According to Bohm, dialogue is the iterative creative process that facilitates the emergence of a common meaning. Two different parties in a dialogue have different assumptions, or maps of meanings. Through dialogue, they note the difference, and eventually arrive to a new common set of meanings. *For example, consider a dialogue. In such a dialogue, when one person says something, the other person does not in general respond with exactly the same meaning as that seen by the first person. Rather, the meanings are only similar and not identical. Thus, when the second person replies, the first person sees a difference between what he meant to say and what the other person understood. On considering this difference, he may then be able to see something new, which is relevant both to his own views and to those of the other person. And so it can go back and forth, with the continual emergence of a new content that is common to both participants. Thus, in a dialogue, each person does not attempt to make common certain ideas or items of information that are already known to him* **My Thoughts** [[A creative collaboration is a co-exploration of a conceptual space]]. But we don’t have the same conceptual space in our heads. [[Conceptual spaces in machine learning terms could be understood as latent spaces]]. Latent spaces encode meaning relationally. Their very definition is that they are semantically meaningful. Dialogue enables the alignment of meaning. Therefore, [[Dialogue enables the alignment of latent spaces in humans and machines]] **From** [[On Dialogue (Bohm, 1990)]] Dialogue has a rich history both in social sciences, philosophy and cybernetics. Dialogue has intrigued thinkers for being a mechanism that builds meaning across agents, and provides a means for coordination. For example, Bohm wrote extensively about the idea of dialoge as something that can build a shared set of meanings and thus he believed dialogue was in fact key to address soome of the key concerns of the time, namely, possible nuclear annhilation. recently, dialogue has received interest from the felds of human computer interaction becaue it provides a useful metaphor for designing interaction between humans and machines. In particular, interactions that enable collaoration. In my research, I explroe the potential of dialogue to enable effective collaboration between humans and Ai in creative activities. I define dialogue as an iterative feedback loop that allows the human and Ai to take a common creation from a less finished state to a more finished state, and were they can interact both trhough and about the artifact. When they interact through the artificat they are acting on the creation. For example, if they are creating a song, this would mean playing a chord. on the other hand, when they are interacting about the creation, they are dicussing goals creative directions and shared meanings. Dialogue is usually understood to involve language but in my conception this is not necessary. In fact, yes dialogic interaction can involve natural language but in fact i define it as mechanism of exchange. When I refer to dialogue, I am more interesting in the dynamics of information exchange, the circular motion, the back and forth, and the interative nature of the interaction, which mantains a context over the interaction and which might also refer to previous context in previous itneractions. To understand the relevance of dialogue, it is important to look at what non'dialogic interaction looks like. Pretty much every interface that we interact with today, at least before the advent of artificial intelligence is non dialogie. When I interact with my graphical interface in my operating system, I am clicking a button, for example, the finder, selecting a file. I am merely operating, but I am not engaging in a dialogue. The interaction mostly flows one way, even though I might receive visual feedback on my acions. There is no context carried over. There is no discussion with the computer about the goal I want to achieve with this interaction. On the other hand, when I interact with a dialogic system, every action elicits a response from the system that is non preditable, or at least non deterministic. In large part, I dont think an operating system requires a dialogic interaction. The interaction is simple enogh that selecting a file would do. Even though one could imagine a natural language based interface like the one presented in the movioe her. In this case, I dont need to engage in a complex feedback loop of communication with the computer to complete the task at hand, deleting a file. However, when I engage in a creative activitiy with a system the task is more complex. Say I want to create an interior design for a building, and I want to use an AI image generator to help me. In this case, it is not as simple as dragging a file to the trash can. It is not entirely determinstic. This task will involve taste and iteration. I need to be able to express to the machine what style I am looking for. I need to be able o receie some suggestions and express that I like more ones or others. I need to be able to take parts from some, and completely discard others. This is the nature of a dialogic interaction. However, most generative systems now, are not dialogic and behave in a similar way to my operating system. Merely operational, not dialogic. I propose that for Ai creative systes to be effective collaboration, they need to be defined for dialogicness. A particular case where the usefulness of a system exponentally increased, at least as measred by the number of users is ChatGPT. ChatGPT uses the same model that had been in existence for many months, namely, GPT3. These mode, a transformer LLM, was able to follow intrsuctions and complete pieces of text. However, at one point, OpenAI decided to train this model, or fine tune it, for dialogue, soit could engage in a back and forth communication where precisely the user and the ai could discus goals, provide clarification, follow up, refinement and mantain a context. GPT3 was a popular model but it was still mostly only know to people within the AI field, or in close proximity. However, ChatGPT became the fastes growing app in history reaching 100M users in less that three weeks. OpenAI hs claimed that this simple interaction adjustmenet, along with making access to the model more open (even though GPT3 was already open outside of beta), is what made it so succesful. In large measure, making the model dialogic increased its usability radically. #Idea