About MICIs

Creativity support is an important area of computing and human-computer interaction (HCI). Historically, work on such creative interfaces broadly clusters around two ends of a spectrum (fig 1.): on the one end, we find traditional creativity support tools like computer-aided design or search, visualisation and collaboration software for creative work. Here, the human is the initiating and deciding agent and the computer a mediating tool. On the opposite end sits computational creativity: the computer is a ‘heroic’ artificial intelligence (AI) agent that autonomously produces creative work, and the human its audience.
The spectrum of human-computer initiative in creative interfaces

Recent years have seen the emergence of a third paradigm between the two: mixed-initiative creative interfaces (MICI) where human and computer interact as collaborators in a tight feedback loop. As in human-human creative dialogue, both sides take turns constraining, suggesting, producing, evaluating, modifying, or selecting creative outputs in response to the other, such that creative agency and initiative cannot be easily ascribed to one side alone. While all creative practice is to some extent a ‘dialogue’ between creator and material, MICI literalise this metaphor by giving the computer the status of creative agency and initiative thanks to artificial intelligence.

Tanagra is a MICI 2D game level generator: the designer adds, removes or modifies high-level ‘beats’ of gameplay and their length (like ‘jump to kill an opponent’), the AI executes level geometry in detail, generates beats undefined by the designer, and continually tests whether the level remains playable.

The vision of augmenting human problem-solving by sharing initiative within a larger human-computer “symbiont” reaches back through the history of HCI and AI to pioneers like Licklider, including early work on dialogue-based interaction with conversational agents and intelligent tutors. Yet in creativity support, it is only today that we find this vision realised, chiefly in the field of procedural content generation (PCG) for games. Here, a lively community of researchers and practitioners is currently exploring AI techniques such as evolutionary computation, heuristic search, or machine learning to (semi-)automatically create and evaluate art assets, game levels, or even entire games (see figures).

Current mixed-initiative, human-AI co-creativity in PCG cover a fraction of possible MICI scenarios. Still, these systems already show a number of highly attractive features: they vastly accelerate the iterative exploration of solution spaces, which also enables divergent exploration that would be prohibitively time-consuming otherwise; they enable and accelerate learning and understanding creative practice through probing and hypothesis-testing; they provide rich lateral stimuli; and they can make creative practice accessible and enjoyable to non-professional and even disabled user groups, with rich benefits to personal wellbeing and societal inclusion.

Yet although MICI and their advantages easily generalise across creative practice, they have so far remained largely disconnected from creative interface researchers in HCI. Examples of MICIs outside games demonstrate the relevance of the approach to a broad range of creative contexts under active consideration by HCI researchers, including: urban design, sketching, interface design, prototyping musical instruments, or data visualisation. Indeed, with the current rise of AI, (semi-)autonomous systems, and conversational agents, human-AI mixed-initiative presents a generally valuable and under-explored interface paradigm.

While HCI could richly benefit from current MICI work in game AI, MICI in turn present a number of formidable interface and interaction design challenges: enabling creators with little computational literacy to readily express ideas, constraints, or criteria in a formal(isable) and thus, computable manner; avoiding user fatigue and making interaction with the machine engaging; rendering computer design decisions legible and transparent to users; identifying visualisation patterns for computational design evaluations and modification and selection suggestions. Since MICIs in games have been chiefly designed by AI researchers for similarly computationally literate user groups, they would greatly benefit from the design expertise of HCI researchers.

Workshop Goals

In summary, mixed-initiative creative interfaces, as currently developed particularly in procedural content generation for games, hold rich opportunities for creative and (semi-)autonomous interfaces far beyond games. At the same time, they present unsolved interface and interaction design challenges that would immediately benefit from HCI and interaction design practices and patterns in creative interfaces. The goal of this workshop is therefore to bring together researchers from the game AI and CHI communities to advance the MICI paradigm by exploring its opportunities for HCI beyond game design and address its unsolved interface design challenges.

Workshop Questions

By creating dialogue across communities, we wish to identify first answers to and facilitate collaborations around questions such as:

  • How do people experience and interact with MICIs?
  • How can we evaluate the quality of human-AI co-creativity?
  • How do MICIs and human-AI co-creativity challenge and advance theories of creativity support tools, computational creativity, and human-human creative collaboration?
  • What interaction and interface design challenges do MICIs hold? What established interface and interaction design methods, principles, and patterns can address these?
  • What AI challenges of MICIs have HCI solutions, what HCI challenges have AI solutions?
  • How can MICIs up-skill not de-skill end users?
  • What new kinds of human-AI co-creativity can we envision across and beyond creative practice?
  • How do opportunities and challenges of designing MICIs relate to different genres and forms of creativity?
  • How can we extend creative relationships between humans and AI beyond collaboration, such as competition (e.g. DJ battles), rivalry or disruption?
  • How might MICIs further accessibility, inclusion, and participation?

Want to know more?

Read the workshop paper:

Sebastian Deterding, Jonathan Hook, Rebecca Fiebrink, Marco Gillies, Jeremy Gow, Memo Akten, Gillian Smith, Antonios Liapis, and Kate Compton. 2017. Mixed-Initiative Creative Interfaces. In Proceedings of the 2017 CHI Conference Extended Abstracts on Human Factors in Computing Systems (CHI EA ’17). ACM, New York, NY, USA, 628-635. DOI: https://doi.org/10.1145/3027063.3027072