Can humans always recognize AI generated images

We hypothesize that older kids can detect AI better then younger kids. We hypothesize this because older kids have probably been exposed to more AI and know more about AI then younger kids.

Our background research found there are not many studies testing children's ability to tell if images are AI generated. Our research suggests that accuracy may increase with experience and age.

Dependent variables: whether the child answers correctly or incorrectly, percentage of correct answers and accuracy rate. Independent variables: 9 AI images and 7 non AI images. Controlled variables: same number of AI and real images for every child. same image size, same printed images used to show, same lighting conditions and same instructions given to each child.

We tested our hypothesis by conducting a survey of St. Luke's students. We chose 9 AI generated images and 7 real images. We instructed the students to tell us if they thought the image was real or AI generated and showed the images one at a time to each student. We recorded our data in a chart for later analysis.

We recorded in a spreadsheet if the student correctly identified each AI image so we had the number of AI images correctly identified for each student. We recorded the students grade and gender. We didn't record any other personal information. observation chart is attached

We grouped our data into Division 1 (K-Grade3) and Division 2 (G4-6) and also grouped by gender. We added up the number of correct responses for division 1 students and divided by the number of students (to get the average correct responses). Then divided this number by 9 to convert to a percentage. We did the same for division 2 and girls and boys. Division 1: 68% accuracy Division 2: 90% accuracy Girls: 74% accuracy Boys: 75% accuracy

We proved our hypothesis that older kids were able to identify AI images with more accuracy than younger kids. Future studies that test the accuracy of other age groups would be interesting. We also had some interesting outliers in our results. Two division 1 girls scored 94% accuracy rate, we think it would be interesting to investigate why some kids are much better than others.

More studies on this topic are needed. Our results showed the importance of making kids aware that not all the images online are real. We think kids should be educated in school on how to spot AI generated images.

Our sample size may not have been big enough to represent the general population. There was not equal numbers of division 1 kids and 2 kids and not equal numbers of boys and girls. Equal numbers would have represented the groups more accurately. Some kids might not have tried and just guessed. We tried to use AI imaged of different levels of difficulty but they might have been too easy or too hard.

we got some AI images from ChatGPT https://chatgpt.com we got some AI images from Copilot. https://copilot.microsoft.com we got a real image from national geographic https://www.nationalgeographic.com/photography

Background research: https://www.getcybersafe.gc.ca/en/resources/recognize-artificial-intelligence-ai-9-ways-spot-ai-content-online https://www.cbc.ca/news/canada/ai-detector-tools-test-1.7482913 https://www.cbc.ca/kidsnews/post/fake-ai-generated-images-on-the-rise-heres-how-kids-can-spot-them

thank you to Mme Cowan who helped us with our science fair project. thank you to Mme Jasmine who took part in running science fair. thank you to Mr. Owoc who ran the science fair and showed up to all the meetings. thank you to Laura (Stella's mom) who helped organize the trifold and filmed our video.

Link to presentation video (the presentation tab will not accept the link) https://photos.app.goo.gl/SnX9A84kWBffkLna8