The advancements in artificial intelligence (AI) technology have increased its importance in creative fields as a creator and tool/collaborative assistant for creativity. This article deliberates on the vast potential of AI in creative domains by focusing on AI creativity technologies and their limitations.
Image Credit: The KonG/Shutterstock
An Overview of AI Creativity
In traditional creative fields like writing and painting, the application of AI is growing rapidly in both practice and theory. Processes related to the creative sector require significantly different levels of skillsets and innovation compared to routine behaviors. AI accomplishments heavily depend on the conformity of data, while creativity exploits human imagination to drive unique ideas that can divert from general rules, which cannot be readily addressed by constrained learning systems.
Many studies have been performed in the past decades to evaluate the possibility of applying AI in creative fields. Recent surveys have shown that a majority of artists in the United Kingdom, the United States of America, Japan, and Germany would consider using AI tools as assistants in non-creative tasks like searching and editing.
This shows a general acceptance of AI across the community as most AI tools have been developed for operating in closed domains where they assist humans instead of replacing them. Thus, better collaboration between AI technologies and humans can maximize the advantages of synergy.
Despite the existing limitations of AI, the first painting solely created by AI was auctioned for a hefty amount in 2018. In the last few years, AI applications in creative industries have increased significantly, with AI being used more in games, marketing, and advertising.
AI in Creative Industries
In creative industries, data compression, information extraction and enhancement, content enhancement and post-production workflows, information analysis, and content creation are the major categories of creative applications using AI. In content creation applications, AI tools are used in script and movie generation, journalism and text generation, music and image generation, animation, and content and captions.
AI tools are employed for advertisements and film analysis, text categorization, recommendation services, and content retrieval, and as intelligent assistants in information analysis. In content enhancement and post-production workflows, AI is leveraged for contrast enhancement, colorization, upscaling imagery, restoration, inpainting, and visual special effects.
Segmentation, recognition, salient object detection, tracking, image fusion, and three-dimensional reconstruction and rendering are the key AI applications in information extraction and enhancement. AI techniques commonly utilized in creative domains include recurrent neural networks (RNNs), deep reinforcement learning (DRL), generative adversarial networks (GANs), and convolutional neural networks (CNNs).
AI Creativity Technologies
Tools for Painting and Drawing: AI and deep learning techniques have been proven effective for artistic painting and drawing. For instance, AI artworks created by GANs trained using 15,000 painted portraits spanning several periods and by a robot called Sofia attracted significant interest from buyers and sold for large amounts, which indicated the AI art's commercial potential.
Google DeepDream is used by artists to create surreal animated journeys through the neural network layers by recursively processing images, magnifying details, and generating new variants. Similarly, DeepArt.io, an algorithmic artistic style transfer algorithm that provides output in the form of new works of art in the style of other artworks, is leveraged by artists in unique ways, like drawing dinosaurs made of flowers.
Style transfer techniques like DeepArt.io are embedded in many popular Android and iOS mobile apps like PicsArt editor. Tools like GauGAN can convert segmentation maps into lifelike imagery. Artists using this GauGAN only need to sketch a broadly-stroked image and ask the AI tool to fill in the details, colors, reflections, and texture. GauGAN then performs the task by referencing its vast training image set.
Tools for Writing, Poetry, and Illustrated Stories: AI tools are also gaining prominence in poetry, writing, and illustrated stories. For instance, language prediction models like the OpenAI generative pre-trained transformer (GPT) can generate human-like text using massive billion-parameter neural networks. This is possible as models like GPT-2 have learned from eight million internet web pages.
Using GPT-like AI, human writers can produce new kinds of creative output involving textual phrases that were previously impossible to realize. For instance, an artificial neural network (ANN) leveraged its training on existing candies to generate new Candy Heart messages.
Great advances in ANN encodings enable AI to generate coherent novel sentences that interpolate between two given sentences. This capability could play an effective role in a collaborative human-AI poem-writing process, where the human writer bounds the prose and the AI tool fills in the middle parts.
The AI in the Verse by Verse application effectively generates a line of poetry to follow any line written by a human writer/a particular poet as it has been trained using full-text poetry of over 12 classical-era poets. Similarly, the Poem Portraits, which combines visual art with poetry creation, creates a long-running poem in 19th-century style based on user-given words along with a visual portrait of the face of every contributor.
The Poem Portraits' ANN has been trained on more than 25 million words of poetry using a long short-term memory RNN. Additionally, OpenAI's DALL-E tool generates images from any text description by leveraging Image GPT and GPT-3. The generated images display a styling that belies the computational nature of their inception.
Tools for Photography and Portraiture: AI also enhances portraits and photography and enables new forms of visual portraiture. For instance, Super Resolution, a technique in which the ANN has been trained in narrow image categories like animals and cars, can effectively upscale small images to larger resolution images by hallucinating and filling in convincing details.
Thus, Super Resolution eliminates the practical scale constraints for human artists. Similarly, GANPaint utilizes ANN GANs to improve the consistency and manner of adding synthetic elements to photographs. Thus, artists can augment imagery using GANPaint by adding convincing new semantic elements. Moreover, the StyleGAN and StyleGAN2 generator efficiently generates unique high-quality facial images while regulating the style aspects like facial features and hair.
A Case Study
Brandmark.io is an AI technology-based smart logo-generating tool that demonstrates the technical possibilities of AI and its impact on creative processes. This case exemplifies the creative development of visual elements' design using digital technologies. The Brandmark website offers different AI tools for developing a creative logo. For instance, the AI Colour Wheel is an example of an AI-powered tool on the Brandmark web application that automatically colorizes various graphic arts like wireframes, illustrations, and logos.
Similarly, Brandmark utilizes deep learning tools to generate logos consisting of an icon, color scheme, and typography, based on specifications made by the user. The neural network approach of Brandmark can group highly similar icons, and its derived uniqueness score and legibility score can find legible and less common shapes.
Although Brandmark is an innovative and capable logo maker currently available, it has several risks and limitations. For instance, the logos created on the Brandmark website can be too generic, which is a major limitation. Thus, the biggest constraint of Brandmark is to find uncommon icons and logo ideas. This development also raises questions regarding the ability of AI technology to replace the creative input of humans. Overall, Brandmark has the potential to assist human creativity and contribute to breakthrough innovations and ideas.
Navigating Challenges
Several research challenges remain despite the success of AI tools in creative domains, including training set bias, sustainability, accessibility, and fairness. Moreover, it is yet to be determined how the notions of experience, personality, empathy, language, ethics, and consciousness relate to a machine's ability to perform as a creative partner.
Thus, AI tools must not be used in isolation as a "black box" solution. These tools must be designed as part of the associated workflow, and a feedback framework must be incorporated with the human "in the loop." In the near future, humans must check the AI systems' outputs, provide feedback on faults to adjust the model, and make critical decisions.
To summarize, the maximum benefit from AI can be derived in creative domains where the focus of the technology is human-centric. AI as a tool or collaborative assistant for creativity is more effective compared to AI as the sole creator.
References and Further Reading
Anantrasirichai, N., Bull, D. (2022). Artificial intelligence in the creative industries: a review. Artificial intelligence review, 55(1), 589-656. https://doi.org/10.1007/s10462-021-10039-7
Hisrich, R. D., Soltanifar, M. (2021). Unleashing the creativity of entrepreneurs with digital technologies. Digital Entrepreneurship: Impact on Business and Society, 23-49. https://doi.org/10.1007/978-3-030-53914-6_2
Cetinic, E., She, J. (2021). Understanding and Creating Art with AI: Review and Outlook. ArXiv. https://doi.org/10.48550/arXiv.2102.09109
Falchuk, B.(2021) How AI is Enabling a Creativity Renaissance. ACHI 2021: The Fourteenth International Conference on Advances in Computer-Human Interactions. https://personales.upv.es/thinkmind/ACHI/ACHI_2021/achi_2021_3_60_20028.html