Beware of ChatGPT Auto-Detection!
Updated: Feb 2
Using AI to Police AI Can't Solve the Problem - Human+AI is Needed
OpenAI, the maker of ChatGPT, released a tool to detect auto-generated text. Fundamentally, they made a tool that identifies when their other tool, and similar tools, are being used. For anyone who is excited to hear this - its success rate is around 26% - so let's not celebrate just yet.
It's important to know that AI detection systems - and this is an exploding industry right now with new products and companies popping up almost daily - have their limitations and often won't be correct. They can give both false positives and false negatives as AI's ability to mimic human writing styles make it tough to tell if text is real or fake.
Instead of putting too much faith in AI detection systems, I encourage everyone to approach this situation in the manner that is always best with AI - combine AI with human engagement. Yes, you can use AI detection systems as a tool, but it requires human insight to determine whether the detection is accurate. You CANNOT effectively outsource your ChatGPT "cheating" concerns to another AI tool. Please don't be fooled by those who profess that this will work and do NOT automatically penalize students or professionals based on the results of these detection tools.
For educators, let's consider following the same educational path that has been established for using calculators. Just as calculators were introduced as a tool to support student learning, at the appropriate place in the learning journey, ChatGPT, and other generative AI tools, should also be seen as tools to support the learning process. Perhaps you could immediately mimic the already established processes used for calculators and the internet.
It's an incredibly exciting time to embrace generative AI, particularly ChatGPT! There's tremendous potential to impact productivity, learning, customer experience, revenue, etc. To do so, let's be prudent and think of building a working partnership with these new and powerful tools. Strike a balance between AI detection systems and human judgment.
Note: The detector only works on >200 word samples. It is free here (requires a log-in via Microsoft or Google, or signing up with an email address): https://platform.openai.com/ai-text-classifier