I have just put the new Gemini Nano Banana image generator through its paces. It has impressive capabilities, yet, for me, DALL-E remains the original and best. But how does it compare to one of the premier proponents of human creativity?
DALL-E debuted as a research paper with a working demo in 2021, before ChatGPT became a global phenomenon, but the two were eventually integrated in October 2023. The name “DALL-E” refers to the artist Salvador Dalí and the Pixar character WALL-E. It is striking that OpenAI selected the most eccentric humans and the most human robots to describe their AI product. Nonetheless, these choices provide the perfect starting point to explore the differences between human and AI creativity.
Salvador Dalí and his creative process
Dalí’s work is instantly recognisable, from melting clocks draped over barren landscapes to elephants on impossibly thin legs. His paintings merge the precision of Renaissance masters with the chaos of dreams. His interests ranged from religious mysticism to nuclear physics, from Freudian symbolism to his wife Gala. More than just a great painter, he is an enduring global brand who created an entire mythology around himself, complete with a museum in his hometown where he is now buried.
Dalí’s creative process was as extraordinary as his output. His “paranoiac-critical method” involved inducing hallucinogenic states without drugs to access his subconscious. He published his techniques in 1948 in an entertaining, outrageous, and revealing book, 50 Secrets of Magic Craftsmanship.
One of his signature methods was “a slumber with a key”. Dalí would sit with a key between his fingers over a plate while he induced a one-second micro-nap. Once asleep, the key would hit the plate and wake him. This is what we now call hypnagogia, which is known to induce hallucinations. He had many other tips, including making a telescope out of an urchin skeleton, organising an art studio based on a spider’s web, and deliberately wearing uncomfortable shoes.
“The difference between me and a madman, Dalí famously said, “is that I am not mad.”
He was right. His seemingly bizarre approach allowed him to maintain technical control while accessing the irrational. His induced visions were meticulously rendered with astounding technical precision, resulting in a realistic depiction of the impossible.
DALL-E’s creative process
The creative technology behind DALL-E is known as a “diffusion model”. A diffuser must be trained on the relationships between images and their descriptions using vast datasets. DALL-E eventually encodes objects, styles, spatial relationships, and artistic techniques in addition to simple labels.
The diffusion process begins with random noise, which is gradually refined into a coherent image, much like the crew of the Starship Enterprise materialising during a transporter “beam”. The model is guided by a text prompt at every step of the de-noising process, using a specialised pretrained text-image encoder to map abstract concepts like “sadness” or “minimalism” into visual directions that shape what emerges.
As such, DALL-E does not search a database of existing images; it generates entirely new depictions pixel by pixel based on the text and its semantic context. This is extremely impressive technically and genuinely creative.
Dalí vs DALL-E
Dalí and DALL-E’s dramatic differences in creative methodology, result in fundamental variances in how works are initiated, constraints are navigated, and ownership of the end product is determined.
Every element in a Dalí painting is intentional. The melting clocks represent the fluidity of time in dreams, and the crutches symbolise human frailty. Even seemingly random elements emerge from his systematic exploration of the unconscious. In contrast, DALL-E has no intentions and only passively responds to prompts. When it places a shadow or chooses a colour, it is following learned patterns, not expressing a synthetic will. Even WALL-E (DALL-E’s partial namesake) shows artistic vision in its expert repurposing of pre-existing components.
Dalí worked with physical constraints such as canvas size, paint properties, and notoriously cramped studios. He was also highly conscious of the physical limitations of his senses and hand-eye coordination, plus the natural processes of fatigue and hunger (he had a lot to say about sleep).
DALL-E has constraints of a very different kind, including compute power, data availability, ethical guardrails, and economic realities. Nonetheless, an AI diffuser can produce the 1,500 works of Dali’s whole life every hour, although the question of their relative significance and value is obvious.
Given clear authorship under copyright law, every Dalí painting was originally owned by the artist and now by his estate. A DALL-E image, however, exists in a contested space. Is it owned by the AI platform, the engineers who designed it, the person writing the prompt, or even the original creators whose works were used to train the AI model? Currently, the law requires a human author, meaning AI-generated works often fall outside copyright altogether. The issue is far from settled, and rights-holders are pushing back hard. For example, Anthropic agreed just last month to settle a $1.5 billion class action from authors. This is likely to be the first of many similar cases that will bring clarity in the medium term.
Deeper questions
Moving from methodology to the nature of the “artists” themselves, humans often create out of their lived experiences, particularly of suffering, whereas AI lacks emotions, long-term memory, sensory experiences, or social needs.
To illustrate, I listened recently to Nick Cave’s album Ghosteen, which grapples with the tragic death of his son Arthur. Cave communicates raw emotion, disorientation, and a struggle to reconcile tragedy with faith. There is no doubt that AI could “write an album in the style of Nick Cave about the tragic loss of his son”, but could it ever convey anything authentic? Or perhaps it is the audience who supplies the meaning, fusing the work with their own experiences.[1]
This raises an even deeper question: is consciousness a prerequisite for creativity? Does it further require the capacity to hold beliefs, ascribe meaning to experience, or reach toward transcendence? These are contested human attributes at the intersection of anthropology, neuroscience, and metaphysics, but creativity brings them into sharp relief. When Dalí painted The Sacrament of the Last Supper, he grappled with transcendence directly, encoding his evolving relationship with Catholicism into the geometric precision of a dodecahedron framing Christ. By contrast, when DALL-E creates, it performs statistical operations over patterns in its training data. It has no beliefs and does not understand the significance of the forms it generates.
Ultimately, what it means to be creative cannot be separated from what it means to be a creator. Until we address those questions, debates about AGI will remain speculative. Nonetheless, the lens of AI is already bringing generative new perspectives to these old questions (no pun intended).
Conclusions
Comparing Dalí and DALL-E illuminates the nature of creativity and the different creators behind it. The melting clocks of The Persistence of Memory endure not because Dalí performed brushstrokes in a striking pattern, but because a human consciousness wrestled with the subjective nature of time and found a way to make us see it too. DALL-E can generate infinite melting clocks, each unique and technically accomplished – but they are reflections of reflections, simulacra without the existential weight of the original.
Yet we should not dismiss AI creativity as mere mimicry. DALL-E represents something genuinely new: machines that can generate novel images at scale. Perhaps, then, there are at least two kinds of creativity. One, with a small “c”, that recombines existing patterns in derivative but novel ways. The other, with a capital “C”, that pushes the boundaries of human meaning and understanding. Humans can do both. For now, AI can only manage lower-case creativity.
Looking ahead, AI creativity will only grow more sophisticated. The field is advancing rapidly: multimodal systems are beginning to see, speak, sing, and compose, and may eventually take embodied form as robotics converge with software. Already, artists are experimenting with AI not as a replacement, but as a collaborator – a digital brush, an instrument, even a co-composer. Perhaps this will lead to a new era of hybrid art. Yet for now, fundamental creativity remains irreducibly human.