Artificial Intelligence
Creativity at the Expense of Novelty – Are Large Language Models (LLMs) Engagement Equalizers?
Securities.io is not an investment adviser, and this does not constitute investment advice, financial advice, or trading advice. Securities.io does not recommend that any security should be bought, sold, or held by you. Conduct your own due diligence and consult a financial adviser before making any investment decisions.
Artificial Intelligence is one of the biggest trends of this decade. Its ability to automate repetitive tasks, reduce human error, make unbiased and quick decisions, and be available 24/7 makes AI extremely beneficial across sectors.
Given these benefits, it is projected that by 2025, AI will eliminate 85 million jobs while creating 97 million new ones.
As AI permeates our daily lives, a growing worry has been whether AI, one day, will replace humans. Well, when you hear and read about AI passing bar exams and achieving a better score than human experts on benchmarks like Massive Multitask Language Understanding, it makes sense that people are wondering about artificial intelligence replacing human intelligence and leaving us obsolete.
While AI usage has risen significantly, with ChatGPT alone having 180.5 million registered users and 100 million users actively using it, this may be too soon to be excited or fearful. This may especially be true when it comes to creativity.
Creativity is a fundamental feature of being human, but it has actually been a challenge for AI. However, with the help of generative AI, stories can be made more creative and enjoyable.
Besides enhancing the content by providing potential starting points and helping overcome writer's block, generative AI can also restrict the variability of a writer's own ideas from the start.
So, while it has been seen that humans can increase quality and improve productivity with the help of AI, little is known about the impact of technology on fundamental human behavior—the ability of humans to be creative.
To find the answer to this question, the latest study focused specifically on the role of AI in affecting creative output through short fiction.
In its study, to understand how the tech affects the participants' ability to produce this particular type of creative written output, the researchers didn't introduce financial incentives but provided guidance to authors to write a story on a random topic with instructions on its length and the target audience.
Creativity is core to innovation and human expression and is usually evaluated based on two main factors: novelty and usefulness. So, for assessment, the study used the novelty index, which captured the story's originality, and the usefulness index, which captured the story's appropriateness for its audience, the feasibility of being developed into a complete book, and the likelihood of a publisher developing the book.
The two-phase study on written creative output was conducted online. In the first phase of the pre-registered experiment, 293 participants were asked to write a short, eight-sentence story for a teenage and young adult audience.
These participants for the experiment were chosen from the Prolific platform and filtered for being based in the UK with an approval rating of at least 95% from between 100 and 1,000,000 prior submissions. What's notable here is that writers weren't chosen based on their writing skills or creativity.
Participants were then randomly segregated into three conditions — human-only with no mention of or access to generative AI, human with one GenAI idea condition with an option to call upon OpenAI's GPT-4 LLM to get a three-sentence starting idea, and human with five GenAI ideas can get up to five different ideas.
In the 2nd phase of the study, a separate group of 600 participants were assigned to evaluate the stories. Much like the writing participants, the evaluator participants were picked off of Prolific, reside in the UK, have an approval rating of at least 95%, and had not previously participated in the writer study. The evaluators represented “regular” readers and were not selected based on prior experience in the publishing industry.
Each evaluator was given six randomly selected stories without being informed which condition they belonged to.
Professionalizing the Story with Increased Novelty and Usefulness
The new study, published in the journal Science Advances, was conducted by Professor Oliver Hauser from the University of Exeter Business School and Professor Anil Doshi from the UCL School of Management.
The study found that having access to ideas generated by AI resulted in stories being evaluated as more enjoyable, creative, and better written.
So, AI technology has been found to improve a writer's creativity by enhancing the rarity and usefulness of stories' ideas. The study found that AI “professionalizes” stories, which tend to have plot twists and are less boring.
To better understand just how the greater availability of generative AI ideas affects the enhancement of creativity, the study looked into the two generative AI conditions separately. Across the two generative AI conditions, the vast majority (at 88.4%) of participants used AI to get at least one initial story idea.
Out of the 100 writers who got one idea from generative AI, 82 of them chose to generate one idea using AI. Meanwhile, a whopping 94.89% (i.e., 93 writers) out of 98 participants in the Human with Five GenAI ideas category chose to do so.
Additionally, participants called upon the AI for ideas more than once on an average of 2.55 times, and 24.5% of participants requested a maximum of five ideas from generative AI. Interestingly, the study didn't “observe any differences in how generative AI was accessed based on the inherent creativity of the writer.”
The study found that having access to one idea from the AI leads to “somewhat greater creativity,“ but the most benefit came from having access to five generative AI ideas.
So, in terms of novelty, writers with just one AI idea experienced an increase of 5.4% over writers with no AI access. Meanwhile, writers with five GenAI ideas exhibited an increase of 8.1% in novelty over those without any AI access.
The most notable difference, however, came in the usefulness of the story. So, those with access to only one AI idea saw it to be 3.7% higher than those with no access. As for those with access to five AI ideas, the usefulness of their stories rose by 9% over those with no access at all and 5.1% with only one AI idea.
According to the study results, “having access to more AI ideas leads to more creative storytelling.”
The study also looked into evaluators' emotional responses to the stories. This was based on reader interest, which included not just how well the story was written but also whether it was funny and enjoyable, the degree of plot twists, and whether the content altered the reader's expectations about future stories.
What the study found was that AI helped make stories more enjoyable and the more the AI ideas, the more likely the plot twists.
Overall, relative to human-only stories, stories written by those having access to up to five AI ideas were considered to be better written, less boring, and have more of an effect on the evaluator's expectations of future stories. However, where they lack is in the fun element, with stories from five AI ideas not being evaluated as funnier than those with no AI access.
Bridging the Gap between Less Creative Writers and Those with High Creativity
With AI on the rise, it's crucial that we understand just how it helps humans, and the latest study does just that.
Conducted by researchers from the University of Exeter Business School, Institute for Data Science and Artificial Intelligence, and the UCL School of Management, the study found that using AI can actually put less creative people on the same level as those who are highly creative. However, those who are already very creative don't really have much use of this technology.
Among the most inherently creative writers, those with high-divergent association task (DAT) writers, there has been little effect of having access to AI ideas on their story's creativity. DAT is a trait measure of creativity, and each participant was asked to complete it.
Across all conditions, the study found that high-DAT writers' stories were evaluated relatively highly for novelty and usefulness, as well as for how well the story was written and how enjoyable it was. Having access to generative AI does not affect high-DAT writers' already good performance.
Those who are inherently less creative, however, were able to improve their creativity and particular emotional characteristics with the help of AI. Among low-DAT writers, those with access to just one AI had their story's novelty improved by 6.3% in comparison to a 10.7% increase for those with five AI ideas. As for usefulness, the jump was 5.5% and 11.5% for one and five generative AI ideas, respectively.
Furthermore, for low-DAT writers with one AI idea, there was an increase of up to 26.6% in how well the story was written, a 22.6% jump in enjoyment, and a 15.2 decrease in the story's boringness. These improvements put low-DAT writers' stories on par with high-DAT writers.
The study further evaluated the objective aspects of the story, which is just how similar or different the stories are. It was found that AI is not as powerful in providing distinctiveness. The generative AI-enabled stories have been more similar to each other than stories by humans alone. So, “writers in the two generative AI conditions are anchored to some extent on the generative AI idea presented to them.”
This points to the new and advanced technology's ability to help increase individual creativity, but this comes at the risk of losing collective novelty. The study stated:
“This dynamic resembles a social dilemma: With generative AI, writers are individually better off, but collectively, a narrower scope of novel content is produced.”
The study further made additional observations, including AI helps less able writers, hence effectively equalizing the stories' assessment and eliminating any advantage or disadvantage based on the writers' inherent creativity. Also, there's no evidence of AI's ability to push the upper bound of creativity beyond what creative humans are capable of on their own.
After disclosing to the evaluators whether the writer got AI ideas and the ideas themselves, the study collected additional outcomes that it says can inspire future directions of research. These outcomes included evaluators imposing an ownership penalty of at least 25% on writers who got AI ideas.
Most evaluators also indicated that content creators, on whose work the models are based, should be compensated and that the use of AI should be disclosed in publications that employ it. Nevertheless, the majority of evaluators considered the use of AI in story writing to be ethical and a ‘creative act.'
Talking about the study's limitations, the authors pointed to length, medium, and output type in addition to writers having no interaction with the AI model or prompt variations.
So, Can AI Really Replace Humans?
With this study, the researchers have made an attempt to understand the impact of AI on something that has both economic and purely expressive value.
Ever since the popularity and adoption of generative AI solutions like Chatgpt, there has been a growing chatter about AI replacing human labor in the not-so-distant future. As we have been seeing, generative AI can create new content in text, images, audio, and video in a fraction of the time it takes humans to produce the same content.
However, as this study has found, the ‘horse race' between human and AI-generated ideas may not be decidedly in favor of AI. Those with less innate talent will surely benefit from it, but for those who already possess high creativity, AI does not appear to be of much use. That said, AI might still be useful to them in other ways.
As the paper noted, the study used a sample of “typical“ participants used in academic studies and did not consider unusually creative individuals or professional writers, an important population segment that remains understudied. The effects of artificial intelligence on this segment can be transformative in other ways, “potentially offering efficiency gains or improved speed of execution.“ Still, less creative individuals are expected to see the largest impact and get the most out of generative AI.
Moreover, the increase in individual creativity comes at the risk of losing collective novelty, which raises the question of whether the stories inspired and enhanced by AI will be able to have sufficient variation in their outputs.
If publications embrace this content, the study says, the produced stories would become more similar and less unique. This “downward spiral“ will mean individuals will be incentivized to increasingly use AI in the future, which will further reduce the collective novelty of stories. It noted:
“Our results suggest that despite the enhancement effect that generative AI had on individual creativity, there may be a cautionary note if generative AI were adopted more widely for creative tasks.“
As a rapidly evolving technology, generative AI‘s full potential is far from being explored. One day, AI will be able to think, learn, perceive, reason, and solve problems like humans.
While that future may one day be a possibility, it might not be coming for decades yet. Still, with technology already enhancing human creativity, further advancements may see future creative possibilities extend far beyond our imagination.
Click here to learn all about investing in artificial intelligence.