The Future of Co-Creative Systems

Artificial intelligence systems may exist at very different levels in the artistic process. They can be considered tools that are eventually the target of curatorship, or agents that can share a path with the artist, building on it as co-creative agents. This paper promotes a discussion on the interaction with artificial systems, the relationships established with them, as well as possible directions to take in the development of a classification framework.

An open question in co-creativity research is “What positive behaviors and outcomes are observed in humans as they engage in human-computer co-creativity?”. The effective design of co-creative systems to achieve these behaviors and outcomes represents an area of great untapped potential for the HCCC field that merits further consideration. We define Therapeutic Computational Creativity (TCC) as the study, design, engineering, and application of Computational Creative systems that fundamentally engage humans in a co-creative enterprise for the purpose of improving the mental or physical health of human participants.

In this abstract I consider as an open question how a distributed view of creativity influences ideas for the design of co-creative systems (Davis, 2012). I propose a distributed approach to co-creative systems embedded in a semantic web framework (Berners-Lee et al, 2001). The distributed view of creativity understands creativity as something that primarily occurs through interactions between numbers of people, with individual creative acts relegated to lower significance than they are assigned historically (Csikszentmihalyi, 2015; Simonton, 2012). Consequently, it views creativity as a properly emergent process, where the value systems according to which things are created coevolve with those creative products. For example, an individual perspective on creativity may consider a musician composing a piece of jazz music, whereas a distributed perspective would consider the evolution of the entire jazz style and the ability for any individual to influence that value system as well as respond to it. From a distributed perspective, there is no clear set of cognitive abilities that comes together in an individual to achieve creative outcomes, although there are key cognitive abilities that contribute to creativity, that have specific effects when combined in certain ways. These include learning abilities and strategies for search. Instead, different people may contribute in different ways to creative outcomes, utilising acts of cognition that are more or less associated with the traditional cognitive view of creativity: a composer coming up with new musical ideas, performers and producers interpreting, label managers promoting, critics analysing, all acting collectively in a distributed emergent process. Understood in this way, co-creative computational systems may not act in obvious ways like human creative agents, but might perform small modular tasks equivalent to these various human roles, such as suggesting ideas, performing evaluation, or embodying specific skills such as painterly rendering or music harmonisation. This distributed perspective is also mirrored in models of distributed creative search that work at a psychological level, such as Neural Darwinism (Edelman, 1987) and Wiggins’ IDyOT model (Wiggins and Forth, 2015). In this presentation I consider how this distributed perspective might influence how we design co-creative systems in more distributed ways. For the purpose of such a discussion, I propose a speculative architecture for creativity support that takes the form of a network of creative systems that could be accessed via a common interface. With this architecture, a creative musician, for example, may submit a request for a new music segment, much as we request search results from Google.com. The request is received from a central ‘genie’ (for example an assistant like Siri or Google assistant), and is then farmed out to a number of competing service-providers, each of which sees if it is able to service the request. The genie collates the results and performs some additional selection, presenting the best results to the user. The user iterates this process of creative search, making adjustments and selections. By enabling multiple algorithm contributors, such a network would exploit the creative power of human networks, and establish a truly distributed human-computer ecosystem of production.

With increased creative capability, co-creative systems may bring risks difficult to envisage. We argue here that, as technology is progressively evolving and playing important roles in the lives of people, we need to start a discussion about these potential risks in the context of creativity. We need to understand the challenges that co-creative systems may represent to humanity and consider the kind of control mechanisms that need to be put in place for advanced human-AI systems.

Recent advances in deep generative models have enabled a broad range of use cases, from drug design to music synthesis. Many of these applications will require a collaborative effort between humans who steer the generative process, and generative models to reach the desired outputs. However, our expressive power to describe interactions with these models has not kept pace. We review frameworks for mixed initiative user interfaces (Horvitz 1999} and mixed initiative creative interfaces (Deterding 2017) and identify gaps due to new capabilities produced by deep generative models. We present a new framework, Mixed Initiative Generative AI Interfaces (MIGAI), that describes human-AI interaction patterns in the generative space.

Technology strongly influences societal systems, and likewise societal perceptions influence the im-plementation of specific technologies as well as their future development. When it comes to co-creative systems, questions arise on how these systems will fit into human social systems. Therefore, a broad and pluralistic discourse about the future of co-creative systems and their societal impact should be aimed for. Our research collects crucial questions from the realm of co-creative literature and embeds them in a collaborative speculative design framework. Firstly, the framework aids collaboration between interdis-ciplinary speculators when imagining future scenari-os about co-creative systems together. And second, it supports the generation of visual speculative prod-ucts that can serve as a starting point for a broad ex-ternal audience to discuss the developments and im-pacts of co-creative technology on possible futures in a pluralistic manner.

In this paper we briefly outline some challenges for creativity domains like; creative ideation, academic research, literature-based discovery, patent creation etc. This paper has two foci; firstly on a common workflow model and secondly on the role of evaluation in the development cycle and finessing such CST

As co-creativity emerges from the interaction between two creative agents, one expects to have a mediator artifact between them. In this paper, we highlight the importance of physicality (or materiality) in the design and development process of human-computer co-creative interactions.

Co-creativity involves several agents working together to achieve a common goal. Agents are often assigned roles because this simplifies the organization and planning of labour, the selection of team members but also the ability to evaluate the agents' performance, including their degree of relevance. All these advantages can be brought into the case of mixed (human, non-human) creative team case. Agents can be assigned roles that help clarify where is it that they are trying to innovate as well as how they are to be evaluated. We can say an agent contributes to the overall creativity of the team if we can measure a variation in the teams level of creativity when we remove or substitute the agent. This measure can be used to quantify the agent's relevance within the team. In (Negrete-Yankelevich and Morales-Zaragoza 2014) we presented a framework to capture some these ideas and stress the importance of issues like roles, responsibilities and evaluation in the context of co-creativity and focused on creativity in the arts when we described roles and aspects (types of innovation). In (Valverde-Pérez and Negrete-Yankelevich 2018), the basis of a new project to experiment with models of nonanthropocentric creativity is described. This project, based on the ideas on the relationship between humans and technological objects by G. Simondon (Simondon 2011), presents a new approach to creativity where agents are autonomous in the sense of having a behavior that can be creative but only by interacting with other agents (relational creativity) and whose evaluation does not depend on human assumptions. A relational, non-anthropocentric view of creativity constitutes a paradigm shift where creativity emerges out of interaction among agents with unspecified roles that are open to the world and seek to interact with it. This creative process is triggered by mutual interest in information sharing but keeping an autonomous and independent evaluation of the relevance of inputs. We look at 1) how information relevance converges (sustaining mutual interest) and 2) how increased engagement increases the ability to detect discrepancies in order to re-frame problems and to detect unexpected links between different sources of information. This view shifts questions regarding roles, creativity quantification, goals, etc., into a more nuanced vision of asymmetrical cooperation in emergent creative processes.

The introduction of new tools in people’s workflow may promote new creative paths. This article discusses the impact of computational tools on perform- ing creative tasks. The study was conducted by a set of semi-structured interviews made to twelve professionals working on graphic design, data science, computer art, music and data visualisation. The results suggest scenarios in which it may be worth investing in the development of creativity-enhancing tools, as well as scenarios where such endeavour is not promising.