AI Technologies
Anit Vallurupalli
August 23, 2024
Does the extensive potential of AI come at the cost of human originality?
Now, the realm of creative innovation faces yet another emerging piece of technology: Artificial Intelligence. This time around though, a crucial question of ethics arises: does the extensive potential of AI come at the cost of human originality?
At its core, creativity is often seen as a human trait, branching from personal emotions and experiences. However, artificial intelligence has demonstrated the ability to mimic and even enhance creative human processes. Recognizing this, major companies including Alphabet, Apple and Microsoft have started to hop on the trend by investing in text-to-image diffusion models. In fact, the global AI image generation market size is projected to grow from nearly $300 million in 2023 to approximately $917 million by 2030 (AI Image Generator Market Size).
Just as the name of the technology suggests, text-to-image models interpret a user’s text input through natural language processing (NLP) and use those parameters to generate images (Kikani).
The two predominant types of deep generative models include General Adversarial Networks (GAN’s) and diffusion models. GAN’s consist of a generator and a discriminator. The generator produces samples that are aimed to “fool” the discriminator into believing they are real, while the discriminator learns to distinguish between real and generated images (Gainetdinov).
This process keeps occurring until the discriminator cannot tell the difference between the real and generated images apart, ensuring a realistic representation of the given text description (Gainetdinov).
Diffusion models on the other hand, consist of two processes: forward diffusion and reverse diffusion. During forward diffusion, noise is added to an image, gradually degrading its quality. Then, by training on large datasets of images and their corresponding diffusions, the model incorporates reverse diffusion, refining the noise until it creates a high-quality image that matches the target description (Ahirwar).
These methods are precisely how AI is able to create such realistic pieces; it combines patterns and features recognized from the thousands of analyzed artworks and pictures to synthesize visuals that align with the textual descriptions provided.
So, if AI can create beautiful pieces of art in whatever style or medium a person wishes, what does that mean for the value of pieces created by human artists?
Text-to-Image models quickly gained popularity because of their accuracy, usability and swiftness. It has become a monumental tool that inspires artists stuck in their version of “writer’s block”. It has allowed business entities to generate captivating visuals for advertisements, brand kits, and web platforms.
Users have even found an investment opportunity within the technology, generating images and selling them as NFTs (Non-fungible tokens); inspired by Pablo Picasso’s revolutionary art styles, two doctoral students from UC Santa Barbara created a neural network called thepicassoproject.
They trained the model by feeding it hundreds of the artist's paintings to produce artificial pieces in Picasso’s signature styles, ranging from cubism to surrealism (Hayden). The works are available to be purchased at NFT's, priced at $5 per piece.
There are two different perspectives when it comes to debunking the pursuit in which these two students took. Some people believe that this project is a way of honoring Picasso’s belief of pushing the boundaries of art, passing on his mindset of open-minded experimentation to the future generations (Hayden).
On the contrary, others argue that it diminishes the uniqueness and emotional depth of Picasso’s original works, reducing his art to a mere commodity.
These arguments don't just apply to the Picasso art generator though. As the popularity of image machine learning models rises, people begin to contemplate on whether the innovation is truly favorable to the evolution of art. The lack of comprehensive guidelines and regulations only add to this confusion, and it may take several years before we fully grasp how the technology will integrate into and impact our society.
But as of now, it is safe to say that text-to-image models can be viewed as an innovation or a disruptive entity, depending on the point of view.
With the development of text-to-image technologies such as Open-AI’s DALL-E and Google’s Imagen, artificial intelligence has sparked both curiosity and controversy in the world of the arts.
While some view these AI tools as a threat to traditional craftsmanship, others see it as an unprecedented opportunity to expand their creative horizons like never before. It is important to not blur the line between human creativity and machine-made content and to view AI as an extension of artistic expression, ensuring the upholding of human artistic value.
Ahirwar, Kailash. “A Very Short Introduction to Diffusion Models.” Medium, 26 Sept. 2023, kailashahirwar.medium.com/a-very-short-introduction-to-diffusion-models-a84235e4e9ae. Accessed 25 Mar. 2024.
“AI Image Generator Market Size & Growth Analysis [2030].” www.fortunebusinessinsights.com, Fortune Business Insights, Nov. 2023, www.fortunebusinessinsights.com/ai-image-generator-market-108604.
Gainetdinov, Ainur. “Diffusion Models vs. GANs vs. VAEs: Comparison of Deep Generative… – towards AI.” Towardsai.net, Towards AI Inc., 12 May 2023, towardsai.net/p/machine-learning/diffusion-models-vs-gans-vs-vaes-comparison-of-deep-generative-models. Accessed 24 Mar. 2024.
Hayden, Tyler. “UC Santa Barbara Doctoral Students Build Neural Network to Paint like Picasso.” The Santa Barbara Independent, Santa Barbara Independent, Inc., 19 Jan. 2023, www.independent.com/2023/01/19/uc-santa-barbara-doctoral-students-build-neural-network-to-paint-like-picasso/. Accessed 24 Mar. 2024.
Kikani, Kalpesh. “How Does AI Turn Text into Images? | Codementor.” Www.codementor.io, 16 Nov. 2023, www.codementor.io/@kalpesh08/how-does-ai-turn-text-into-images-2amzolymx5. Accessed 25 Mar. 2024.
