Can humans recognize AI images

[Rabiahfs]

My 7th grader wants to do this for his school science fair. He wants to do an adult vs student comparision as well. should he tabulate his results for each image ( as shown in your procedure, e.g. image #1: 50% volunteers recognized it correctly), or should he do it for each participant (e.g. participant 1 recognized 11/20 images, participant 2 recognized 12/20 correctly and so on).


Moderator note: https://www.sciencebuddies.org/science- ... ted-images

[DauntlessDog12]

Hi there,

If you care about the individual responses and using the table to generate the percentage that you listed, you may want to stick to the table but use two: one for adults and one for students. It will also depend on the clarity that you want from the data. If you have a lot of participants, it might be useful to consolidate the table so that you won't have to format a lot of data into one table. Personally, I would go with the second option with individual participants, but it will depend on how you want the tables to look in the end. Hope this helps.

[Rabiahfs]

Thanks.
any suggestions on where we can get the real and AI images to use for his experiment?

[Rabiahfs]

is this hypothesis ok??or is it too long??
If adults and students are shown a mixture of real photographs and AI-generated images, then both groups will correctly identify the image type only about 50–60% of the time, which is close to random guessing; however, students are expected to perform slightly better than adults because they are often more familiar with AI-generated media and online imagery.

[CarissaP]

Hello,
For finding datasets of real and AI images, I would suggest using Kaggle (https://www.kaggle.com/). Kaggle contains thousands of datasets that are often used for training machine learning models, but these datasets could definitely be used in your experiment. I think your hypothesis sounds pretty good! It both makes a general claim and compares adults and students.
Have fun with this project!
Carissa