OpenAI's Newest Tool for Detecting AI-generated Text

Have you ever read an article or text that seems just a bit too strange to be written by a human writer? It might just be that in fact it wasn’t! OpenAI has introduced a new tool to detect AI generated text such as by their own chatbot ChatGPT, but also different AI models that are used to generate language. This way human content can be distinguished from that created by AI.

The tool is the latest innovation from OpenAI, who recently launched the content generating model ChatGPT. Their latest model is able to detect and analyze content and determine if and which parts are written by AI. But why would this all mater? Language creating models have raised quite a controversy. For instance, a study from the Middlebury Institute of International Studies’ Center on Terrorism, Extremism and Counterterrorism found that the GPT-3 chatbot is able to create ‘’influential’’ text that has the power to radicalize people to extremist ideologies. That’s an example of why OpenAI feels the responsibility to try to minimize the harmful effects that the chatbot can create. OpenAI stated that “While it is impossible to reliably detect all AI-written text, we believe good classifiers can inform mitigations for false claims that AI-generated text was written by a human: for example, running automated misinformation campaigns, using AI tools for academic dishonesty, and positioning an AI chatbot as a human.” But besides that, there are countless other situations where it can be crucial to know if a piece of content was generated by AI. For instance, if you’re reading reviews of a certain product or service online.

So how does it work? It’s idea stems from the fact that text written by AI tends to carry certain characteristics, such as the way AI-generated texts are more repetitive than content created by humans. The company has trained a classifier to distinguish AI written text from human text. Then you paste a text into the box, and the system will tell you whether it thinks that the text is very likely, unlikely or unclear if its AI generated. But in a press release the creators of OpenAI have warned the public about its accuracy. They stated that it “should not be used as a primary decision-making tool, but instead as a complement to other methods of determining the source of a piece of text.’’ OpenAI also said that it identifies around 26 percent of AI generated text as ’likely AI-written’. Other limitations are that its very unreliable on texts consisting of less than 1000 characters and that its ‘significantly worse’ when used on texts in other languages than English. Besides that, the company also said that the system sometimes ‘’incorrectly but confidently’’ labels human written text as AI written text. But even at this level of accuracy it’s a valuable instrument to detect the influence of AI in written text.

The initiative of OpenAI to safeguard the possible harmful influence of its own system signifies the push for explainable, friendly AI. By distinguishing AI-written text from human work, those possible harmful effects might be reduced in the future.

Have we sparked your interest to learn more about all the possibilities in the world of AI? Contact us via our Baise website and let us know what you're looking for. We offer our personal advice and expertise on AI for your company. Our aim is to connect brains with business by providing you with talented and ambitious students looking to gain real experience in the field.