Human creativity finds new AI expression, but debate rages on future of art

On November 18, Dubai’s Museum of the Future had on display a unique piece of art titled The Dreamcatcher. A 3D generative sculpture, it was created by Fabin Rasheed, an alumnus of IIT Hyderabad, using his brainwaves and a photograph of his subconscious.

At the exhibition, Rasheed went into a state of meditation while wearing an EEG device, which translated his brainwaves into his computer and generated vibrant feathers for The Dreamcatcher. While still in that state, he also zeroed in on an image in his subconscious — a “city with a circle around it”. Then, to create a visible picture of what he’d seen in that hypnagogic state, he fed a prompt into an image-generation artificial intelligence (AI) model. A digital image of the city of his subconscious emerged, which he placed at the top of The Dreamcatcher. With that, his sculpture was ready (See Picture).
 

Art knows no boundaries. Neither does technology. In recent years, the two have come together in ways that have artist-technologists innovating like never before; galleries and auction houses sitting up and taking notice; novices, including school students, “generating” art; and traditionalists cringing.

Giving a whole new dimension to this confluence of creativity and technology is AI, which is penetrating every sphere of human life. So, why should art remain untouched?

AI-generated art isn’t new. Back in October 2018, an AI artwork titled Portrait of Edmond Belamy had sold for $432,500 — nearly 45 times its upper estimate — at British auction house Christie’s. It was the first AI-created artwork to feature in a Christie’s auction. That year, Delhi-based gallery Nature Morte, too, held its first AI art exhibition — a series titled The Anatomy Lesson of Dr Algorithm.

Inspired by Rembrandt’s painting, The Anatomy Lesson of Dr Nicolaes Tulp, it was based on 60,000 images of human surgical dissections which the Bengaluru-based artist, Harshit Agrawal, had curated and fed into a software algorithm. There have since been other such AI-based exhibitions.

But what’s new now are the tools, such as DALL-E 2, Midjourney and Stable Diffusion, which can conjure up AI-generated art in a matter of seconds. Many of these tools are not even a year old, and have turned even amateurs into creators of complex, abstract works. All they need to do is key in some text, any text, and voila! 

Rasheed, whose works are carefully conceptualised, layered and laboured upon with, of course, help from AI, explains how this works. “It depends on the AI model used,” he says. “For example, DALL-E converts text prompts into a corresponding image equivalent using a model called CLIP. CLIP essentially is trained on a dataset of annotated images to link textual semantics to visual representations.”

Kunal Bhatia, co-founder and CEO of Hexo, which helps build custom image generation engines, demystifies it further. “Image generation models like DALL-E 2 and Stable Diffusion have been trained on billions of images scattered around the internet, and they have learnt millions of ‘concepts’ of things — ranging from famous people, art styles, to correlated words. And they are now able to create new images based on text inputs that refer to these concepts,” he says. “Stable Diffusion, for example, has been trained on 5 billion images sourced from the internet.”

What the AI model does is find correlated patterns between words typed in the captions and the corresponding images, he adds. “For example, if the AI model is fed 100 images created by Picasso, it will learn the common pattern in Picasso’s style of painting. Or, if the AI model is fed 100 images of Elon Musk, it will learn what Musk looks like.”

So, you can, say, create an image of “Elon Musk in Picasso style” from the memory of what Musk looks like and what Picasso’s style of art looks like. “These are not cut-paste solutions,” Bhatia clarifies. “These images literally did not exist before the text prompt was fed in as an input and were created the way an artist would create.”

That addresses the creativity question. But what about uniqueness, which is also at the heart of art? What if, say, two people type in the same search term — for example, “Van Gogh 9/11” — at the same time and from two adjacent computers? Will AI then give the same result?

“In most cases, it will be somewhat different,” says Nitin Sharma, general partner and co-founder at Antler India, an early-stage venture capital firm that invests in emerging technologies. “AI operates on certain mathematical principles, which enables it to map out inputs to certain vectors in the backend, which drive the entire generation process.”

Bhatia, too, says the probability of getting the same output is extremely low, since there are billions of possible outputs that AI can create for the same input text prompt. “This is because the output image is dependent on two factors — (a) the input text prompt, and (b) a ‘seed’, which is a starting point for the model to begin creating the image and is represented by a number,” he explains. “Since this number can be anything, the images created can have as many variations as the seed. This seed can be randomly generated or can have a user-defined input. Now, if two people enter the same input text prompt as well as the same seed, they will get the same image as an output.”

Portrait of Edmond Belamy, which sold for $432,500 at Christie’s in 2018. It was the first artwork created using artificial intelligence to be featured in a Christie’s auction

Is there a market for AI-based art?

“Yes, there is, especially through NFTs (non-fungible tokens),” says Rasheed. “I have sold many AI works in the past. For example, one of my works, called Humachine Imaginarium, which I created with my wife Ashna Sahir, was an early exploration into what a language model like GPT3 visualises. This sold for $15,000 two years back.”

Christie’s, too, has launched Christie’s 3.0, an on-chain auction platform dedicated to exceptional NFT art. And Christie’s Contemporary Art Department has designated specialists who are in charge of all digital media developments, including AI, says Suzy Sikorski, a Dubai-based Christie’s specialist in Middle Eastern 20th and 21st Century Art. However, “these works are mainly offered in the primary art market and less in the secondary market where auction houses operate,” she adds.

Some now claim AI is the future of art. Others pooh-pooh the very idea. Yet others believe AI would be just another expression of human creativity.

“AI-generated art should not be seen as an alternative to artwork generated by humans,” says Ratul Sharma (name changed), a PhD student of fine arts at the University of California, Los Angeles (UCLA). “While a few galleries and collectors might pay a hefty price for AI-generated art, the value of such artwork primarily relies on its novelty. The depth and complexity of manmade artwork cannot be rivalled by AI.”

There is also the matter of ethics and possible copyright violations.

Says Prashanth Kaddi, partner at Deloitte India, “There are websites that are putting a ban on AI-generated art due to controversies it might attract in terms of ethical and copyright implications.” AI models, he adds, are trained on thousands of pieces of human-generated art, creating ethical debate and the fear of AI-generated art replacing the artist. “However, no AI has so far delivered the desired results without human intervention.” 

Sikorski of Christie’s agrees that these are early days for AI-based art. “We cannot yet judge its longevity and impact on the art world.”