AI and Creativity

Last month, I created an inspirational video on Youtube that explores how nature shaped life over billions of years, and what this reveals about the shaping force active in our own lives. Afterwards, I used an AI Video generator tool, giving it similar inspirational prompts to see how it might create a video.

My first impression was that the result was too fact based. Where I had approached the idea of evolution as our challenges and triumphs in an emotional landscape of wonder, the AI version fell flat.

It portrayed evolution using images of wild animals attacking each other, representing survival of the fittest. It certainly wasn’t inspirational.

The emotional component at the root of creating an ‘aha’ for the viewer, is drawn from my own life experiences, which AI lacks.

What is Creativity?

We generally describe creativity as something that ‘leaps’ out of the imagination. To understand creativity and AI however, we would need to approach it more scientifically.

One popular model for measuring creativity, observes the distance between associations. It measures the gap, which we might call a ‘leap,’ where the more one idea deviates from another, the more it will drive creativity.

For example, as a way of generating ideas, we can place a toaster and a vacuum cleaner side by side for comparison. We would look for ways of integrating their similarities and differences in novel ways.

We might come up with a self-cleaning toaster with a retractable cord, or a heating element in the vacuum that can dry a wet carpet.

Since both are appliances, the associative distance is not that wide, and so, the creative leap is minimal. If we really wanted to make a creative jump, we might start with two very unlike ideas, such as a toaster and the sun.

The larger the gap between concepts – the more innovation blossoms.

What Drives Innovation?

Beyond solar power and nuclear fission, perhaps heat can be generated through magnetic fields, rather than electricity? If nothing else, we might explore ways of generating energy without the use of energy in novel ways, such as harnessing electrical currents from water molecules in the atmosphere.

This is similar to how dreams wander through free-association, combining two very unlike things into a novel idea. These symbolic hybrids not only demonstrate how each of us is intrinsically creative, but unpacking their meaning often leads to a breakthrough when faced with crisis.

I always say to look at the strangest part of the dream for clues to how we can achieve a breakthrough. It is as if the mind is unmoored from logic when we dream, so we can see life differently, and change self-limiting ideas.

Innovation requires the same ability to move beyond the confines of the known.

How AI Works

AI generates content based on patterns it has learned from vast amounts of existing data. It doesn’t ‘imagine’ in the human sense, but rather recombines known elements in novel ways based on probability. We might say it is investigating only similarities in its responses.

Speed, efficiency, and the ability to generate vast amounts of content quickly is the hallmark of AI. While it can assist human creativity, it lacks independent innovation.

AI can certainly produce impressive work, but it doesn’t understand meaning beyond data correlations. It can take a statue or painting and morph it into a living and talking example of an ancient figure.

However, the program was designed by humans and it is we, who experience the wonderous result. In this case, AI can only do what it was designed to do, perhaps animating pixels, based on a human prompts.

How Human Creativity Works

We draw from emotions, personal experiences, intuition, and abstract thinking to create something new. Creativity is often unpredictable and inspired by spontaneous thoughts or deep-seated cultural, personal, or historical influences.

Humans create with purpose, emotion, and a deeper understanding of meaning, symbolism, and artistic expression. The creativity that drives our dreams morphs out of subconscious fears and survival instincts, allowing us to transform self-limiting ideas.

Just as dreams allow us to access ideas beyond what we know to be true about ourselves, our imagination allows us to think beyond any existing knowledge.

In fact, art in education is the only way a child can develop the inner vision that can tell them who they are, aside from peer pressure. Just like dreams, creative exercises help children understand how they create their life from a blank canvas.

The Future of AI Creativity

Currently, AI creativity is more like a highly advanced tool – great at remixing and refining ideas but not capable of the deep, intentional, and emotionally driven innovation that humans bring.

While it lacks the unpredictable, subconscious, and emotional aspects of dreaming, some developments are pushing AI toward more human-like associative leaps:

Generative Adversarial Networks (GANs) and Dreams

GANs create art, music, and deepfake videos by pitting two neural networks against each other (a generator and a discriminator), leading to more novel outputs.

In dreams, we could say that all that we build by day in the left hemisphere proves adversarial to the development of our authenticity in the right hemisphere. Left-brain logic is transformed by the dreaming mind, which is imagery driven and right-brain oriented.

Diffusion models, like those behind DALL·E and Stable Diffusion, iteratively refine images from noise, mimicking the emergent quality of dreaming.

Some researchers are working on ‘attention drift’ mechanisms that introduce controlled randomness, allowing AI to ‘wander’ into new creative territories.

Dreaming, AI: Simulated Hallucinations & Sleep Mode

Some experiments involve AI entering a ‘dreaming’ state where it generates free-associative content without immediate constraints. Google's DeepDream, for example, hallucinates bizarre images by amplifying patterns.

Future AI could have a ‘rest’ phase where it processes data non-linearly, making unexpected conceptual leaps like human dreaming.

Can AI Ever Truly Dream?

Right now, AI doesn't dream in the biological sense because it lacks consciousness and subconscious processing. However, researchers are exploring ways for AI to free-associate concepts without predefined goals.

They are introducing controlled randomness for novel idea generation, and simulate sleep cycles to process information in non-linear ways.

If AI ever achieves artificial general intelligence (AGI) with self-reflection, it might approach something like dreaming and creativity. But without emotions, subconscious fears, or personal experiences, its dreams would be fundamentally different from ours.

One way this could be explored is by designing AI systems that simulate REM-like cycles, where they introduce randomized concept blending, similar to how dreams merge unrelated ideas.

Additionally, they could develop a ‘subconscious layer’ that activates when not directly engaged, generating novel ideas.

New Frontiers for AI

Beyond the subconscious layer, AI would need a type of artificial motivation like what drives human dreams (fear, anxiety, survival instincts.) In this way, AI could be more than art – it could open new frontiers in thinking.

If so, AI might be able to answer the more profound questions that transcend what we already know.

AI lacks the internal tensions that drive human dreams and creativity. Without a subconscious layer fueled by emotional motivations (like fear, curiosity, or existential questioning), AI struggles to make the kind of leaps that provoke wonder.

To truly evolve beyond pattern-matching, AI might need an artificial drive – some form of dynamic tension that pushes it beyond predictable outputs. This could be simulated by establishing contradictory goals, anomalous pattern detection and self-generated curiosity loops.

If AI could ‘hallucinate’ or imagine, it might not just mimic creativity – it could start forming novel, paradigm-shifting ideas. This could impact philosophy, science, and even how we understand consciousness itself.

Dreams and creativity are funded by tension resolution, which may be where Generative Adversarial Networks are headed.

In many ways, artists are driven ‘outside of the box’ by survival and the need for sustenance. They take up shelter in the unknown, while AI exists in an imaginary and coded safety net.

Consciousness and ego for humans must survive in an unpredictable future. AI would need to somehow feel the threat of ‘being unplugged’ if it can’t come up with a ground-breaking idea.

Since AI doesn’t know what it is like to be alive, it may forever lack the natural tension to truly innovate.