The stunning capabilities of artificial intelligence (AI) large language models (LLMs) challenge the long-held belief that creativity differentiates humans from machine learning algorithms. Has AI technology exceeded humans in the creative realm? A new study compares the abilities of AI versus humans in creative divergent thinking with potential insights on the future of work in creative domains.
The Future of Jobs Report 2023, by the World Economic Forum (WEF), states the most important skills for workers in 2023 are the cognitive skills of analytical and creative thinking. According to the WEF report, creative thinking is growing more in importance compared to analytical thinking.
Increasingly, AI technology is being used for creative purposes. According to a 2023 Statista survey of 4,500 American professionals, 37 percent of those surveyed who are working in advertisement or marketing had used AI to assist with work tasks.
“With AI systems becoming increasingly capable of performing tasks that were once solely within the purview of humans, concerns have been raised about the potential displacement of jobs and its implications for future employment prospects,” wrote the study co-authors Simone Grassini and Mika Koivisto, PhD.
Grassini is an Associate Professor at the department of Psychosocial Science of the University of Bergen, Norway, and at the Cognitive and Behavioral Neuroscience Laboratory at the University of Stavanger, Norway. Koivisto is a University Lecturer in Psychology at the University of Turku in Finland.
“The development and widespread availability of generative artificial intelligence (AI) tools, such as ChatGPT (https://openai.com/) or MidJourney (https://www.midjourney.com), has sparked a lively debate about numerous aspects of their integration into society, as well as about the nature of creativity in humans and AI,” the authors wrote.
Large language models are AI deep learning algorithms that are trained using unsupervised learning with massively large data sets, often scraped from the Internet, in order to “understand” existing content and generate new content. Examples of large language models include OpenAI Codex, and the OpenAI LLMs for its AI chatbot ChatGPT (GPT-4 and GPT-3), GPT-4 for Microsoft’s AI chatbot Bing Chat, BLOOM by HuggingFace, the Megatron-Turing Natural Language Generation 530B by NVIDIA and Microsoft, Anthropic’s Claude (for AI chatbot Claude 2), Meta’s LLaMA, Salesforce Einstein GPT (Using OpenAI LLM), the PaLM 2 that powers Bard, Google’s AI chatbot, and Amazon’s Titan.
To measure the creativity of humans versus AI, the researchers used the Alternative Uses Test (AUT), a test designed by American psychologist J.P. Guilford, one of the eminent psychologists of the 20th century according to the American Psychological Association (APA). The AI chatbots evaluated include ChatGPT3 (version 3.5), ChatGPT4, and Copy.Ai, which is based on GPT3 technology.
Guilford views intelligence as an aggregate of many mental factors or abilities, rather than one dominating general ability. Guilford’s theory of human intelligence consists of the three dimensions of operations (cognition, memory, divergent production, convergent production, evaluation), products (units, classes, relations, systems, transformations, and implications), and contents (visual, auditory, symbolic, semantic, behavioral).
Guilford considered creativity as a form of problem-solving and a part of intelligence. Problem-solving abilities could be further defined as sensitivity to problems, fluency (ideational, associational, and expressional), and flexibility (spontaneous and adaptive).
Guilford is credited for introducing the terms “divergent and convergent thinking” in the 1956 theory of human intelligence called the Structure of Intellect Model (SI). Brainstorming is an example of divergent thinking, where many ideas are generated in response to an open-ended task or question. In contrast, the output of convergent thinking is a single correct answer to a well-defined problem.
In this study, the tasks included generating creative and original uses of everyday objects, such as a rope, box, pencil, and candle. The researchers found that, unlike the response generated by the AI chatbots, the 256 human study participants had generated a relatively high proportion of what could be considered sub-par ideas, or common responses.
“The results suggest that AI has reached at least the same level, or even surpassed, the average human’s ability to generate ideas in the most typical test of creative thinking (AUT),” the researchers concluded.
However, the AI chatbots lacked consistency and the top human performers achieved better results than AI, the study results showed. The research has provided a snapshot of AI’s creativity versus humans. Grassini and Koivisto caution that this may change six months from now as AI technology continues to rapidly advance in the future.
Copyright © 2023 Cami Rosso All rights reserved.