Co-designing with AI: Design Approaches to Creativity Support with Imbedded Artificial Intelligence

As functional AI becomes more accessible, there is a growing recognition that design methods,
particularly within Human Computer Interaction, need to adapt to factor computational
intelligence into design outcomes. Holmquist (2017) considers emerging practicable
implementations of AI as "a new design material" and defines some challenges to designing with
this material, which include "Designing for unpredictability", "Designing for evolution", and
"Designing for shared control".
Yet the challenges of AI can also represent opportunities. By shifting 'designing for..' to
'designing with..', these particular challenges of AI could potentially be reframed as
methodological advantages (e.g. unpredictability within the ideation stage of a project leading to
unique and creative responses). Underlying these factors is the autonomy and "proactive
intelligence" of AI (Holmquist, 2017). It is here that the 'material' metaphor encounters some
limitations, with AI conspicuously lacks the consistent and predictable physical properties that
define materials.
In contrast Manovich (2013) describes digital experiences being "constructed by software in real
time" as momentary "software performances". Considering the computer system as a performer
(an actor following a script, making decisions and assumptions on behalf of the system designer
and in response to audience actions) has particular resonance with the heightened autonomy of
AI. In this enfranchising view of AI it seems more appropriate to treat it as participant rather than a
material during the design process.
Despite the emerging nature of practical AI systems, there is already an established body of
research examining this approach. For example Boden's (1994) concept of programmatic agents
for creativity, Lubart's (2005) personified categories of computational creative collaborators (in
particular the role of "computer as colleague") and many of the creativity support systems
documented by Gabriel et al (2016).
Furthermore, the introduction of AIY Kits (Google, 2018) provide a practical, adaptable, and
usefully embodied method of representing AI in the design process.