Accuracy of people and machine learning models in distinguishing between real and AI-generated images
Authors
Advisors
Issue Date
Type
Poster
Keywords
Citation
Abstract
Over the past decade, especially in the past few years, media generated by AI, artificial intelligence, has improved a lot. Advancements in this field have been so great that it is now fair to ask whether or not people can even tell the difference between real photos and AI-generated images. In the case that people are unable to correctly identify the majority of images, this would mean that industries like photography, film, and others would be at risk. To protect such industries, we asked whether AI could be used against itself by using a machine learning model to tell the difference between the two types of photos. To answer this question, we collected various images, some photos, some AI-generated images, and tested people against the image sets to see how many images they would correctly classify as real photos or AI-generated. After collecting these results, we identified the average percent error of participants and compared that to the average percent error of a model we built with the sole purpose of distinguishing between real photos and AI-generated images. We found that people were, more often than not, unable to correctly classify images as photos or AI-generated, with an accuracy rate, at best, of about fifty percent, while our model was able to correctly classify images nearly perfectly. This shows us that people have trouble correctly classifying images as photographs or AI-generated and that machine learning models can be used to help people distinguish between photos and images generated by AI.
Table of Contents
Description
Research project completed at the School of Criminal Justice.

