural technologies that enhance cultural transmission and imitation. It notes that while these AI systems can effectively imitate patterns found in data, they fall short in areas such as innovation and causal understanding, which are vital for genuine learning and adaptation. The findings emphasize that true educational outcomes depend not only on the capabilities of AI but also on the inherent cognitive abilities and embodied experiences of human learners. Consequently, the paper underscores the importance of integrating human-like learning processes with AI applications to foster deeper understanding and creativity in education. Overall, it suggests that although generative AI can support educational practices, it cannot replace the innovative and contextual learning that human experiences provide.

Key Applications

Large language models (e.g., OpenAI's GPT, Anthropic's Claude)

Context: Cognitive learning comparison between AI models and human children aged 3-7 years

Implementation: AI models were tested alongside children in tasks requiring imitation and innovation using novel tools and causal inference.

Outcomes: LLMs excelled at tasks requiring imitation but struggled with innovation and causal reasoning compared to children.

Challenges: LLMs rely on statistical patterns and lack the ability to infer novel causal relationships.

Implementation Barriers

Cognitive Barrier

LLMs struggle to understand and infer novel causal relationships and innovate beyond imitation.

Proposed Solutions: Further research into cognitive development techniques could inform improvements in AI systems.

Project Team

Eunice Yiu

Researcher

Eliza Kosoy

Researcher

Alison Gopnik

Researcher

Contact Information

For information about the paper, please contact the authors.

Authors: Eunice Yiu, Eliza Kosoy, Alison Gopnik

Source Publication: View Original PaperLink opens in a new window

Project Contact: Dr. Jianhua Yang

LLM Model Version: gpt-4o-mini-2024-07-18

Analysis Provider: Openai

Let us know you agree to cookies