Catching AI-Generated Content

Artificial Intelligence (AI) has been making headlines with the new development of OpenAI’s newest creation, ChatGPT. ChatGPT follows a generative content format which allows the program to create answers as if it were human. This means that a user could ask almost any question and receive what would appear to be a plausible answer. 

This also applies to drafting documents such as research papers and other written pieces. You can even have an AI write an article about detecting AI content:

“One way to detect ChatGPT-generated text is by analyzing the text for common patterns. For example, ChatGPT-generated text often follows a predictable pattern of repetition, with the same phrases and words being used multiple times. Additionally, ChatGPT-generated text tends to use simpler sentences and language than would be used by a human.” – Portion of an article written by GPT-3

This is revolutionary to search engines and as a research tool, but it also poses some serious challenges to things like education and authenticity.

What Is AI?

AI, or “artificial intelligence,” refers to a program designed to perform tasks that normally require a human to perform. Two types of AI exist out there: generative and retrieval. Retrieval AI will pull from a set list of responses, while generative AI will formulate its own answer from a database.

An example of Retrieval AI would be a chatbot on a website asking if you need help. The bot doesn’t know your answer but knows what phrases to look for and what information to retrieve for that question.

 

ChatGPT, GPT-3, and other AIs all fall under generative AI. These programs will take a question like: “Write me a book report on The Great Gatsby”, and do just that – write you a full-length book report. Generative AI has even been known to cite court documents and pass law school exams!

For a free legal consultation, call (725) 900-9000

How to Detect AI Content

Detecting AI-generated work is a difficult task since the entire point of the system is to replicate the work of a human. This replication isn’t perfect yet, but it is original work that isn’t always picked up by standard plagiarism detection.

The first trick to detecting AI-generated work is to read the document carefully. AI tends to use a simpler language model for writing these documents. AI can also draw conclusions or use information that may not make sense or that may be outright false.

There are many tools we can use to analyze the writing style or pattern of the work and determine if the pattern fits that of an AI. Such detection will determine the likelihood whether a piece was written by AI or a human being.

Issues with AI Detection Software

One of the newest detection software programs is Classifier from OpenAI. This detection software is still in the beginning stages of reliability. OpenAI even admits on the site that false positives or negatives are possible and not to use the resource as the only test.

To input a prompt, you must provide at least 1000 characters, which then prompts the following warnings:

• The classifier isn’t always accurate; it can mislabel both AI-generated and human-written text.
• AI-generated text can be edited easily to evade the classifier.
• The classifier is likely to get things wrong on text written by children and on text not in English because it was primarily trained on English content written by adults.

Classifier also gives examples of what human-written, AI-generated, and misclassified text are. Other websites and tools suffer from similar accuracy issues. As databases build and more papers are submitted, the more accurate the readings will become, but analyzing the documents yourself is still the best solution for the moment.

Click to contact our personal injury lawyers today

Why Is AI Detection Important?

Detection is important for the verification of the author of a document. This typically becomes an issue in the area of education but has been worrying in other sectors due to the tendency of ChatGPT and other AI to provide misinformation. 

Education is the largest concern due to the possibility of students using AI to cheat. With the capability to write papers in a few minutes or even seconds, AI is tempting to use for those last-minute research papers. 

Another area of concern is blog posts and other content that can post misinformation in areas that people rely on. Subjects such as medicine and law are the top concerns. Chatbots have also been spreading misinformation linked to politics and other conspiracy theories.