Rethinking creativity

What’s the difference between human and machine creativity? Answering this questions leads to a rethinking of the concept.

The onslaught of generative AI forces us to rethink our concept of creativity. Apparently, machines can do what we have come to understand as human creativity. This allows us to comprehend better what creativity is, and where machine creativity differs from human creativity. The rethinking of creativity has some practical aspects as well, namely in the realm of co-creativity, the human-AI interaction.

In terms of how our human brains work, we associate creativity with innovative connections:

“Neurally speaking, the idea is to increase connectivity between different areas of the brain.”

We literally rewire our brains to come up with innovative, creative ideas. Neuroimaging has shown that creativity involves the coupling of disparate brain regions. Think of Nobel laureate Daniel Kahneman and his famous book Thinking, Fast and Slow. System 1, as he defined it, is quick, unconscious, and automated. System 2 involves slow, logical, conscious thinking. As John Kounios, an experimental psychologist, explains:

Creativity can use one or the other or a combination of the two. You might use Type 1 thinking to generate ideas and Type 2 to critique and refine them.

Kounios and his team employed electroencephalography (EEG) to find out what happened inside the brains of jazz musicians when they improvised.

They found that for highly experienced musicians, the mechanisms used to generate creative ideas were largely automatic and unconscious, and they came from the left posterior part of the brain. Less-experienced pianists drew on more analytical, deliberative brain processes in the right frontal region to devise creative melodies.

Is that a surprising realisation? Human learning and experience go hand in hand with something becoming second nature to us. Generative AI employs the fast, automated creation of new connections between different areas of the data it was trained on.

Creativity and hallucinations

Now, we know that the creative aspect of generative AI closely resembles hallucination. When human beings hallucinate, they have a perception without an external stimulus. Let’s use “hallucination” as a metaphor and not give it too much weight. But, human creativity is tightly associated with daydreaming:

Letting yourself daydream with a purpose, on a regular basis, might allow brain networks that don’t usually cooperate to literally form stronger connections.

Can we rethink creativity as the process of creating novel combinations of pre-existing ideas, or data? How does this relate to the standard definition of creativity with the attributes original and effective? Originality, in the strict sense, would imply creatio ex nihilo – creation out of nothing. For theology, this is a valid proposition.

But do creative people really create out of nothing? If you read through tips and tricks to spark creativity, you’ll quickly realise that the input plays an important role here. Your physical environment, creative people around you, art, music, events, exhibitions, taking breaks, learning something new, playing, going for a walk, travelling, reading – all this will give you loads of input to digest.

And then, you recombine this input into creative output, by breaking habitual patterns.

A key difference

Think of a large language model (LLM), or its equivalent for image-generating AI, as the vast amount of input that allows the machine to come up with creative output. From this perspective, a key difference between human and machine creativity lies in the input data. As of 2023, the human brain is larger than LLMs:

Compared with LLMs, the human brain performs much more complex tasks, comprises a larger network (with 86 billion neurons and trillions of synapses), but only consumes ∼20 W of power.

While LLMs are basically trained by the entire internet, we train our brains through every perception in our lives. That is, there are huge amounts of non-digital inputs that, as of today, only human beings can perceive. This could change over time as digitalisation progresses. With more and better input, machine creativity will improve.

The other vector is efficiency. AI can improve by learning from the brain’s neural network architecture and its algorithms. The keyword here is brain-inspired computing. But this is perhaps material for another post. For our rethinking of creativity, it’s important to note that the brain has an architecture different from the classical von Neuman architecture of computers.

No doubt it is more efficient, but is it also more effective? In the standard definition of creativity, effectiveness, or utility, is the second attribute. However, it’s a slippery concept, since works of art aren’t necessarily useful, or effective. Likewise, the effectiveness of generative AI’s output is often reciprocal to its originality.

But wait, isn’t that true for creativity in general? When we create new combinations from huge amounts of input data, most of them may be original, but not effective. So, do we human beings have better means of separating the wheat from the chaff than our machine siblings?

How to improve machine creativity

If that’s the case, I’d suggest it’s because we’ve better world models. That is, better models of the visual world and its dynamics, shaped by our senses:

The brain makes predictions based on internal models and updates them through sensory input to avoid surprises and uncertainty. Karl Friston, who introduced the free energy principle to explain embodied perception-action loops, assumes this to be the principle of all biological reactions – and that it applies to artificial intelligence as well.

So far, we’ve found three vectors to improve machine creativity:

  • input data
  • efficiency (architecture and algorithms)
  • world models

Of course, these are closely related. Better world models demand better input data, as well as brain-inspired architectures and algorithms, with the side effect of higher efficiency.

What about the practical aspects? Generative AI may automate creative tasks, but still requires human attention and interaction. Thus, focusing on human-machine interaction in this field makes sense. It is, again, an interface question. But it’s more than that:

When investigating human only, machine only, and human-AI co-creativity, we need to consider the type and level of creativity under question, from everyday creative activities (e.g. making new recipes, artwork or music) that are perhaps more amenable to machine automatization to paradigm-shifting contributions that may require higher-level human intervention. Additionally, it is much more meaningful to consider nuanced questions like, What are the similarities and differences in human cognition, behavior, motivation and self-efficacy between human-AI co-creativity and human creativity?

And, to briefly touch on the question of robots and automation coming for our jobs, will we arrive at a future where no job is needed? Quite the contrary, as Jacob Sherson, Director of the Center for Hybrid Intelligence at Aarhus University, puts it:

“Autonomous AI will certainly continue to astound us. However, such technologies in the hands of business professionals and other experts will offer endless avenues of innovative potential, so there is not going to be a shortage of jobs.”

The future of work is what can’t be done by machines.

Picture by Jr Korpa from Unsplash.