Workplace Everyday-Creativity through a Highly-Conversational UI to Large Language Models

Paper · Source
Workplace ApplicationsDesign FrameworksCo-Writing and Collaboration

We explore everyday co-creativity for collaborative human-AI teams in workplaces via a conversational user interface to a large language model. Previous short papers explored human-AI team-creativity methods such as framing and reframing. This experiment examines aspects of brainstorming.

In this study, we used a highly-conversational user interface (UI) to the Llama2 large language model (LLM), similar to [23] Each conversational turn by the AI was “tuned” to be brief, humble, and helpful, through prompt design and some intermediate software between the UI and the LLM. Conversations with the AI took place in an internal group chat application on a company-internal server.

Figure 1 shows excerpts from an informal, one-user dialog during the divergent thinking phase of the brainstorming activity. The human proposes a business problem as the topic of brainstorming at Figure 1A. The AI proposes “a few brainstorming ideas” at Figure 1B. While these ideas may not be brilliant, we believe they are plausible and potentially useful.

We also note that the work of the human and the AI moved between exploratory activities (divergent thinking, Figure 1A-B, D, G, J-K) and optimization activities (critical and convergent thinking, Figure 1C, F, H). These fluid changes in focus – within the broader context of brainstorming – appear to be related to Schön’s analyses of designers’ reflective practice [24] as constantly developing and emerging through conversations with the design materials [25].

A screenshot of a computer

3 DISCUSSION: THE LOCUS OF CREATIVITY

Unlike projects that elicit creative outputs from generative AI models [2, 16], we pursue the concept of co-creativity as a process that is distributed [5] between the two parties of a human-AI collective [11]. We propose that creativity emerges through interaction, in the hybrid space between human and AI (see e.g., [14]). For example, the human framed the problem space in Figure 1A, and the AI proposed ideas into that frame in Figure 1B. Within a more critical subframe, the pattern repeated in Figure 1C-D and again in Figure 1F-G.

Lines of inquiry this paper opens 24

Research framings built by reading the notes related to this paper — the questions it feeds into.

Does AI assistance enhance or undermine professional skill development? How does AI adoption reshape employment and wage outcomes? Can verification capability keep pace with AI generation speed? How do different editing strategies affect output quality and agent safety? How has ChatGPT adoption actually reshaped labor markets and information ecosystems? Can we reliably measure and demonstrate AI's actual productivity impact? Does generative AI replace coordination with individual productivity? Does AI-assisted coding actually improve developer productivity and code quality? How should conversational agents balance proactive goal-pursuit with user autonomy? How does AI writing assistance affect reader perception of writer authenticity? How should control and autonomy be allocated in human-AI collaboration? What explains the gap between LLM semantic understanding and production?