Many students today view the use of artificial intelligence as a quick fix for writing an assignment, or worse, a dissertation. It is a fact that awareness of its use is growing rapidly within schools and universities. But another problem is its emergence in numerous online sources. Without using AI, one can therefore find oneself citing ideas... from AI. How can we currently avoid this and verify the human nature of our sources?
Why try to identify a text written by AI?
The reliability of sources is a topic covered from secondary school onwards and assessed throughout your studies. Cross-checking sources, verifying the author’s identity, and understanding the context in which a text was written ; are all tasks that may seem tedious to you. However, with the rise of AI, this has become even more important. Even before reading a source, it is essential to verify its authenticity. An AI would indeed have far less legitimacy than a genuine researcher specialising in the subject, which could risk wasting your time reading an irrelevant document.
By using a text written by AI, you also run the risk of plagiarism. Even if the text is not detected as such—since it has been reworded and is therefore difficult for anti-plagiarism detectors to match with the original text—it will still fail to cite the multitude of sources used, posing a serious intellectual property issue.
These two major issues make it important to know whether a text has been written by AI in order to determine whether or not you can use it as a reliable source, thereby ensuring the academic integrity of your work.
How does generative AI work?
Before you can spot text generated by AI, it is essential to understand how it works. How does generative AI work?
Generative AI operates using machine learning models that draw on data from a multitude of online sources (such as web pages, news articles, blogs, forums, etc.), as well as interactions with its users and trainers. The AI has been trained to formulate sentences with perfect syntax and spelling, but it remains an algorithm that strings words together without understanding their true meaning. And this is important to understand, because it is this weakness that will enable us to identify its output!
How can you tell if a text has been written by AI just by reading it?
All texts written by AI share similarities that make them identifiable at a glance.
A text written by AI tends to have a flat style. This is evident in its consistently monotonous rhythm, simple vocabulary and neutral tone that avoids taking a stance. Whilst there will certainly be no syntax errors, the text will leave the impression of being flat and uninspiring.
This aspect can be exacerbated by repetitions. The vocabulary used is repetitive, and the phrasing is often the same. AI does indeed have identifiable linguistic tics, with certain words and expressions recurring regularly: crucial, innovative, that being said, delve into, at its core, facilitate, are just a few examples. Double adjectives and phrases such as ‘not only… but’ are also heavily favoured by AI.
In essence, text generated by AI will consist of generic content featuring long, generalised sentences, lacking depth or analysis. The text does not stray from the topic, but it is of little interest. The algorithm picks up on the most common ideas, which results in a lack of contrasting views and dissenting or minority opinions. Furthermore, if you see several sentences repeating the same idea without adding anything new, this is almost certainly the work of AI. The text is also often devoid of concrete examples and real references.
AI is indeed prone to recurring inconsistencies. This results in contradictions within its own statements and in ‘hallucinations’. These errors can be found particularly in the references, with references that are partially or entirely fabricated: you will therefore need to check whether you can find the exact source and whether all the information (author’s name, date, title, etc.) is accurate. Furthermore, an ‘author’ who has requested a bibliography from an AI and simply copied and pasted it may also display an unfortunate tendency to include, after each reference, the supposed contribution of that reference to the subject.
Finally, a text that has simply been copy-pasted from an AI can be detected by its layout, particularly by sequences of ideas presented in bullet-point form.
What tools can help identify text written by AI?
If an algorithm can be created to generate sentences, then one can certainly be created to detect them too! However, generative AI has been one step ahead of anti-AI detectors, which have had to engage in a frantic race to develop by training themselves to recognise AI writing styles and offer effective services to users. Today, several anti-AI detectors are available online, some free and others paid. Unfortunately, their reliability is often called into question, with frequent false positives and AI-generated texts going undetected. Free software unfortunately seems less reliable, as the development and maintenance of algorithms come at a definite cost.
Frequently Asked Questions (FAQ)
Is text generated by AI and then edited by a human still detectable?
Detection is more difficult, as many clues have been removed. The form will be affected and may not necessarily be detected, particularly by algorithms, but if the content remains the same, the quality of the document will not be any better and this will remain visible even to the most experienced eyes.
Can we still use AI?
Yes, you can, for example, use it as a brainstorming tool, but you will need to make the ideas your own by rephrasing them in your own words and enriching them with your own thoughts and experiences.
What should I do if I’ve used a source that turns out to be AI?
We recommend that you remove it from your bibliography and revise the sections where you have used it, searching for new, reliable sources using the same keywords; this will avoid any risk of plagiarism and may even enable you to enhance your work.
What should I do if I have doubts about a source?
Be sure to check the writing style and bibliographical references carefully; these are the easiest clues to spot and the most revealing. If you are still in any doubt, it is best not to use it.
Could I face penalties if I use a source that has been written by an AI?
If this is unintentional, there is almost no chance that you will be penalised. However, if you are writing a literature review and the authenticity of your sources counts towards your mark, and several of your sources are generated by AI, then this could indeed affect your mark.