A study conducted by Kent Hubert, Kim Ava, and Daria Zabelina demonstrated a more creative approach to artificial intelligence in solving creative problems than previously thought. During testing, the GPT-4 neural network from OpenAI provided more original answers compared to humans.
AI Outperforms Humans
The study, titled “Current AI Generative Language Models Are More Creative in Lateral Thinking Tasks,” utilized three tests. In the first test, participants were tasked with brainstorming unusual uses for household items such as cutlery or rope. The second test required presenting solutions to hypothetical situations, such as if people no longer needed sleep.
The third test, dubbed “Task for Remote Associations,” challenged participants to generate ten nouns with the greatest semantic distance from each other. For example, while “wolf” and “bear” have a small semantic distance, “wolf” and “bronchitis” have a significantly larger one. The study’s results indicated that GPT-4 provided more original responses than humans across all types of tests.
However, the study did not consider factors such as individuals’ engagement in creativity or potentially lower motivation among humans to generate responses. Additionally, artificial intelligence lacks independence and relies on human assistance. Consequently, the potential of AI may stagnate without proper stimulation through tasks provided by humans.