The Dialogic Loop is a novel concept for human-AI interaction design that I am proposing. It emphasises the importance of a feedback loop when a user interacts with a generative AI system. Here's a breakdown of the concept: ### Definition: A Dialogic Loop exists when a user can instruct a generative AI system to create something, view what the AI generates, and subsequently modify or refine the output, thus initiating a dialogue-like interaction with the system. ### Significance: I hypothesise that the inclusion of a Dialogic Loop in co-creative AI systems tends to make the interaction more intuitive and effective. The essence of this idea is that the tool becomes more of a collaborator than just a passive instrument. ### Examples: 1. **Text-to-Image Systems:** Consider earlier versions of systems like Dall-E or Mid Journey. Users could feed a textual prompt to these systems, prompting them to generate corresponding images. Using the example of a "modern chair inside a well-lit living room in a leafy apartment," the system might produce a set of images. However, if a user wanted a minor alteration (e.g., a wider chair), the system did not offer a mechanism to make such specific refinements. It was a one-way generation process. ![[Files/image 6.jpg]] 2. **Image Reference in Modern Systems:** Newer iterations, such as the updated Mid Journey, allow users to input an image as part of their prompt, indirectly facilitating a sort of Dialogic Loop. While this lets users influence the output by providing visual context, the AI doesn’t have a memory of previous interactions or the capacity to directly modify the initial image. Instead, it merely uses the image as a reference for generating a new one. Thus, the loop still lacks fluidity. 3. **Evolution of Text Generators:** Before the rise of chat models like ChatGPT, we had systems like GPT-2 and GPT-3, which functioned primarily as autocomplete tools. Users provided a priming text, and the AI would extend or complete it. This structure did not facilitate a genuine dialogic interaction. For instance, if a user inputted a starting line for a story and wanted a specific genre direction after seeing the AI’s output, there was no mechanism to communicate that preference and refine the output accordingly. The introduction of ChatGPT changed this dynamic. Even though it still operates on an autocomplete principle, it's designed to autocomplete text in the form of a dialogue, allowing users to provide feedback and guide the AI’s responses. The user can communicate with the AI both through and about the creattion. This model simulates the idea of a Dialogic Loop more closely, and has made GPT more effective as a collaborative tool. ![[Files/image 7.jpg]]