**Outline for further development: ** **What gap is the DCAI filling?** - there is a divide between creative Al models and users. - Most SOTA creative AI models have not been widely adopted in creative workflows - This is not a machine learning problem but an interaction design one. - the DCAI fills that gap with a dialogic proposal. **Why dialogue ** - advantages: more explainable and understandable (from [[A Speculative Exploration of the Role of Dialogue in Human-Computer Co-creation (Bown 2020)]]). The user clearly expresses what they want. - there is a conversation that helps build shared meanings and contexts, a sort of intersubjectivity. [[Gordon Pask]]'s conversation theory: conversation is the process in which shared understandings emerge - The direction where multimodal creative Al is going.[[Codex]],[[Dall-E]], [[NÜWA]] **Why a typology** The DCAI does not intend to produce a completely objective classification, but rather a general typology that classifies models according to dialoguic characteristics. The fist aim of creating this typology is to propose those characteristics, and second, to group the produced prototypes within them. **Use for this typology in the field** - Given that adoption of AI in creative models is an interaction problem rather than an ML one (back this claim), a DCAI typology would give designers an empirically grounded tool for designing interactions between creative AI and creative humans. - Moreover, given the increased interest in [[human-AI collaboration]] such that human capabilities are augmented rather than replaced, the DCAI provides a clear route to design for that. **Future work** The typology can be understood as a part of a set of deliverables for the DCAI, which includes: - a typology - Empirically grounded design principles for dialogic interactions (accounting for creative self-efficacy, creative block, etc) - A set of working and tested prototypes Relevant literature: - [[conversation theory]] - [[Lambros Malafouris]] [[Material Engagement Theory]] - [[Lambros Malafouris]] [[Material Agency]] - Other info: > From "Typologies, taxonomies, and the benefits of policy classification" by Kevin B. Smith (Policy Studies Journal, Sep 2002): > > There are two basic approaches to classification. The first is typology, which conceptually separates a given set of items multidimensionally… The key characteristic of a typology is that its dimensions represent concepts rather than empirical cases. The dimensions are based on the notion of an ideal type, a mental construct that deliberately accentuates certain characteristics and not necessarily something that is found in empirical reality (Weber, 1949). As such, typologies create useful heuristics and provide a systematic basis for comparison. Their central drawbacks are categories that are neither exhaustive nor mutually exclusive, are often based on arbitrary or ad hoc criteria, are descriptive rather than explanatory or predictive, and are frequently subject to the problem of reification (Bailey, 1994). > > A second approach to classification is taxonomy. Taxonomies differ from typologies in that they classify items on the basis of empirically observable and measurable characteristics (Bailey, 1994, p. 6). Although associated more with the biological than the social sciences (Sokal & Sneath, 1964), taxonomic methods–essentially a family of methods generically referred to as cluster analysis–are usefully employed in numerous disciplines that face the need for classification schemes (Lorr, 1983; Mezzich & Solomon, 1980).