About     Design and AI     SxD

Design and AI

For an industry that once championed disruption, the disruption wrought by AI and Large Language Models has proven a challenge. Design, be it user experience design, interaction design, user interface design, information or service design, is now forced to adapt to and adopt AI. Where once AI was considered a threat do design jobs, it's now an augmentation, a tool, and accelerant. The performance, speed, and time gains created by AI are simply too great to ignore.

When it comes to design contributions to AI, however, the picture is less clear. Not only is conversation a dominant mode of interaction with LLMs, AI is increasingly automating tasks and workflows such that we provide oversight and supervision. Even this is at risk, as the human-in-the-loop paradigm threatens to become the human-as-bottleneck paradigm.

My primary interests are in conversational AI, and the nuances that apply to the experience design of interactions with AI by chat, text, and voice. Having practiced social interaction design (SxD) previously and for social networking, social tools, and social media, social dimensions of AI are a nascent interest as well.

I maintain a large vault of AI white papers and excerpts, mostly from Arxiv.org, in an Obsidian vault that has been mapped to create topic notes. I use Claude against this vault to research concepts and insights, and to write. A growing number of examples of these projects can be found here, along with prior SxD articles, research, and blog.

Large language models "read" and generate text and documents, generating writing and documents in response. Language models also converse, using styles that are chat conversation or voice-based conversation. From an interaction desigbn persepctive, these are different modes of communication. The conventions common to speech and talk are distinct from those that shape discourse, even while both use language (and language is the foundation of LLMs).

Knowledge Graph

Browse this knowledge graph for a look at the topic notes in my Obsidian vault. The legend at bottom left represents categories with the highest density of connections. Use it to select just a single topic category. Hover over nodes for a mini summary of the topic note. Click to see related connections and associated white papers. White paper titles are active links.

Topic notes are concepts and insights shared across categories in the vault of white papers. They are not directly tied to individual white papers. When writing, Claude traverses topic notes to pull research ideas and insights from across the vault. A similar search with AI or on the web would surface just a fraction of these insights and papers, and each paper alone is a heavy lift and drain on the context window. These excerpts have been embedded already and can be used with vector and deep search.

Conversation is the Medium

An essay for ML and experience designers on conversation as shared context. I work with Claude through research on LLMs to shape an argument that the goal of AI-human interaction is interestingness in the experience.

The Reader Who Wasn't Reading

One research question, three reasoning approaches, three prompts, and three posts. A look at how prompting with different reasoning modules produces posts that make different arguments. Another experiment using Claude on an Obsidian vault of white paper research.

Knowledge Custodians

Knowledge Custodians: What impact AI will have on knowledge work, and on the role and status of the expert? I examine what makes expertise and how AI will reshape how it is produced and what qualifies it.

More posts written with Claude: Posts written from inside my Obsidian vault of AI white papers

• The Question Answer Trap: Do LLMs become trapped by the question, or the answer?
• Prototypical Writing: Why AI writes prototypes.
• Personas Impersonated The debate on AI personas, seen from the interaction designer's perspective.
• False Punditry: How AI Writing Is Changing Social Media From the Inside Out A collaborative AI reflection on AI-generated expertise, the silence it creates, and the displacement it causes.
• What If the Problem With AI Isn't That It Lies — But That It Doesn't? Six dominoes. Each one had to fall for the next, and none of them are getting back up.
• Two Ways to Build an AI Society (And Why the Choice Matters More Than You Think)
• Epistemic Hyperinflation: What Happens When AI Floods the Knowledge Economy
• Cognitive Surrender: How AI's Competence Becomes Our Vulnerability
• Intellectual Camouflage: A Linguistic Analysis of AI Writing on Social Media
• The Ventriloquists: How AI Counterfeits the Social Contract of Language
• The Meaning Gap: Why AI Talks Like It Writes — and Why That Changes Everything
• Why AI Writes Like It's Always Selling — and Can't Tell When That's Wrong
• Moltbook agents don't be social

• Conversational AI: LLMs expand the conversational capabilities and applications of AI, from open dialog to task-oriented dialog. Engaging and effective conversation are a matter for design.

• Design Theory: Design needs new concepts for the plastic and adaptive nature of generative AI. Speech and language as an interaction system, real-time generation as a technical capability, and many agentic architectures and integrations require us to update design concepts if not also methods.

• Domain-Specific AI: Domain-specificity is needed to make generative AI applications capable of handling the requirements of vertical commercial use cases. Design can contribute to assessing these needs within a domain and to shaping the performance of AI-based applications within it.

• Emotional AI: Emotions are added to voice mode conversational AI, but in order to make AI seem more human. The use of emotions creates novel design issues and considerations.

• Interestingness: In replacing search and creating experiences based more on discovery and exploration, language models can be designed to engage user interests more widely than with focused search. Design concepts for interest, over use, could help to shape these applications.

• Organizational Design: AI integrated into software and organizational applications will need to be designed, from task and job workflows to front-end interfaces.

• Personalization: AI will greatly expand its usefulness for end users when it is personalized to user preferences and data. Personalization is a design topic.

• Prompts: Prompts are an obvious application of design but are often considered engineering. Given the impact of prompting on LLM response generation, I think design theory needs concepts specific to prompting.

I have a list of the whitepapers I have read here.

       

©2007 - 2025 by Adrian Chan. All Rights Reserved.