Autumn 2022 was when Artificial Intelligence designed to facilitate the creation of art, design, and writing began to rapidly ‘grow a beard’. An idiom popularised by Star Trek fans, ‘grow a beard’ refers to the point at which a television series suddenly and significantly improves in quality at the same time a key character grows a beard. In 2023 AI ‘beard growth’ is exponentially beyond where it was less than a year ago. It is rapidly changing how we create.
Obvious or telltale signs of AI-human collaboration can be desirable. Glitch Art and the 2019 AI rug collection by Shame Studios, London are two such examples. LLM (large language models) software chatbots create ‘natural’ human language text and conversations. The launch of LLM ChatGPT at the end of 2022 disrupted word-based media. Shallow fluency is an LLM drawback, as is its proclivity to ‘hallucinate’—to make up facts and references. Lucy Upward’s comment piece in this issue’s News section (page 33) was written by ChatGPT following her prompt: ‘In 500 words discuss the benefits that AI could bring to the designing of designs for contemporary handknotted carpets made in places such as India and Nepal.’ The resulting article includes ChatGPT’s false statement: ‘With the advent of Artificial Intelligence (AI), the potential for revolutionizing the process of designing these carpets has become evident.’ The ‘potential’ is not yet ‘evident’ and the ‘revolution’ has only just begun.
Lucy asked me to explore how AI may impact the rug world. In turn, I invited Hector Coombs, founder of Shame Studios and an early adopter of AI, to share his AI rug journey. This article is illustrated with prompt-based AI generated images created by Coombs. The images feature real rugs from Shame Studio’s AI capsule collection of flatweaves designed by Coombs using GAN (generative adversarial network) databased-AI.
Four years ago, publicly available AI was rudimentary. In 2019 Coombs happened upon the website This Horse Does Not Exist (now defunct) where users could generate a photograph-based horse by refreshing the page. ‘It would sometimes look like a horse’ Coombs remembers, ‘but the next refresh might mangle it while the next could give you something quite alien looking.’ Online image hosting service Imgur’s archive of these horse images range from recognisably ‘horse’ to body parts randomly stitched with extra limbs. ‘It’s always a challenge to put a new spin on contemporary rugs, so I figured if AI can do this for a horse, it can do it for a rug.’
Coombs began to build an image database. ‘We started with a range of rugs, but the rugs “it” designed were a mess like the horses. So we narrowed the database to about 500 tribal kilims.’ The database learned the rug patterns and generated designs that could hide in plain sight with the real kilims. The next step was ‘outpainting and uncropping’ meaning ‘logical extensions’ which expand an image beyond its original borders.
‘AI picked up on the “mistakes” and inconsistencies of the database rugs and amplified them whilst adding its own glitches,’ says Coombs. ‘We translated the images into three designs that our weavers in India could understand and produce. Almost instinctively they began to correct the AI designs to be a bit more regular whilst also making their own mistakes and design choices. So the rugs became like a game from machine to human and back again, with each participant adding their own understanding and choices to the process.’ What Coombs describes is ‘infidelity in memetic replication’; a process when cultural information transfer alters based on perceived Darwinian evolution. The New Scientist in July 2023 predicted the exponential growth of captioned AI-generated images uploaded to the internet has the potential to create a doom loop of declining replication and decreasing visual diversity as generative AI cannibalise their own images.
New technologies have disrupted and then assisted the creative process for hundreds of years. ‘The Beggars Complaint: Some observations on the Conduct of the Luddites in Reference to the Destruction of Machinery’ was published in 1812, the year after English artisan weavers and knitters first rioted for the destruction of machinery that was replacing them or causing wage deflation. The name of their spiritual leader —Ned Lud—evolved into the pejorative ‘luddite’ used today to identify a person perceived to be against technological progress. More than 200 years after the riots, sophisticated professional rug design software is used by the majority of rug and carpet designers and manufacturers. Coombs reflects on these changes. ‘I’ve seen our industry move from sending hand drawn designs back and forth between designer and producer to now designing a perfect pixel-to-knot naksha [pattern] weavers can follow with very little room for interpretation.’
The author of the The Beggars Complaint believed whoever ‘invents any thing that will lessen human labour, is a benefactor to mankind . . . and the public [will benefit] by the free use of such inventions.’ Coombs confesses he finds the potential of AI ‘absolutely terrifying’, but does he also believe there is potential for public benefit? ‘It’s democratised aspects of design; we couldn’t a ord to create the images used in this article; a photo shoot like this would cost £5,000 at an absolute minimum.’ But Coombs questions the future of AI design. ‘I really enjoyed AI when it was quite messy, but as it gets so good at generative creative ideas, I’m not sure it’s in line with our values as a company. I’m very proud of our AI rug project, but the deep and rich cultural history of tribal kilims couldn’t be further away from AI.’
The acronym GAN includes the word ‘adversarial’. It describes how two neural networks vie in a form of ‘combat’ to create AI data that mimics what a human might create. Coombs and I agree that artists, designers and writers need to be in combat mode and learn to use AI so it remains a useful assistant and ‘friendly adversary’ and doesn’t evolve to be a combative competitor with a ‘beard’ that never stops growing.
