Is AI now more creative than humans?
Is AI now more creative than humans?
Putting ChatGPT through a divergent thinking test...
Paid supporters can hear a narrated version of this article by clicking here.
Written by Davide Piffer.
It is difficult to say definitively whether artificial intelligence (AI) is more creative than humans, as creativity is a complex and multifaceted concept that cannot be easily compared.
On one hand, AI has the potential to be more creative than humans in certain ways. For example, AI can process and analyze vast amounts of data quickly and accurately, which can allow it to generate new ideas and solutions by synthesizing and analyzing information in a way that is beyond the capabilities of most humans. AI can also learn and adapt to new situations through the use of machine learning algorithms, which can allow it to come up with creative solutions to problems that may not have been anticipated by its human creators.
However, there are also limitations to the creativity of AI. One of the main limitations is that AI is only as creative as the data and algorithms that are fed into it. In other words, AI is only capable of generating ideas and solutions that are based on the information that it has been trained on, and it is not able to come up with truly novel ideas or solutions that are completely unrelated to its training data.
Another limitation of AI is that it lacks the ability to understand and interpret the world in the same way that humans do. While AI can process and analyze data, it does not have the ability to experience emotions, perceive the world through our senses, or understand the context and meaning behind the data that it is processing. This means that AI is not able to generate ideas or solutions that are inspired by emotions, personal experiences, or cultural context in the same way that humans can.
In conclusion, while AI has the potential to be more creative than humans in certain ways, it is limited by the data and algorithms that it is trained on and lacks the ability to understand and interpret the world in the same way that humans do. As such, it is difficult to say whether AI is more creative than humans overall.
Why the italics? Because this is the answer ChatGPT gives to the question: “Is artificial intelligence more creative than humans?”.
If you’ve been living under a rock, ChatGPT is a form of artificial intelligence (AI). It is a language model that has been trained on a large dataset of human-generated text and can generate human-like responses to prompts. Language models like ChatGPT are commonly used in natural language processing (NLP) tasks, such as dialogue generation and chatbot systems.
However, the answers provided by ChatGPT on a test designed to measure creative potential in humans suggest that the chatbot may overestimate its limitations.
Psychologists often use divergent thinking tests, such as the Alternate Uses Test (AUT), to assess creativity. The AUT, developed by J.P. Guilford in 1967, asks individuals to come up with as many uses as possible for a simple object like a brick or shoe. Results from this test are typically evaluated based on four dimensions:
Fluency, or the number of alternative uses generated
Originality, or the uniqueness of the ideas
Flexibility, or the range of ideas in different domains and categories
Elaboration, or the level of detail and development in the ideas.
Research has shown that the ability to generate a high number of relevant ideas or answers (known as fluency) is strongly related to the other three dimensions of divergent thinking (Takeuchi et al., 2020). Because it is simpler and less problematic to measure fluency by counting the number of answers given rather than evaluating the originality of those answers, we will focus on the number of responses provided. The reader can then determine the originality of the answers for themselves. From personal experience administering this type of test to humans, I have found that the perceived originality of the answers given by ChatGPT is not lower than that of human responses. I prompted ChatGPT with this item from a divergent thinking test:
Provide as many answers as you can to this divergent thinking test:
"Other than drinking milk, how can we use milk bottles?”
Before continuing, try to answer yourself and set a 5-minute time limit. After completing the test, count the number of answers you have provided. I’ll list ChatGPT’s answers at the end.
I noticed that the AI stopped at 10 answers because this is the pre-set number of responses. I then asked it to provide more answers. Of course, the AI could have continued indefinitely and would have outperformed any human in terms of the number of ideas generated.
According to research by Benedek et al. (2013), the average university student generates about 13 ideas (standard deviation = ~5) in 5 minutes. In contrast, the AI provided 10 answers in 1 minute, 20 in 1:30 minutes, and 30 in 2 minutes. At this rate, it could reach a score of 90 in 5 minutes, which is about 15 standard deviations above the average for university students.
It is possible that the AI is simply too modest to admit that it has surpassed humans in terms of creative potential. One explanation for this paradox is that the divergent thinking tests used to measure creativity may not actually be reliable indicators of creativity.
I’ve previously written about how creativity should result in a realistic product (2012). Obviously, people can quibble about definitions, but most recognize that nonsense answers aren’t the hallmark of a truly creative person. If I said I could use a milk bottle as a skyscraper, for example, that’s undoubtedly a creative use, but is it actionable? The most creative people aren’t the ones daydreaming random thoughts, but the people who take action. Another person might have the same thought, recognize it is ludicrous but use it as ludic inspiration for an actual building or blueprint. Hence, there’s a distinction in the literature between creative potential and creative output.
ChatGPT outperforms us because it’s a language model based on statistical associations between words. That is, it learns patterns and relationships between words and sequences of words, and uses these patterns to generate new text or understand and respond to input.
In the same way, divergent thinking is based on associative processes. People with less structured semantic networks have looser association hierarchies, and can make remote associations more easily (Mednick, 1962; Kennet, Anaki and Faust, 2014, Forthmann et al., 2016). In humans, both bottom-up processes such as associations and top-down abilities such as memory retrieval and fluid intelligence are involved in divergent thinking tasks (Beaty et al., 2014).
ChatGPT can obviously draw from a huge amount of knowledge, dwarfing the memory capacity of any human being. Hence, not only can it make associations more rapidly, but it can make associations using a much larger pool of concepts. So, ChatGPT clearly possesses one aspect of creativity (divergent thinking) to an extremely high degree. But does this result in real creative behavior?
The answer is yes…sometimes. DALL·E 2 is an AI system that generates realistic images and art based on verbal descriptions provided by users. It is currently difficult to evaluate the artistic value of its creations, but it is clear that the output of this system is significantly greater than that of any human artist in terms of both the quantity of its production and the level of technical skill. It is also likely that the originality of DALL·E 2's creations surpasses that of the average non-artist human.
The widespread, long-held belief that AI is only capable of excelling at logical and cognitive tasks, and is incapable of emulating human creativity, has been challenged. Indeed, it may even be easier for AI to surpass humans in terms of creative thinking — at least in some domains — than in raw IQ due to its ability to utilize its associative network and extensive knowledge base for generating creative ideas. Its ability to identify relationships and connections allows AI to come up with novel ideas and solutions that humans may not have thought of, or to produce paintings by combining concepts or visual features that are not obviously related. A significant open question is whether a lack of emotions will prevent AI from true artistic mastery or whether this is yet another thing that can be mimicked with enough data, processing power, and algorithmic ingenuity.
References
Beaty RE, Silvia PJ, Nusbaum EC, Jauk E, Benedek M. The roles of associative and executive processes in creative cognition. Mem Cognit. 2014 Oct;42(7):1186-97. doi: 10.3758/s13421-014-0428-8. PMID: 24898118.
Benedek M, Mühlmann C, Jauk E, Neubauer AC. Assessment of Divergent Thinking by means of the Subjective Top-Scoring Method: Effects of the Number of Top-Ideas and Time-on-Task on Reliability and Validity. Psychol Aesthet Creat Arts. 2013 Nov 1;7(4):341-349. doi: 10.1037/a0033644. PMID: 24790683; PMCID: PMC4001084.
Boris Forthmann, Anne Gerwig, Heinz Holling, Pınar Çelik, Martin Storme, Todd Lubart. The be-creative effect in divergent thinking: The interplay of instruction and object frequency. Intelligence,Volume 57,2016, 25-32, https://doi.org/10.1016/j.intell.2016.03.005.
Y.N. Kenett, D. Anaki, M. Faust (2014). Investigating the structure of semantic networks in low and high creative persons. Frontiers in Human Neuroscience, 8 , 1-16, 10.3389/fnhum.2014.00407
S.A. Mednick (1962). The associative basis of the creative process. Psychological Review, 69, 220-232, 10.1037/h0048850
Piffer, D. (2012). Can creativity be measured? An attempt to clarify the notion of creativity and general directions for future research. Thinking Skills and Creativity, 7(3), 258–264. https://doi.org/10.1016/j.tsc.2012.04.009
Hikaru Takeuchi, Yasuyuki Taki, Rui Nouchi, Ryoichi Yokoyama, Yuka Kotozaki, Seishu Nakagawa, Atsushi Sekiguchi, Kunio Iizuka, Sugiko Hanawa, Tsuyoshi Araki, Carlos Makoto Miyauchi, Kohei Sakaki, Yuko Sassa, Takayuki Nozawa, Shigeyuki Ikeda, Susumu Yokota, Daniele Magistro, Ryuta Kawashima (2020). Originality of divergent thinking is associated with working memory–related brain activity: Evidence from a large sample study, NeuroImage,Volume 216,2020,116825,1053-8119, https://doi.org/10.1016/j.neuroimage.2020.116825.
Appendix:
ChatGPT answers
First prompt:
Milk bottles can be used as a container for storing and transporting liquids, such as water, juice, or homemade smoothies.
Milk bottles can be used for making homemade yogurt or fermented dairy products, such as kefir or crème fraîche.
Milk bottles can be used for growing herbs or small plants, by filling the bottle with soil and seeds, and placing it in a sunny spot.
Milk bottles can be used for making homemade cleaning solutions, by mixing water, vinegar, and essential oils in the bottle.
Milk bottles can be used as a makeshift funnel, by cutting off the top of the bottle and using it to pour liquids or dry ingredients into small containers or bottles.
Milk bottles can be used as a DIY bird feeder, by filling the bottle with birdseed and hanging it up in a tree or on a balcony.
Milk bottles can be used for storing small objects, such as buttons, beads, or craft supplies.
Milk bottles can be used for making homemade soap or shampoo, by mixing together various oils, fragrances, and other ingredients in the bottle.
Milk bottles can be used as a makeshift watering can, by filling the bottle with water and using the spout to water plants.
Milk bottles can be used as a DIY lamp, by cutting a hole in the bottom of the bottle, inserting a light bulb, and wiring it to a power source.
Second prompt:
Milk bottles can be used as a portable container for drinks, such as coffee, tea, or hot chocolate, by attaching a lid and carrying it with you on the go.
Milk bottles can be used as a makeshift ice pack, by filling the bottle with water and freezing it.
Milk bottles can be used as a DIY sprinkler, by cutting small holes in the sides of the bottle and attaching a hose to the top.
Milk bottles can be used for storing and organizing small items, such as coins, paper clips, or rubber bands.
Milk bottles can be used for making homemade salad dressing or marinades, by mixing together oil, vinegar, and spices in the bottle.
Milk bottles can be used for making homemade insect repellent, by mixing together essential oils and water in the bottle and spraying it on your skin or clothing.
Milk bottles can be used as a DIY snow globe, by filling the bottle with water, adding small figurines or other decorations, and sealing the top.
Milk bottles can be used as a makeshift telescope, by cutting a small hole in the bottom of the bottle and looking through it to magnify objects in the distance.
Milk bottles can be used as a DIY aquarium, by filling the bottle with water and adding small fish or other aquatic creatures.
Milk bottles can be used as a makeshift drum, by filling the bottle with beans, rice, or other small objects and tapping it with a spoon or other object to create different sounds.
Third prompt:
Milk bottles can be used as a DIY birdhouse, by cutting a hole in the side of the bottle and attaching it to a tree or other structure.
Milk bottles can be used as a makeshift terrarium, by filling the bottle with soil, plants, and other decorations to create a miniature ecosystem.
Milk bottles can be used for storing and organizing art supplies, such as markers, crayons, or paints.
Milk bottles can be used as a DIY candle holder, by filling the bottle with melted wax and inserting a wick.
Milk bottles can be used as a makeshift maraca, by filling the bottle with beans, rice, or other small objects and shaking it to create different sounds.
Milk bottles can be used as a DIY pinhole camera, by covering the top of the bottle with aluminum foil and making a small hole in the center.
Milk bottles can be used for storing and organizing small tools, such as screws, nails, or bolts.
Milk bottles can be used as a DIY sand art project, by filling the bottle with different colors of sand and creating designs or patterns.
Milk bottles can be used as a makeshift drumstick, by attaching a piece of foam or other soft material to the end of the bottle and using it to hit a drum or other percussion instrument.
Milk bottles can be used as a DIY sensory bottle, by filling the bottle with water, food coloring, and small objects, and sealing the top to create a visually appealing and tactile sensory experience.