Can you spot AI generated text?

Paul Gavrikov
3 min readJun 12, 2022

AI has already been responsible for some impressive feats of writing, including penning a best-selling novel and composing music that has been performed by world-renowned orchestras. But can you spot AI generated text?

Photo by Andrea De Santis on Unsplash.

AI is quickly becoming a force to be reckoned with when it comes to writing.

When it comes to writing, AI is quickly becoming a force to be reckoned with. Can you spot AI generated text? Some people may be able to tell if a piece of text was written by a machine, but this is getting increasingly difficult as technology advances. There are already many examples of AI-generated writing that have been mistaken for the real thing. With this in mind, it’s important to be aware of the potential for AI-generated text in the future and learn how to spot it.

AI has already been responsible for some impressive feats of writing, including penning a best-selling novel and composing music that has been performed by world-renowned orchestras.

In recent years, AI has become increasingly good at creating novel and accurate text. This is especially impressive when considering that, until recently, AI was not able to generate text that sounded natural to humans. One of the most famous examples of this is when an AI system composed a song that was performed by the world-renowned Vienna Philharmonic Orchestra. Similarly, an AI system wrote a best-selling novel in Japan. In both cases, it is amazing that the AI was able to create something new and compelling that humans enjoyed reading or listening to.

But can you spot AI generated text?

The question of whether or not you can spot AI generated text is one that has been asked a lot lately. With the advances in technology, it’s becoming more and more difficult to tell the difference between human-generated text and text generated by artificial intelligence. There are a few things that you can look for, however, that may help you to identify AI-generated text.

Repetition

If you are willing to type one word over and over, computers are not far behind you, and you will soon see similar words and phrases repeated in a text. This is most obvious when the writer is just regurgitating the same words without any real thought.

If you find a lot of repetition, chances are you’re reading AI-generated text.

Formality

AI has a habit of making text too formal.There is little or no variation in tone. No matter the subject, the text remains very formal, making it sound robotic and unnatural.

If you come across something that sounds too formal, it could be artificially generated.

Incorrect Grammar

AI has made great leaps in recent years, but it has little understanding of grammar. That’s why errors in a paper or a report may point to a lack of human involvement in the process.

Lack of proofreading

One of the biggest reasons why AI-generated text is less than perfect is that it simply can not be edited to the same level as human-generated text. AI is often trained on huge volumes of information and if the training data is not very good, the system would be unable to catch mistakes.

Lack of Knowledge

AI does not have a general knowledge. This is fairly obvious for any sort of AI system, because you have to teach it everything.

Consider this example:

Microsoft’s AI had a weakness for cat pictures. The MS team set up a competition to find the pictures of cats in the company’s data sets. They gave the best example of a cat to the AI and then created a game out of it. The prize went to the person who could show the neural network a cat image it could not spot. The company is now being flooded with cat pictures.

So, what now? Well, you might have guessed it already — or perhaps you have not. Everything written above was generated by an AI. None of the text except for the title was written by me and all I had to do is pick from suggestions to guide the AI to the final post. Shocked? What are your thoughts?
PS: I used www.copysmith.ai which is based on OpenAIs GPT-3 language model.

--

--

Paul Gavrikov

PhD student in Computer Vision working on Representation Learning in Convolutional Neural Networks