Can Machines Be Creative?
In the past year and a half, we have seen a lot of Generative AI (Gen AI) models emerging, be it Large Language Models (LLMs) or Image Generators. Some folks have started to call these "Creative Machines," and there's a belief they may replace the jobs of creative personnel such as journalists, copywriters, and artists.
I wanted to write this piece to shed more light on creativity and the current struggle in accepting the statement, "Machines are creative." I am not here to change opinions but rather to share what I have gathered so far.
Margaret A. Boden has defined three types of creativity:
Combinational Creativity: The unfamiliar (but interesting) meshing of two or more familiar ideas.
Exploratory Creativity: Taking two or more existing structures or styles, meshing them together for novelty, and then determining their feasibility.
Transformational Creativity: Greatly manipulating the dimensions of existing styles and structures.
With these three types of creativity stated and (hopefully well) defined, let us ask the question: Which creativity can machines satisfy or meet?
We know that LLMs and Image Generation models have "read" and "seen" a vast amount of text and images, far beyond what any human can achieve in their lifetime. As such, when LLMs or ImageGen models produce something, humans may find it novel because, again, the human experience of the world is relatively small compared to the machines' "experience" of the world.
However, these novelties are, at most, brought together by what the machine has experienced, since what it "experiences" is greatly dependent on the data that is fed into it.
With this in mind, you can see that machines can certainly generate a lot of combinational creativity! You could say that machines provide a weighted average of combined familiar ideas.
Machines can possibly help with exploratory creativity, which requires the generation of prototypes for human feasibility testing. Machines can be used to quickly generate prototypes by taking advantage of their speed, allowing humans to make a quick assessment on whether to move forward with an idea.
Where machines cannot be creative is in transformational creativity. This involves branching out from an existing idea into something truly new that the world has never seen before. This kind of creativity is not achievable by machines due to the lack of appropriate training data, and it might take humans much longer to assess its possibility and feasibility.
Conclusion
Here lies the biggest difference between human creativity and machine creativity. Machine creativity can perform combinational creativity and assist with exploratory creativity, but it is very unlikely to achieve transformational creativity. Humans, on the other hand, can do transformational creativity, which is important if humanity is to advance, build novel ideas, develop better technology, and push the knowledge frontier further.
So, are Generative AI or machines in general creative? I would love to hear your thoughts!
I very much enjoyed the discussion which presented many perspectives. If you do, consider reaching out to me on LinkedIn.
Consider supporting my work! You can make a “book” donation and drop me some wisdom! :)
Recommendations
This is a good question with an easy answer: yes. It is able to take novel ideas and draw pictures about them. If humans can be considered as creative, AI can too.
Nice way in deconstructing the construct of creativity. Too often, we think of creativity in the realm of art, but ALL work requires creativity. Combinatorial and Exploratory creativity is spot on about Gen AI. But there is also the challenge of "eye of the beholder" – the receiving end of the creativity. Humans make and evaluate simultaneously, while Gen AI doesn't yet have that sense of 'evaluation'. Not sure how that space will evolve, but it would have to emerge at some point.