Abstract
The future of work and workplace is very much in flux. A vast amount has been written about artificial intelligence (AI) and its impact on work, with much of it focused on automation and its impact in terms of potential job losses. This review will address one area where AI is being added to creative and design practitioners’ toolbox to enhance their creativity, productivity, and design horizons. A designer’s primary purpose is to create, or generate, the most optimal artifact or prototype, given a set of constraints. We have seen AI encroaching into this space with the advent of generative networks and generative adversarial networks (GANs) in particular. This area has become one of the most active research fields in machine learning over the past number of years, and a number of these techniques, particularly those around plausible image generation, have garnered considerable media attention. We will look beyond automatic techniques and solutions and see how GANs are being incorporated into user pipelines for design practitioners. A systematic review of publications indexed on ScienceDirect, SpringerLink, Web of Science, Scopus, IEEExplore, and ACM DigitalLibrary was conducted from 2015 to 2020. Results are reported according to PRISMA statement. From 317 search results, 34 studies (including two snowball sampled) are reviewed, highlighting key trends in this area. The studies’ limitations are presented, particularly a lack of user studies and the prevalence of toy-examples or implementations that are unlikely to scale. Areas for future study are also identified.
Highlights
The emergence of artificial intelligence (AI) and machine learning (ML) as a crucial tool in the creative industries software toolbox has been staggering in scale
The progressive growing of GANs (P-GAN) algorithm (Karras et al, 2017) was applied to produce a set of generated designs of varying resolutions outputted at three discrete training epochs
Due to the computational cost, architecture complexity, and difficulty in training GAN models, the current state of the art outputs images at 512 × 512 px. While some of these results are very impressive and have garnered much media attention, how much real-world value and penetration will these systems achieve without a marked increase in visual fidelity? There is no doubt that the quality of results will continue to improve as architectures evolve, but it remains a major limiting factor
Summary
The emergence of artificial intelligence (AI) and machine learning (ML) as a crucial tool in the creative industries software toolbox has been staggering in scale. Feature selection, preprocessing, model development, parameter tuning, and final assessment of the resulting model’s quality are all made without consulting with end-users on how they will interact with the resulting system This has led to systems where the enduser involvement consists of little more than providing some input and hoping for a good result. There have, been many excellent recent surveys on the state of the art (Alqahtani et al, 2019; Hong et al, 2019; Pan et al, 2019; Khursheed et al, 2020), performance (Kurach et al, 2018), advances in image synthesis (Wu et al, 2017; Wang et al, 2019), and approaches to improving stability (Wiatrak et al, 2019) These reviews showcase the prevalence of GANs in research and indicate it as a growing area of importance. The systematic literature review methodology will be described, and the results of the review will be presented
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
