AI Detection: Should You Be Worried?

In this article, we will delve into the topic of AI detection and address the concerns that many people still have about it. Despite the fact that I have repeatedly shared my stance on AI detection in my previous videos, it is surprising to me how many individuals still express worry about it. Let me reiterate my position once again: there is no need to be concerned about AI detection unless you belong to specific niches such as academia, writing for clients who use AI detection, or are engaging in questionable practices in the online world.

For the majority of people, AI detection should not raise any red flags. There is already ample evidence on the internet that AI-generated content ranks on the first page of search engine results, and I can personally vouch for this with my own websites.

Now, in this video, I will not only address the concerns around AI detection but also show you a trick to check your outputs without ever touching Turnitin, a popular anti-plagiarism software. Additionally, for those who are not interested in Turnitin specifically, I will compare various AI detection tools against each other in terms of their accuracy.

Let's begin by going through the whole process step by step:

1. Generate a List of Popular Essay Topics: Start by using a resource like JGPD to generate a list of the ten most popular essay topics. These topics will be used for further experimentation.

2. Create 1000-word Essays: Write a 1000-word essay for each of the ten topics generated in step one. Keep in mind that the actual word count may vary between 500 and 600 words.

3. Select AI Detection Tools: To conduct the experiment, it is crucial to choose the right AI detection tools. For this purpose, I have considered several factors, such as availability, limitations, and cost. The three tools I have chosen are:

   • Content at Scale: A free tool with no limitations.
   • CopyLeaks: Another free tool with some limitations, but easily bypassable by using a different browser or VPN.
   • Cross Black: Offers free trials per session, but again, by switching browsers or VPNs, you can use it repeatedly.
   • Originality.ai: This tool was selected to assess its performance as well.

4. Prepare the Essays: Now that the tools are selected, it's time to run the experiment. I have prepared a total of 20 essays for testing purposes. The first ten essays are the originals written by Chegeberry, and the second set consists of altered versions generated using an undetectable AI. For consistency, I used the "High School Essay" setting to ensure closer adherence to the human writing style.

5. Correlation Analysis: The objective of this experiment is to examine the correlations between Turnitin and different AI detection tools. By comparing the readings obtained from each tool for the original essays, we can assess their reliability. If Turnitin shows a high correlation with, let's say, Cross Black, we can conclude that we don't need Turnitin for future tests. The same applies to the altered versions; the tools that show the closest numbers to Turnitin will serve as proxies for further comparisons.

Now, let's explore the results of the experiment:

Original Essays:

• Originality.ai: This tool scores all the essays as 100% AI-generated. However, it doesn't prove to be a reliable metric when used in conjunction with Turnitin. There were outliers among the essays, indicating discrepancies between the two evaluations.
• Cross Black and Content at Scale: The readings for these two tools were inconsistent and showed negative correlations. Therefore, they are not suitable if you want reliable AI detection results.

On the other hand, the Basic Model of CopyLeaks, which is the free version, produces better results. Although none of the essays scored 100% AI, there was a 75% correlation between CopyLeaks Basic Model and Turnitin. Comparing the results on a graph reveals similar peaks and dips for these two tools, implying a considerable level of agreement.

Altered Essays:

• Originality.ai: Strikingly, this tool consistently marked every transformed essay as 0% AI and 100% human. This complete absence of variation across multiple iterations made correlation analysis impossible.
• CopyLeaks and Content at Scale: Here, CopyLeaks and Content at Scale aligned more closely, both indicating that the essays were 0% AI and 98% human. If you have transformed outputs and want to assess their compatibility with Turnitin, you can rely on these tools as proxies.

In summary, if you possess reliable methods of bypassing AI detection or use effective prompts found on websites like Strictly.a, you can confidently use CopyLeaks. Achieving consistent high scores with CopyLeaks suggests similar outcomes with Turnitin. Similarly, if you have transformed outputs and can't access Turnitin directly, using Content at Scale or Cross Black and observing the results can give you an indication of how Turnitin will evaluate your work.

To conclude, don't fret about AI detection as long as you're not involved in suspicious practices or operating in niches where AI detection is vital. Hopefully, this video and the experiment conducted shed some light on the topic and help you navigate the challenges associated with AI detection.

Remember to like, share, and subscribe to our channel if you find this information helpful. Stay tuned for more insightful content, and until next time!