Research

This page provides a shortened overview of my research, which includes the abstract, key findings, and statistical analysis.

My Research Contributions

I independently initiated and conducted this research under the mentorship of a University of Georgia faculty member.

My contributions include:

Research Design: Created the research questions, experiment conditions, and methodology.
Stimuli Production: Produced three versions of a mixed-media animation using fully AI-generated, hybrid, and manual production methods.
Data Collection: Designed and administered survey through QuestionPro, recruiting 155 participants through CloudResearch Connect, and obtaining 145 completed responses.
Data Preparation: Used Java in order to clean, arrange, and recode participant data through surveys into variables to be used in statistical analysis.
Statistical Analysis: Worked with my mentor to conduct chi-square tests, independent samples t-tests, and one-way ANOVAs using JASP.
Interpretation: Worked with my mentor to interpret statistical results and analyze patterns in classification accuracy and artistic evaluations.

Abstract

As artificial intelligence (AI) becomes increasingly prevalent in digital media production, distinguishing AI-generated work from human created work has become increasingly difficult. This study investigates viewers’ accuracy in distinguishing production methods of Mixed Media Animation, an animation method that involves a combination of traditional and digital techniques. Additionally, it examines how different authorship methods impact perceptions of artistic quality, creativity, visual cohesiveness, effort, and skill.

Participants (N=145) were assigned to one of three production methods: entirely AI-generated (n=47), entirely human-created (n=47), or a hybrid combining AI-generated and human-created elements (n=51). After viewing the animation, participants evaluated artistic qualities using a seven-point Likert Scale, identified the production method, and reported their confidence in their classification. Additional questions assessed participants’ editing experience, familiarity with Mixed Media animation and generative AI, AI usage frequency, and attitudes towards AI.

Preliminary Findings

Sample Size: 145 Participants (AI: n=47; Hybrid: n=51; Human: n=47)

1. Classification Accuracy

Overall classification accuracy was relatively low, with only 29.0% (42/145) of all participants successfully classifying the production method of the animation they viewed.

Classification accuracy varied considerably across production condition:

Fully AI Generated: 6.38% (3/47)
Hybrid: 62.75% 32/51
Fully Human-created: 14.89% (7/47)

Participants most frequently classified the animation as hybrid, regardless of the production condition. A chi-square test of independence found no statistically significant association between actual and believed production conditions: χ²(6, N=145) = 6.015, p = .421.

2. Actual Production Condition and Artistic Assessments

A series of one-way ANOVA tests were performed to examine whether the actual production condition would have an effect on the participants' evaluations of originality, artistic quality, visual cohesiveness, effort, and skill.

AI-generated animation scored the highest ratings for originality (M = 5.55), artistic value (M = 5.43), perceived effort (M = 5.13), and perceived skill (M = 4.92). None of the differences between the production conditions was significant (all p > .05).

Artistic quality showed the strongest evidence of a potential difference between conditions (F(2, 142) = 2.960, p = .055).

3. Believed Production Condition and Artistic Assessments

A series of one-way ANOVA tests were performed to examine whether the believed production condition would have an effect on the participants' evaluations of originality, artistic quality, visual cohesiveness, effort, and skill.

Participants' evaluation of skill differed significantly across believed production condition, F(3, 141) = 2.970, p = .034. Those participants that were uncertain of the production method had the highest average score of skill (M = 5.55), while those perceiving the animation as an AI creation had the lowest (M = 3.55).

Effort approached, but did not reach, statistical significance (p = .062), whereas no significant differences were found among originality, artistic quality, or visual cohesiveness.

These results imply that there is a possible relationship between perceived authorship and evaluations of artistic characteristics.

Page updated

Google Sites

Report abuse