As developments in AI continue to advance, it has become more and more challenging to distinguish between content that’s human-written versus content that’s AI-generated. According to a Tooltester survey, over half of Americans couldn’t accurately distinguish between content written by humans and content written by ChatGPT. This has led to the development of a wide range of AI detection tools, like Originality.AI which can help identify AI-created text with greater accuracy. Other tools exist to help discern AI-edited videos or AI images.
Before we jump into how to check if an article is written by AI, it’s important to better understand how AI generated content is created in the first place. AI content is created by algorithms, or systems which have been trained on a variety of databases that are made up of lots of different materials. AI systems use techniques like machine learning and natural language processing (NLP) and deep learning to mimic human writing and structure across many different languages.
In the early days of AI, it was easy to identify content written by machines because it sounded robotic and formulaic. Oftentimes it wouldn’t make sense or was outright incorrect.
AI writing truly hit its stride in the 2010s though, with the introduction of neural networks. These breakthroughs enabled the AI to “remember” content across larger swaths of text which in turn made its writing more coherent and sophisticated.
Add to that the introduction of transformer architecture (seen in models like Google’s BERT and ChatGPT) and you had the potential for billions upon billions of different parameters, making it possible for AI to produce a wide range of content on different subjects, as well as in different styles and tones.
If AI is so adept at “writing like a human”, why not let it? There are several underlying reasons why we don’t give AI free reign to write content on every topic under the sun, and these reasons span academic, ethical, legal and even societal reasons, including:
How would students develop critical thinking, creativity and a deep understanding of subjects if they just let the AI write all of their assignments? In academic settings, there’s also the very real risk associated with plagiarism. Students and researchers are expected to produce original work. AI generated content has already been to blame for high profile incidents like an attorney citing fake cases that were generated by ChatGPT in a legal brief.
Academically, there are stiff penalties already in place for cheating and plagiarism across numerous universities. As with many other writing assistants out there, like Grammarly and ProWritingAid, ChatGPT and other systems like it are just that – tools. They’re not meant to substitute for the discerning, deeper, more innovative and creative nature of humans. They can be great for outlining articles, breaking writer’s block and jump-starting the creation process, but they can’t take the place of a well-versed writer or student, which is precisely why AI detection tools are needed.
Journalism is built on trust. Knowing that articles were researched, written and published by humans helps keep journalists and the media as a whole accountable and ensures that they adhere to a quality editorial process that people can follow.
In addition, journalists provide the kinds of insights and analysis that go deeper than what ordinary AI, even cutting-edge systems, can provide. This is particularly true of breaking news or more nuanced discussions where a deeper understanding of the forces or issues at play is vital.
Beyond journalism being rooted in credibility and deep investigative research, there’s also the ever more convincing specter of fake news, misinformation and disinformation campaigns. AI can be used to create highly-realistic looking (but false and misleading) likenesses of celebrities, political leaders and many other professionals, which can exacerbate the spread of fake news and misinformation. Knowing the source of such content, video, images or otherwise, helps contain its spread.
Legal questions come into play with the use of AI when it comes to the authorship of a piece. Since AI cannot “own” a copyright, and since it has been trained on how to “write like a human” by using human texts, legal scholars have discussed the implications, especially derivative works, giving credit to the original authors and much more.
Although there’s no set-in-stone agreement in terms of how to handle intellectual property and copyright when it comes to AI, it’s necessary to be able to detect it and differentiate it from human writing in order to assign copyright in the first place. Doing so can also help with copyright infringement and fair use issues.
You may have heard of a company using Tom Hanks’ likeness in the promotion of a dental plan which he never endorsed. AI can be used to mimic the likenesses of others with convincing realism. Understandably, not only are celebrities and politicians upset but marketers and advertisers too are very interested in charting unexplored territory with the likenesses of others in creating AI-driven commercials and ads.
Going back to the aforementioned dental plan. Imagine if all of those smiling customer testimonials were fake? Shouldn’t you, as a consumer, have the right to know that in order to be able to make an informed decision about the quality of the service you’ll receive, if you choose to sign up at all? You can’t put the genie back in the bottle, so these are the kinds of questions we have to answer now, as a society.
You may not realize it, but AI generated content is already starting to have an effect on the public at large. This content can be skewed, biased or completely erroneous. It can also have a ripple effect on others and society at large.
Imagine a scenario where an AI created a highly-influential blog series about living as a global nomad. The blog would detail the adventures of a person traveling the world, sharing the life lessons they learned and the challenges they faced. If people knew that the blog was written by AI, rather than a person, it would change their perspective completely in that they’d be much more skeptical.
Such a blog written by AI might romanticize the idea of leaving a stable job to travel the world, making drastic life changes without considering the consequences. It would also set a high bar for what an ideal life should look like (think glamorous Instagram photos) which in turn would influence societal norms and expectations.
However, it wouldn’t be all negative – the blog would help bolster local communities and boost the travel and outdoors industries. It might also become so well-known that it shapes public discourse and forces us to talk about key topics like world-life balance, sustainability in travel or cultural-exchanges.
At the same time, all of these would be rooted in purely AI-generated experiences which begs the questions – would you follow the advice of such a blog knowing that it was AI generated? Shouldn’t the blog have the transparency to inform you so that you can make your own decisions?
Now that you better understand the history and the implications of AI writing, how do you actually go about detecting whether or not an article has been written by AI? There are several different routes one can go to arrive at the same destination. Here are a few of the ways to tell whether or not an article has been written by AI:
Because AI content is generated using patterns, it makes sense that content written by AI would exhibit those same patterns. Through a linguistic analysis, we can see certain features like repetitive phrases, strange syntax or a lack of nuanced expression that you’d normally find in human writing.
Despite its advances, AI still has trouble maintaining consistency – particularly in longer texts. Looking for factual inconsistencies or shifts in tone or theme is a sure sign of AI authorship.
Many advanced platforms use statistical models to look at the complexity of the text, the sentence structure, and the use of vocabulary to determine if the content has been written by AI. Many AI-written texts tend to show uniformity in their sentence structure and length which is what separates it from human writing at-a-glance.
Machine learning models (like Originality.AI) are trained to differentiate between human-generated and AI-generated texts. They leverage many different features of the text, including the aforementioned methods, to learn how to identify patterns that are telltale signs of AI writing.
Although AI article checkers are able to keep pace with advances in AI writing tools, the lines are starting to blur as AI becomes more adept at understanding nuance, subtlety, and other key signs of human writing. AI is also learning how to mimic the writing styles of specific authors, which makes it more difficult to determine based on the content alone.
What’s more, content that combines human and AI writing often presents a considerable challenge for many of the detection methods we have in place today. There’s also the risk, with any AI article checker, of false positives: text that is completely human written that has been flagged as AI. No AI detection software can guarantee 100% accuracy, 100% of the time.
AI is on a path toward exponential and explosive growth and development, and today’s AI article checkers, software and platforms must keep pace with it or risk falling behind. You’ve seen how such a platform is necessary, especially considering that scholars, researchers, legal analysts and other professionals continue to wrangle with the major questions of using AI on an academic, legal, philosophical, journalistic and worldwide basis.
It’s entirely possible that in the near future, we’ll see more advanced machine learning models, better linguistic analysis and potentially even AI systems designed to detect AI material. One thing is certain, however, and that’s that we’ll continually need the collaboration and insights of linguists, computer scientists and subject matter experts all working together to help create and develop more advanced detection models.
The fact is, AI isn’t going anywhere. Despite some countries’ attempts to ban, circumvent, stall or otherwise try to apply a quick-fix to dealing with it in their own way, it’s here to stay. And with it, AI-generated content. Although the methods shared here will give you a solid starting point toward how to detect if, AI article detection in and of itself isn’t just a technological endeavor. It’s something that we, as its developers and creators, should work together. The future of authenticity and credibility in the digital age depend on it.
Originality.AI was built to not only check for AI writing, but much, much more. With a thorough and robust option to also check for plagiarism and future-proof your site from the potential of Google penalizing AI content (especially as it continues to become more and more pervasive across the web). As we continue to develop and refine our AI article checker and other tools, you can trust that our advanced technology and machine learning algorithms will keep pace with ChatGPT and other AI-driven writing platforms as new developments emerge.
These features, combined with our thorough readability checker and fact-checking aid, help to ensure that every piece you publish is uniquely human, with all of the credibility, authenticity and originality that comes with it. You can also choose to scan different articles individually, using AI detection for some pieces, plagiarism detection for others, or all of the features together simultaneously through our easy-to-use credit system.
See for yourself how easy it is to check if an article is written with AI as well as scan your entire site to see how AI usage can affect your potential future ranking on Google and much more.
When you stop and think about it, writing is actually pretty amazing. We’ve got this manifestation of thoughts and ideas that we put into a structured form of language so that others can understand us. But within that language itself, there are irregularities and unpredictabilities.
