TLDR: Orchid is a new system designed to improve how users interact with Generative AI in creative projects. It allows users to specify project context, personal preferences, and stylistic personas, and then seamlessly reference and monitor this information across multiple interactions and sessions. A user study demonstrated that Orchid leads to more novel and feasible creative outcomes, better alignment between user intent and AI responses, and a stronger sense of control and collaboration with the AI, effectively addressing issues like context drift and repetitive prompting in current GenAI tools.
In the rapidly evolving landscape of Generative AI (GenAI), context is paramount for meaningful interactions. However, current tools often fall short in orchestrating this context effectively, especially across complex creative projects that span multiple interactions, sessions, and AI models. This often leads to users repeatedly re-specifying details, juggling various artifacts, and struggling with context drift, which can hinder creativity and obscure user intent.
Introducing Orchid: A New Approach to Context Orchestration
To tackle these challenges, researchers Srishti Palani from Tableau Research and Gonzalo Ramos from Microsoft Research have introduced Orchid, a novel system designed to empower users with robust mechanisms to specify, reference, and monitor context throughout their creative workflows. Orchid aims to keep creators and AI aligned, thereby augmenting their creative potential.
Orchid’s core functionality revolves around three key affordances:
-
Specifying Context: Users can define various types of context. This includes ‘Project context’ through workbook pages, uploaded supporting materials, and a home page for project planning and task management. ‘User’s Personal context’ is captured on a ‘Me page’ where users can share their emotional state, design preferences, and personal insights. Lastly, ‘Stylistic context’ is managed via ‘Persona pages’, allowing users to create and channel the perspectives of simulated collaborators or experts with specific skill sets and personality traits.
-
Referencing Context: Orchid provides flexible ways to integrate relevant context into GenAI interactions. Users can explicitly reference context using an ‘@’ shorthand, similar to common notebook interactions. They can also use in-line selection to specify context for a prompt or implicitly ground interactions by simply prompting within a specific page or associated persona. These mechanisms help generate ‘meta-prompts’ that seamlessly integrate context, ensuring outputs remain context-aware and aligned with user intent.
-
Monitoring Context: To maintain transparency and control, Orchid features a ‘Transparency Lens’. By hovering over the generate button or a result block, users can see exactly what context (page, persona, etc.) the AI interaction is grounded on, allowing them to understand and adjust the information being used.
User Study Highlights Orchid’s Impact
A within-subjects study involving 12 participants demonstrated Orchid’s significant benefits. Participants using Orchid for creative tasks, compared to a baseline toolkit of web search, LLM-based chat, and digital notebooks, produced more novel and feasible outcomes. They also reported greater alignment between their intent and the AI’s responses, higher perceived control, and increased transparency.
Key findings from the study include:
-
Better Creative Outcomes: Expert raters found that ideas generated with Orchid were significantly more novel, feasible, and valuable.
-
Improved Context Alignment: Participants issued fewer prompts with Orchid, suggesting more efficient context management. They frequently used explicit and implicit grounding mechanisms, reducing the need for repetitive prompting and manual context switching.
-
Perceived as a Collaborator: Users actively edited their ‘Me’ pages and ‘Persona’ pages, personalizing the AI’s understanding. Participants felt Orchid provided empathetic support and viewed the GenAI as a collaborative partner rather than just a tool, fostering a more engaging and supportive creative process.
-
Enhanced Creative Control: The Transparency Lens gave users confidence in their inputs and outputs, allowing them to understand and influence the AI’s reasoning.
Also Read:
- Beyond Chatbots: How LLMs Are Learning to Build Interactive User Interfaces
- Rethinking How We Interact with AI: Insights from a Design Thinking Workshop on LLM Interfaces
The Future of Creative Workflows with AI
Orchid represents a significant step towards next-generation GenAI tools that support complex, iterative workflows. By prioritizing context orchestration, it addresses critical gaps in existing systems, enabling creators and AI to stay aligned and augment their creative potential. The research suggests that such context-aware systems can not only enhance creative outcomes but also transform the user’s perception of AI into that of an empathetic and accountable teammate.
For more in-depth details, you can read the full research paper: Orchid: Orchestrating Context Across Creative Workflows with Generative AI.
