The release of AI tools, like MidJourney, that allow people to create graphics in seconds has prompted widespread concern about the impact on jobs in the creative sector. Indeed, in the UK, creatives are fighting back against proposals that would weaken copyright law and allow tech companies to freely use their work to train algorithms.
There’s an element of justification to these fears, with research from Imperial showing that there was a 30% drop in the number of graphic design jobs posted on a freelance job platform since the introduction of generative AI tools that promise to do this work for us. This was reflected by those in the industry, who argued that we’re largely sleepwalking into a pretty dire future for many creative professionals.
Humans and AI
For many years in the pre-GenAI era, the narrative was that AI would augment humans rather than replace them. Indeed, IBM used the tagline of “augmented intelligence” to try and popularize what has largely been referred to as Kasparov’s Law, which states that a human and AI will always be better than either a human or an AI on their own.
Despite the pessimistic jobs figures, it’s a narrative that creatives themselves seem to have taken on board. Research from Duke University shows that professional artists are bullish about their prospects, and firmly believe that they are best placed to utilize the new technology.
The emergence of generative AI has prompted much discussion around its impact on creativity. For instance, a study from the University of South Australia looks at the relationship between AI and creativity, found that while AI can produce creative-looking work, it still depends on human input.
A broader study, from Cambridge’s Judge Business School, looked at a total of 13 creative tasks, ranging from story writing to scientific problem-solving. The tasks were designed to cover creative writing, problem solving, and divergent thinking, which are believed to be at the heart of creativity.
The results show that the collective creativity of LLMs is broadly equivalent to 8-10 humans when the LLM was questioned 10 times. For every 2 additional responses after that, it was equivalent to around one extra human. The researchers argue that, if we use LLMs in the right way, they may be competitive with a small group of humans on creative tasks.
Augmenting humans
The Cambridge researchers argue, however, that humans will still have the edge providing they learn to work effectively with the technology. It’s a perspective shared by the Duke study, which found that professional artists have largely retained the upper hand, even in our AI age.
The researchers quizzed 15 professional artists, all with several years of experience, and paired them with 15 lay people with next to no artistic training. All of them were asked to write prompts that were fed into DALL-E 3 to generate a series of pictures. The researchers also put the same instructions into ChatGPT in order to get some AI-generated prompts to put into DALL-E 3.
Then, they recruited around 300 volunteers to assess all 45 of the generated artworks, with a rating provided for their creativity. The results show that the work created by the professional artists was uniformly rated as the most creative, with the AI-generated artwork in second, and the work produced by the novices last.
More creative
When the researchers explored why this was, they found that both the professional artists and ChatGPT would often use a wider range of words in their prompts, which then helped DALL-E produce more creative work.
The professional artists, however, would use words with a much higher level of so-called semantic distance than ChatGPT did. This describes how the words used were not ones that we would commonly use together. It’s a phenomenon that has been linked with higher levels of creativity in previous studies.
“Whether AI is truly ‘creative’ depends on how we define creativity,” the researchers explain. “AI undeniably generates images and other outputs that people perceive as creative. But if creativity is tied to human experience, emotions and intentionality, then AI appears to fall short.”
What’s more, the researchers believe that the results could be even starker if the participants gained more experience with DALL-E rather than approach the task as complete novices. Similarly, none of them were allowed to edit their prompts to fine tune the images, which might also further enhance the creativity of the outputs.
Suffice it to say, the researchers accept that we’re in a rapidly shifting environment, and while the professional artists did very well in this study, if the capabilities of AI continue to advance, this may not be the case in the future.
“If AI becomes better at understanding artistic intent or mimicking human-like intuition, the gap between AI and professional artists could narrow or even disappear,” they conclude. “However, it’s possible that the uniquely human aspects of creativity – such as emotional depth and lived experience – will continue to set human artists apart. Whether that distinction matters in the long run remains an open question. Only time will tell.”
