Human-AI Creativity

PI: Mr Poon King Wang (LKYCIC, SUTD)
Co-PI: Dr Dinithi Nilanga Jayasekara (LKYCIC, SUTD), Dr Benjamin Prisse (LKYCIC, SUTD
Team: Dr Deng Ruotong (LKYCIC, SUTD), Mr Ho Jun Quan (LKYCIC, SUTD)

We evaluate the creativity of Artificial Intelligence (AI) by asking individuals to assess the creativity of human artwork and their AI-generated equivalents. Participants assign a grade from 0 to 10 to pencil drawings of animals, watercolor paintings of mountain landscapes, and commercial logos. We then compare the creativity ratings of human and AI artwork. Results indicate that AI is judged to be more creative in all experimental conditions.

The role of AI in society remains an ongoing debate. The introduction of Chat GPT into society has resulted in job losses for workers in writing-related tasks, creating a context of generalized fear about job displacement by AI. Currently, there is a discussion about the extent to which programming and other high-paid jobs that require intellectual performance can be automated. One perspective is that AI will not replace humans, as it executes commands conceived by humans. Instead, AI serves as an assistant that can automate simple and laborious tasks, thereby enabling humans to focus on the more challenging and reflective aspects of their work. The goal of this study is to provide evidence on whether AI can match humans in a task that traditionally requires a component thought to be unique to humans: creativity.

The project investigates whether artistic work made by Artificial Intelligence (AI) is judged as creative by humans. It is accomplished by comparing drawings created by human artists and their AI equivalents generated using the state-of-the-art Generative AI, Midjourney.

We hypothesize that the ability of a state-of-the-art AI like Midjourney to perform similarly to humans decreases with the technical complexity of the task increases.

Participants will be asked to complete three tasks. In the first task, they will grade the creativity of drawings on a scale from 0 to 10. In the second task, they will bid tokens for each of the same drawings, with bids ranging from 0 to 10. In the third task, they will grade the creativity of a new set of drawings on a scale from 0 to 10.

By analyzing the results, we will determine whether participants attribute higher grades to human or AI drawings, both when subjects are giving an opinion or taking an action incurring a personal cost.

To assess different levels of difficulty, we will use three distinct experimental treatments: Business logos, simulating a task of simple difficulty; Animals pencil drawings, simulating a task of intermediate difficulty; and watercolor paintings, simulating a task of advanced difficulty.

Project email: trustinai@sutd.edu.sg