Session 4: Building Shared Conceptual Grounding for Interacting with GenAI

Session 4: Building Shared Conceptual Grounding for Interacting with GenAI

by Stanford HAI

Humanities, Social Sciences & Thought Computing & Cybersecurity UYComputer scienceUYZHuman–computer interaction
🎙 Stanford HAI 👥 34K 📅 October 30, 2025 ⏱ 46 min 👁 441 📄 expert opinion 🧭 2026-08-06
Available in: English (current) Français

Keywords

shared conceptual groundinggenerative AIhuman-AI collaborationdesign choicesprompting

Summary

The talk, presented by researchers from Stanford HAI, addresses the challenge of achieving shared conceptual grounding between humans and generative AI tools. The speaker uses the analogy of Ansel Adams and his assistant Alan Ross to illustrate how human collaborators establish common understanding through conversational repair and shared concepts. The talk demonstrates the difficulties of prompting AI for specific design outcomes, showing a series of failed attempts to generate an image of Stanford’s Memorial Church with desired tonal qualities. The core argument is that lack of shared conceptual grounding leads to trial-and-error prompting and unsatisfactory ‘AI slop’. The project aims to develop tools that enable shared grounding by first studying how humans communicate during collaborative creation, using experiments where pairs work on CAD drawings. The team includes experts in cognitive psychology, HCI, AI, and education. Progress includes developing experimental protocols to analyze multimodal communication. The talk concludes with implications for improving AI collaboration and future goals.

156 words

Critical Evaluation

The talk provides a compelling and well-structured argument for the importance of shared conceptual grounding in human-AI collaboration. The use of the Ansel Adams analogy is effective in illustrating the concept and grounding it in a familiar context. The speaker’s personal experience with prompting AI for a specific image vividly demonstrates the frustrations of current tools, making the problem tangible. The research approach, which involves studying human-human collaboration to inform AI design, is methodologically sound and aligns with established practices in HCI and cognitive science. However, the talk presents preliminary findings without detailed data or peer-reviewed publications, so the evidence is largely anecdotal and based on expert opinion. The experimental setup described is promising but lacks specifics on sample sizes, controls, and analysis methods. The talk also does not address potential limitations or alternative approaches. The sources cited are primarily the project’s own website and general references to articles, which are not detailed. Overall, the talk is insightful and thought-provoking, but its scientific rigor is limited by the lack of published evidence. The title accurately reflects the content, and the talk is well-suited for an academic audience interested in human-AI interaction.

191 words

Title / Content Match

The title accurately reflects the content, which focuses on establishing shared conceptual grounding between humans and generative AI.

Quality & Reliability

8/10

The talk is presented by researchers from Stanford HAI, a reputable institution, and discusses ongoing research with a clear methodology. The content is based on expert knowledge and preliminary findings, but lacks peer-reviewed publications or detailed data, so a high but not perfect score is given.

Chapters

Cited Sources

Concurring Sources

Contribution & Novelties

The talk introduces a novel framework for understanding and improving human-AI collaboration by emphasizing the need for shared conceptual grounding. It proposes a research agenda that combines cognitive psychology and HCI to study human communication and then apply those insights to AI systems. The use of the Ansel Adams analogy provides a clear and memorable illustration of the concept.

Pour aller plus loin :

  • Theory of Mind — Relevant to the mutual understanding between humans and AI.
  • Conversational repair — Key mechanism for establishing common ground in dialogue.
  • Zone System — Ansel Adams’ technique for controlling tonal range, used as an example of shared concepts.

105 words

Radar Profile

The radar profile shows high scores in information quantity, quality, and reliability, with a slightly lower technical level, indicating a well-balanced and credible presentation that is accessible to a broad academic audience.

Reliability 8/10