AI-generated text, produced by tools such as ChatGPT, is increasingly shaping our daily experiences. Teachers are experimenting with using AI in the classroom, marketers are eager to replace human interns, and meme-makers are having a blast. However, many individuals, including writers, are feeling anxious about the possibility of robots taking over their jobs.

AI tools that generate text are now widely available to the public, making it likely that we will encounter more synthetic content online. This could range from harmless auto-generated quizzes to malicious propaganda campaigns from foreign governments. As AI becomes more advanced, it can be challenging to distinguish between human-written and AI-generated content.

How AI-Generated Text is Impacting Daily Life

The rise of AI-generated text has been a hot topic of discussion in recent years, with some hailing it as a revolutionary tool that will change the way we interact with technology, while others are wary of its potential to cause harm. Despite the varying opinions, there is no denying that AI-generated text is already starting to impact our daily lives in a variety of ways.

In the classroom, teachers are experimenting with using AI-generated text as part of their lessons. The aim is to help students understand how language works, and to provide them with engaging and interactive content that will keep them engaged. Meanwhile, marketers are eager to replace their interns with AI tools that can generate high-quality content faster and more efficiently. Memers, on the other hand, are using AI-generated text to create new and exciting content that is sure to go viral.

The rise of AI-generated text has also raised concerns about the quality and reliability of the content that we are exposed to. For example, sophisticated propaganda campaigns could use AI-generated text to spread false information and manipulate public opinion. This is why researchers are working to develop tools that can detect whether a piece of text has been generated by an AI tool like ChatGPT.

One of the ways in which AI-generated text is being detected is by evaluating its randomness and variance. Tools like Harvard and MIT-IBM Watson AI Lab's experimental tool, GPTZero, scan text and highlight words based on their level of randomness. This allows users to see whether a piece of content was generated by ChatGPT or not. However, as AI text generators become more sophisticated, it is likely that these detection tools will become less effective, which is why researchers are exploring alternative methods to identify AI-generated text.

One such method being explored is watermarking, which would involve building rules into the large language models that power AI text generators. This would allow a watermark to be placed on certain words or patterns, which would be off-limits for the AI text generator. If the text is scanned and the watermark rules are broken multiple times, it would indicate that a human being wrote the text.

While watermarking is an interesting idea, it is not foolproof, and some experts are skeptical about its effectiveness. For example, Micah Musser, a research analyst at Georgetown University's Center for Security and Emerging Technology, expresses skepticism about whether this watermarking style will actually work as intended.

Detecting AI-Generated Text

Academic researchers are exploring methods to identify text that has been generated by AI programs like ChatGPT. Currently, the most straightforward indicator of AI-generated text is its lack of surprise or unpredictability. In 2019, Harvard and the MIT-IBM Watson AI Lab released a tool that scans text and highlights words based on their randomness.

Edward Tian, a Princeton student, created, a tool aimed at educators to determine the likelihood that a piece of content was generated by ChatGPT based on its "perplexity" and "burstiness." OpenAI has also developed a tool to scan text over 1,000 characters and make a judgment call on its authenticity. However, these tools are still limited in their accuracy and are most effective for English text.

Challenges and Opportunities for Detecting AI-Generated Text

While AI detection tools are useful at present, computer science professor Tom Goldstein from the University of Maryland believes that they will become less effective as natural language processing improves. He worked on a recent paper exploring the possibility of building watermarks into AI-generated text to help distinguish it from human-written content.

Micah Musser, a research analyst at Georgetown University's Center for Security and Emerging Technology, is skeptical about the effectiveness of watermarks in AI-generated text. He contributed to a paper that looks at strategies to counteract AI-fueled propaganda and highlights potential misuse and detection opportunities. The paper builds on Meta's 2020 study of detecting AI-generated images and focuses on relying on changes made by the creators of the model rather than those in charge of it.

In conclusion, AI-generated text is already starting to impact our daily lives, and it is likely to become an even more prominent feature of our world in the coming years. Whether this is a good thing or a bad thing remains to be seen, but it is important that we remain vigilant and work to develop tools that can help us identify AI-generated text and protect ourselves from its potential harm.