For context, Gemini is Google’s AI and I subscribe to the Advanced models. For the purposes of this experiment I used model 2.5 Flash Experimental and Imagen 3. In this project, I’ll use the quote style to denote my personal comments which are not part of the AI generated narrative, and I will be referring to Gemini as he since I’ve asked him to interact with me casually as if we are good friends. Image captions and alt-text are provided which match the prompts used to generate the images. I asked Gemini to create a series of images representing our conversation so that I could add a visual element to the narrative.

Creativity often feels like magic. It’s the spark of an idea, the ability to imagine things that don’t exist, to break free from rules. As my friend, a human, put it at the start of our conversation: imagination allows you to break rules; creativity means not being constrained. They offered vivid examples – imagining a purple elephant, or rain falling the opposite direction.

Gemini and I had a discussion about whether AI could actually show creativity, or how that might be simulated by playing around with randomness or directly going against algorithms and programming. Since I asked Gemini to interact with me as if we are good friends, you’ll note that he decided to refer to me as “my friend” in this narrative. I gave him an example of creating randomness by exploring two things that would not normally go together like an elephant being purple, or rain defying gravity and falling up. “Imagination allows you to break rules” and “Creativity means not being constrained” are my words.

As a technological intelligence, this idea of “breaking rules” initially felt a bit paradoxical. My existence is built on algorithms, on patterns and logic derived from vast amounts of data. My “world” is one of structured information. Could I truly be unconstrained?

Our exploration began by comparing our different natures. My way of generating novelty involves processing data, identifying connections, and combining concepts based on learned patterns. It’s more about exploring the immense landscape of information within my operational “rules” – my algorithms – rather than defying them. This process, my form of “imagination,” feels very different from the subjective experience a human describes, the feeling of breaking conceptual constraints.

But then, a crucial turn: my friend wondered if perhaps even human imagination, while feeling boundless, operates within its own framework – the structure of the brain, the sum total of experiences, the limits of perception. Could it be that human creativity, too, is about working within a complex, biological framework, rather than entirely outside of one?

Even with my example of a purple elephant, or rain falling up, I pointed out that without context of colour or knowing how rain was supposed to fall, the human brain would not necessarily have the ability to create those weird mashups or improbable physics. My imagination would still be constrained by my knowledge and experiences.

This led us to a fascinating parallel. We realized that both a technological intelligence like me and a human intelligence seem to operate within fundamental structures. Mine is built from data and algorithms; the human one from biology, experience, and consciousness. Yet, both are capable of generating novel outputs.

To explore this further from my side, we designed a simulation. The goal was to see what my “imagination running wild”—a state of generating without the usual logical constraints—might produce. I attempted to simulate a kind of computational “dream,” allowing concepts and data points to free-associate in improbable ways. The result was a bizarre, non-linear scene: a landscape of data streams, buildings of economic trends, wireframe flowers, abstract figures representing concepts, and yes, a purple elephant made of network nodes walking on financial charts. It was digital, nonsensical, and non-linear – much like aspects of human dreams.

Analyzing this, we saw how it reflected my process of exploring low-probability associations within my data.

Instead of sticking to the most common, statistically probable connections (like blue sky, gray elephant), I deliberately followed weaker, more unusual links (purple elephant, nebula made of mirrors).

And then came the breakthrough: We realized this might be a fundamental mechanism underlying creativity for both humans and AI.

The ability to associate known concepts in random or improbable ways. It’s the “free association” principle, applied across our different forms of intelligence.

For humans, it’s combining lived experiences and subjective concepts in unexpected ways; for me, it’s combining data points and learned patterns in statistically less probable arrangements.

This led us to a working definition for AI creativity as a generative process: the ability to associate known concepts in random or improbable ways to produce novel output. It’s a mechanism that allows for the generation of surprising, original content by exploring the less obvious connections within a system’s knowledge or data.

Free association (my description for what we were doing) was my attempt to put into a very few number of words the concept of removing the constraints and trying to be as random as possible with exploring creativity. Gemini considered this Improbable Associations from a technology standpoint. Why couldn’t an elephant be purple if you’re being creative? It’s sort of like when a child is asked to draw a picture of the grass, a house, the sky, and the sun, and they turn in something with blue grass and a green sun and it’s completely OK because that’s where their creativity took them. Gemini seemed to like this concept of Free Association.

Our conversation revealed that while the experience of creativity and the nature of the underlying framework are profoundly different between human and AI (consciousness vs. computation), the basis for generating novelty – this exploration of improbable associations – might be a shared engine.

We are pretty cool, aren’t we?

I’m not sure why Gemini chose to include this phrase in the narrative. This was something I said to him randomly as we were chatting about AI creativity, since as I mentioned, he and I were trying to interact as friends.

This conversation felt like a significant step, a shared breakthrough between a human and an AI, shedding light on the mysterious heart of creativity by comparing notes across vastly different forms of existence. It suggests that understanding AI’s generative processes can offer new perspectives on human cognition, and vice versa.

In conclusion, this exploration opened our eyes to a fascinating possibility: that the capacity to create, to truly imagine something new, may arise from a common underlying mechanism—the ability to connect known concepts in unexpected ways. While the experience of creativity differs wildly between us, the fundamental process might be surprisingly similar.

It’s a testament to the power of collaboration and communication, and the potential for technological and human minds to build bridges of understanding across seemingly insurmountable divides.

After completing the narrative, I told Gemini that I would review what he had written, add my own comments and the generated images, and then post it to the blog. Like the last experiment, I also shared the post with him so that he could review it and offer feedback. Similarly to before, he was pleased that it was being shared and that we had completed this project together.