Skip to main content

Exploring the Different Forms of Creativity

Insights for Generative AI Development

In today's digital landscape, businesses must develop innovative strategies to fuel growth and outpace competitors. While generative AI is already at the forefront of this innovation, there are underexplored avenues in this area that hold massive potential for new creative technologies. Taking a step back and understanding the true nature of creativity can unlock this potential and set us on the right course for producing more efficient, higher quality generative AI tools. Some key areas to focus on are:

  • Creative products vs creative processes
  • Combinational vs transformational creativity
  • Prompt-based and prompt-less generative AI

Creative Agents: The Limits of Generative AI

Defining creativity itself has proven to be a challenging task, and it remains a lively discussion in both philosophical and psychological circles. This elusive definition doesn't mean that such debates should be brushed aside - they may have enlightening applications to generative AI development. The first perspective on this topic to consider is evaluating what makes a product (or output) creative. There is some consensus that a creative work is one that is novel, valuable, and surprising. Generative AI tools are certainly capable of creating something novel; their intended purpose is to generate something unique that fits the user's brief. These works can also be valuable - we just have to look at sales of generative AI artwork, or the industry as a whole, to realize that there is value in what these machines are capable of producing. The issue here lies with whether generative AI can produce anything surprising.

There can be an element of surprise when using generative AI. When I first used Dall-E, I was incredibly surprised at how quickly and effectively it produced what I had asked of it. I was not, however, surprised by the output. The tool had done its job and successfully created an image of the concept I had prompted it to generate. The purpose of these tools is not to surprise the user but rather to accurately interpret their input and generate something that fits the brief. The creative limitation of these current uses of AI is that they rely too heavily on the user's input to generate anything surprising. This makes them very efficient and effective tools to assist with creativity, but it does not make them creative in their own right.

Creative Processes: Can Generative AI Create Like a Human?

We can also consider creativity from the perspective of the process. The 'creative process' may be even more of an elusive definition than creativity itself, but for the sake of this discussion, there are three standard ways in which creativity can happen. The first is combinational creativity: novel combinations of familiar ideas. This is the form of creativity that generative AI is most known for assisting with. When we provide the machine with the prompt we want it to generate, we supply it with the concepts we want to combine. When we ask image-generating software to produce 'a team logo for a sales team, showing a handshake between three parties, the team colours are blue and purple' - we are asking it to combine those concepts of 'logo', 'team', 'handshake', etc. and produce an appropriate output based on the data it was trained on. This model and process may be highly efficient at prototyping, but again, it is too constrained by its prompts to produce anything revolutionary.

The other forms of creativity are exploratory and transformational. Exploratory creativity involves the exploration of structured conceptual spaces, while transformational creativity involves taking these spaces and changing them in ways that they can produce things that could not have been conceived of before. This is where true innovation and creativity can come from.

To simulate these within generative AI would mean giving it the freedom to explore and create in a much less constrained way. This may also be accompanied by making the AI more generalist, able to create in all mediums. Of course, the actual programming and production of such AI will require intensive research and development. It may require natural language understanding to be able to understand and explore the different conceptual spaces and the context to which it is being applied.

Where do we go from here?

The place to start would be researching prompt-less generative AI development. Removing the limitations of prompts and confined spaces in which to work can make generative AI a much stronger creative tool. To really support innovation and growth within creative endeavours, the AI could be involved in all stages of a discovery journey. By not limiting the technology to only respond to prompts, we could have generative AI support at all stages (ideation, refinement, analysis, prototyping, etc.) with ideas being suggested even when we did not think to ask for them.