Can ChatGPT be Detected? The answer is maybe, Here’s why

You must be familiar with ChatGPT and what it can do. It has some great uses including researching for any topic, generating content as per the prompt, and beyond. The ongoing AI war has got us caught up as many tech companies have released their AI solutions. ChatGPT rose to its epitome after it was launched last year and reached 1 million users in less than 5 days. 

It has a total of 180+ million users with more than 1.6 billion prompts from people-generated content left and right. Undoubtedly, many people have doubts that you can detect AI-written content. Well, if that’s your concern too, let’s find out more about it.

Understanding Generative AI and ChatGPT

It’s crucial to know what ChatGPT is and why it is useful. This will act as a stepping stone to uncover the mystery around whether AI writing is detectable. 

ChatGPT is essentially a generative AI tool powered by large foundation and language models. These models can process user prompts (or queries/ commands), analyze the meaning behind using natural language processing algorithms, find relevant content from the internet, and generate responses carrying summarised and often informative responses. 

ChatGPT can generate text responses, however, there are multimodal generative AIs such as Google Gemini that can generate responses in multiple formats such as text, images, audio, videos, and music among others.  

Talking ChatGPT is extremely useful for educational purposes as you can ask the AI to write an entire project with correct prompts. Similarly, it can generate anything from content strategy, business plans, scripts, pull insights from data, and so on. 

We have come a long way since the original ChatGPT was released. Today, you can converse with the AI in a natural human-like manner, ask questions, seek out explanations, and generate creative content, or pretty much everything you can ever think about.

Keeping ethical and moral thinking aside, people have concerns if ChatGPT is detectable. What it means is, that whatever you write using ChatGPT (or any other tool), is it detectable using AI detectors like Turnitin or ZeroGPT? 

The short answer is NO to a great extent. It is because there are a lot of factors that could link AI-written content as such as most of the time, there are false positives. There’s an abundance of tools that specialize in detecting AI-written content, however, the results have always been a mix of positives, negatives, and false positives. 

Can ChatGPT Be Detected?

Over a couple of months, many AI detector tools have mushroomed on the internet that claim to detect the use of ChatGPT or other AI tools. These are in addition to plagiarism checkers that can detect copied or plagiarised content.

For instance, there’s ZeroGPT, Turnit in, GPT-2 Output Detector, and Writefull GPT Detector to name a few. Of course, some of the tools mentioned above may have a higher degree of AI-written content detectability, however, you need to understand that most of the people in charge may deploy one or two tools to discover AI content. 

The folks at ZDNet gave out several prompts to generate AI content and ran it through five ChatGPT detector tools. Let’s see a summary of these tests to understand it better.

	ZeroGPT	GPTZero	Content at Scale AI Content Detection	GPT-2 Output Detector	Writefull GPT Detector	Overall Score
Test 1	28.9% AI GPT generated	Likely written by human	100% likely written by human	99.98% real content	1% likely written by GPT	4/5 tools say it’s a human-written content
Test 2	51.22% AI GPT generated	–	3% likely written by AI	99.69% real content	12% likely written by GPT	3/4 tools say it’s a human-written content
Test 3	91.35% AI GPT generated	May include parts written using AI	99% likely written by human	99.98% fake content (AI)	77% likely written by GPT	3/4 tools detect use of AI

Analysis: After the tests, it is safe to assume that the results have been mixed. Some of the aforementioned tools were able to detect AI, however, it wasn’t always 100% true. There are instances when these tools gave out false positives while in some cases, the results showed the use of ‘human-written content’ as ‘AI-generated’. To summarize, these tests aren’t always conclusive at least at the time of writing this. 

At the end of the day, these apps are designed to detect AI content, however, many factors could influence their findings. 

How Can You Detect ChatGPT Content?

An AI-written content detector tool may not have the highest accuracy rate, however, humans are legit made differently. 

When someone writes a particular piece of content, they usually follow a unique writing style. On the other hand, generative AI tools rely on what you can call ‘average writing style’ that makes it easier to detect such content if it is just good but not too good or too creative based on the prompt These tools usually rely on a lot of extravagant words such as ‘elevate’ or ‘unleash’ or something that quite sticks out and may not look or sound natural. 

Another indicator is when ChatGPT or other AI tools repeat sentences with different connotations throughout the piece. You may come across repeated words or phrases that the AI could’ve simply summarised instead. This may not be apparent when asking for a simple query like ‘Define AI’ but when you add multiple prompts like ‘Define AI and how it is different from sentients in Terminator movie series’ the writeup may look different.

You can also inaccuracies or illogical statements that as a person of subject matter expertise may distinguish while AI may not, at least at this stage.

Another pointer is fairly easy to gauge, ChatGPT might reference ‘using AI to generate content’, especially in the concluding sentences. Of course, we would say a person detecting AI writeup will be more accurate if the person knows where to look out for it. On the other hand, ChatGPT detectors may not have 100% accuracy and thus, you cannot solely rely on their analyses as false positives may exist to a greater magnitude than you can expect.

You can follow Smartprix on Twitter, Facebook, Instagram, and Google News. Visit smartprix.com for the most recent news, reviews, and tech guides.