Conserved Land: Keep Out

10 / 10 confidence in my own ability to create believable AI-generated content.

10 / 10 confidence in the general public’s ability to create believable AI-generated content.

While I’ve been trying to mentally focus on the positive applications of AI tools for most of these activities, I find this activity terrifying.  It was FAR too easy to generate a false news article through ChatGPT, and while I can spot some problems with it, they’re not ones that would likely be picked up by someone skimming for info about the latest public policies and how they should feel about them.  While AI Content Detector accurately spotted my false article, I cringe to think that I could someday (or already?) feel the need to take the time and energy to run things through such a tool, and even then with unreliable results.

Interestingly, I find the article much more believable than the image (Stable Diffusion), and yet Content At Scale found a there to be a 79% chance the image was human-generated.  In this case, I think I could have picked the fake image myself better than the detector.  However, I followed the NY Times quiz suggested, and I found it impossible to choose between the real and AI-generated images in that activity.

One note about image creation:  I tried to make a sign with words in my image on both Craiyon and Stable Diffusion, and this proved challenging.  Craiyon “refused” to make the sign at all, and Stable Diffusion kept misspelling words (spelled correctly in my prompt) or even repeating words.  I ended up settling for a very simple message, which worked.

1 / 10 confidence in my ability to detect fake content.

1 / 10 confidence in the public’s ability to detect fake content.

2 / 10 confidence in AI’s ability to detect fake content.

(Answered above.)

Monica Przyperhart

Leave a Reply

Your email address will not be published. Required fields are marked *