The Grounding Problem and Co-Creative AI

Source: AI Mind Labs / Medium
Author: Mary Lou Maher
Original source: https://medium.com/ai-mind-labs/the-grounding-problem-and-co-creative-ai-1b3348a0150e
Source type: essay

Private backup: the full article text is archived in the private repository at archives/articles/medium-com-the-grounding-problem-and-co-creative-ai-1b3348a0150e.source.md. It is not published on the public Quartz site.

Summary

The article uses the grounding problem to explain how generative AI can produce outputs perceived as meaningful or creative despite operating over symbols or vectors. It distinguishes symbol grounding and vector grounding, then argues that referential grounding can connect symbolic cognitive models and vector-based generative models through shared concepts in the world. The author proposes that co-creative AI can be strengthened by integrating cognitive models of creativity, expectation, surprise, novelty, and human values with deep learning models. The larger claim is that grounding can support alignment, instructability, and more meaningful human-AI co-creation.

Big ideas

• Students need to bring the purpose; AI should not supply it for them
• AI is changing what knowledge work asks people to do

Claims

• Co-creative AI needs shared concepts about the world
• AI can reshape a student’s purpose, but it should not replace it

Key evidence and examples

• The article defines grounding types including sensorimotor, communicative, epistemic, relational, and referential grounding.
• Symbolic representations and vector representations are both described as mapping to concepts in the world.
• Computational creativity models include expectation, surprise, similarity, novelty, semantic distance, and syntactic distance.
• Co-creative design systems are described where a human sketches and AI provides inspirational sketches through cognitive-model/generative-model interaction.

Pull quotes

The grounding question

The Grounding Problem asks how AI systems can be grounded in a world with which they have no direct interaction.

Meaning across models

This is where the insight for how we find meaning in both LLMs and cognitive models of creativity emerges.

Human values as a bridge

Models of human cognition can therefore provide a bridge between language models trained on large data sets to human values.

Education relevance

Indirect but useful for advanced AI literacy, computational creativity, design learning, co-creative tools, and discussions of alignment and meaning-making.

Durability note

This article is durable as a conceptual bridge between grounding, creativity research, and human values; specific model examples may age, but the grounding question remains central.

My notes