Generative Adversarial Networks–Enabled Human–Artificial Intelligence Collaborative Applications for Creative and Design Industries: A Systematic Review of Current Approaches and Trends

Tags

Top Highlights

• Zeng et al. (2019) saw AI being integrated into the design cycle. It shows how the human–AI relationship can be collaborative;

• The families of generative models can be broadly categorized into two distinct categories: explicit density and implicit density models.

• heir system generates design variety from user input, which can then be explored by the user and incorporated into the next iteration

• systems that can generate landscape paintings

• or terrains

• An optimally trained GAN generator should recreate the training distribution and therefore cannot directly generate an image based on new governing principles because such an image would not be similar to anything like it has seen in its training data. Therefore, one must ask if users would be prevented or discouraged from exploring more exciting directions.

• how GANs are being integrated into design support tools.

• One of the critical human–computer interaction (HCI) considerations when creating these tools is how the end-user will communicate or interact with the system

• The reviewed articles present various interface modalities, with the most common being sketch-based interfaces

• most articles are situated within the art-design space

• Natural-language user interfaces (NLUIs)

• two fundamental modes of operation became apparent: variation and beautification.

• explore machine-generated variations

• The machine then generates a selection of design variants. The designer can then examine the variants and select one or adapt their design, taking inspiration from the generated examples

• beatification

• designers provide the system with course level input (e.g., sketches, graphs, and language-based instruction), and the system outputs a more fully realized design (e.g., image, landscape, and game level).

• Due to the complexity of AI and ML, from an algorithmic and architecture perspective, there is a gap in knowledge between interaction/user-experience designers and ML engineers when it comes to understanding ML’s limits, what it can and cannot achieve

• while also identifying a range of limitations in this field of research, primarily finding a lack of focus on the end-user when developing training sets and designing interfaces, and limited outcomes in terms of scalability or professional usability.

• f this technology is going to make the break-through to mainstream adoption, a stronger focus on collaboration and the end-user is needed