The Picasso Paradox: How AI and Human Creativity Are Converging
By Sairam Sundaresan
A seasoned AI research scientist by day and creative storyteller by night, Sairam Sundaresan transforms complex AI concepts into engaging and accessible narratives. He is on Substack, LinkedIn, and X at @dsacience.
In a quiet studio, a painter stands before a blank canvas, brush poised. Miles away, a generative AI algorithm hums within a sleek server. Both are ready to create, but their methods could not be more different. Or could they?
As AI becomes increasingly sophisticated, it isn't replacing human creativity—it's converging with it in surprising ways. Both humans and machines rely on a fundamental creative principle: remixing the old to create the new. By examining this convergence, we can glimpse the future of creativity itself—a future in which humans and AI collaborate to remix ideas in powerful new ways, reshaping the very nature of art, music, and design.
This convergence of AI and human creativity is already happening in various fields. Consider Tim Fu, an architect who co-designed futuristic streetwear with AI, and the recent partnership between OpenAI and Shutterstock to bring generative AI capabilities to all. These examples show that AI isn't just for analyzing trends—it's helping creatives produce bold, fresh pieces that push the boundaries of what's possible.
But how does AI remixing happen?
This process closely mirrors how human creatives work. Imagine a chef experimenting with new ingredient combinations or a perfumer mixing scents to create a unique fragrance. They start with existing elements, but through experimentation and refinement, they arrive at something original. The key difference is that AI can perform these creative remixes at vastly greater speed and scale.
"Good artists copy, Great artists steal." – Pablo Picasso (maybe)
Google DeepMind's Lyria is a good example—an AI model that generates music from text like an automated composer. It pulls together genres, instruments, rhythms, and even vocals to create unique pieces of music. With a tool like this, millions of people like you and me can make music by simply texting a model our preferences. Under the hood, it's like a jazz musician improvising on stage, blending what it knows into entirely new expressions at a previously unimaginable scale.
Wait a minute. How does it know?
Painters learn by studying masterworks. To excel, they meticulously imitate each brushstroke. They carefully mix colors and layer the composition just like in the original. Over time, they can replicate masterpieces at a high level. However, during this process, they also learn how to innovate. They learn taste while copying several paintings. They begin to incorporate more of what they like and less of what they don't. They begin to remix. AI systems like DALL-E, Midjourney, and Adobe Firefly are trained on vast datasets of existing content (called training datasets). They repeatedly see the same examples over and over again to learn from them.
How? These generative models are shown a partially complete image and asked to finish it. When they do, they are graded on how well their version matches the original. The same applies to language models like BERT or GPT. This process is like solving a jigsaw puzzle. By hiding parts of an image or sentence, the AI is forced to predict what's missing, training itself to understand the context. This makes models adept at filling gaps—whether completing an unfinished poem or generating a visually coherent landscape from just a few words.
On seeing a prompt, the model believes it’s been handed a new jigsaw to solve. It uses its experience from the training process to generate new text, images, and even videos. So, when Midjourney generates a surreal image—a room filled with dream-like colors and impossible shapes—it's not creating from nothing. It's recombining visual elements it's "seen" before, producing something that feels both new and oddly familiar, like collage artist Deborah Roberts or music producer J Dilla stitching together obscure samples into fresh beats.
What's both awe-inspiring and scary at the same time is how fast these models are improving. This is thanks to another machine learning technique called transfer learning. Just as an experienced chef who's mastered French cuisine might more easily learn to prepare Japanese dishes, AI systems use transfer learning to adapt what they've learned from one domain to another. For example, if a model has been trained on images of animals, landscapes, and buildings, it can quickly adapt to generating fashion designs. All it needs are a few examples of this new task. Through transfer learning, it can pull from a reservoir of prior knowledge on general objects—patterns, silhouettes, and colors to create new designs.
Think of artists or writers who do residencies. They already have a wealth of experience, and during their residency, they lean on this experience to adapt and improve. Now, imagine an amateur doing the same thing. Wouldn't they get much less out of it?
As AI and human creativity converge, it raises profound questions about the future of art, music, and design. Will we see more AI-human collaborations where machines help amplify and augment human creativity? Will AI-generated content become indistinguishable from human-made works, blurring the lines between artist and tool? These are the questions we must grapple with as we step into this new era.
But this convergence also brings challenges. In 2022, Midjourney's AI-generated artwork won an art competition, sparking debate over what constitutes creativity and authorship. Was this AI-human collaboration "creative" in the same way as a painting by Rembrandt? Is AI the artist or just a tool? Is it just the same as a photographer setting up their camera and pressing the shutter? As machines become our creative partners, we may need to rethink our definitions of creativity, authorship, and originality.
Questions around consent and attribution are even thornier. Let's revisit the Lyria example. It can generate compositions that sound like your favorite artist. Does the original artist have a right to this generated music? Recently, actress Scarlett Johansson claimed that OpenAI used a voice for Sky that closely resembled her own without permission. This similarity was particularly noticeable to her friends, family, and fans, who found the AI voice eerily similar to Johansson's, incredibly reminiscent of her role as the AI assistant in the movie "Her". In an age of AI remix, issues of consent and attribution become incredibly complex.
Despite these challenges, the remix revolution offers thrilling possibilities. Imagine Runway Gen-2—an AI that generates video clips from text prompts—helping architects visualize sustainable buildings or AI remixing molecules in the lab to create life-saving drugs. AI tools could democratize creativity, enabling more people to express themselves, even without formal training.
But to unlock this potential, we'll need robust systems for tracking attribution and rewarding original contributions, even as they're endlessly remixed. Content fingerprinting, like the Content Authenticity Initiative (CAI), can embed secure provenance trails into images and videos, ensuring creators receive proper attribution. Watermarking can help distinguish AI-generated content and track its use. Several big tech companies like Google, Meta, OpenAI, and others have developed their own watermarking solutions. These technologies could form the basis of automated royalty systems, where original creators are compensated each time an AI remixes their work.
We also need policy and regulation. Governments and international bodies should set guidelines requiring AI developers to use these attribution technologies. Regulatory frameworks can ensure that models trained on copyrighted content include mechanisms for tracking and compensating creators, protecting the integrity of creative works.
But technology and policy alone aren't enough. We must also cultivate human skills that complement AI, like critical thinking, emotional intelligence, and ethical judgment, as we navigate this new landscape. Assessing whether an AI-generated piece is used responsibly or if its origins are acknowledged will be key to fostering a healthy creative ecosystem. Moreover, as AI takes over more routine creative tasks, human creators may need to focus on higher-level skills like problem-framing, concept development, and creative direction. They'll need to become adept at guiding and curating the creative process, working in symbiosis with AI tools.
This shift could lead to new roles and specializations within creative fields—AI trainers, prompt engineers, algorithmic curators—that require a blend of technical and creative skills.
Educational institutions and professional organizations must adapt curricula and training programs to prepare creatives for this new landscape. This might include courses on AI ethics, collaborative workflows with machines, and techniques for leveraging AI while maintaining human authorship and intent.
Imagine a future where AI and human creators collaborate seamlessly. AI handles the grunt work of generating endless variations and novel combinations, while humans guide the high-level vision, curate the best outputs, and imbue the work with emotional resonance. Automated attribution systems ensure that every contributor, human, and machine, is recognized and rewarded fairly.
In this future, creativity is not a mysterious solo act but an open collaboration between humans, machines, and our shared cultural heritage. By thoughtfully steering the convergence of AI and human creativity, we can expand the boundaries of what's possible and create works of art, music, and design that we can scarcely imagine today.
The painter makes a first mark on the canvas. The AI outputs its first take on the prompt. Both are remixing the old to create the new. They are not rivals but partners in a grand, unfolding act of co-creation.