Why AI Detectors May Mislabel Human-Written Text (Full Transcript)

[00:00:00] Speaker 1: Can you detect AI content with some of the most common tools available today? Let me show you an example from a paper that I'm about to submit in response to a Nature paper as a commentary. I've come here to CopyLeaks which is one of the most advanced AI content detectors on the market today and this is a free version available but it functions similar to other ones. Copied and pasted my text here and lo and behold this generator detects AI content. If I scroll down here let me see what the content is and this actually thinks that I've written about half the paper here what it's highlighted in red is as AI. Now I'm 100% sure because I wrote this myself without AI that there is no AI content here. I've been accused before by some students as being an actual robot and AI but that's another conversation. These detectors have a ton of false positives and are currently not reliable not safe for use at work or academic purposes in my view.