What is an AI detector and how does it work?

Although they are not infallible, these tools analyse issues related to language, coherence and meaning in texts to determine whether they have been written by artificial intelligence or not.

Communication Team

Telefónica Follow

Reading time: 3 min
  • AI detectors do not provide yes or no answers, but rather a degree of certainty that the text analysed was created by AI or by a person.
  • Through analysis of the texts, from the perspective of language, context and coherence, they analyse whether they have been written by a person.

What are Artificial Intelligence detectors?

AI detection, as its name suggests, refers to the ability of certain platforms to identify whether a text has been written by a person or by a computer program.

Parallel to the increase in the popularity and use of Artificial Intelligence, the use of these detectors has also been on the rise, with particular incidence or relevance in academic, creative or professional environments where authenticity must be paramount.

Through algorithms that have learned from numerous examples of texts written by both humans and machines, AI detectors do not offer a black-and-white answer, but rather a score to measure the degree of certainty with which the text was written by a person or not.

But let’s see how these AI detectors work.

How Artificial Intelligence detectors work

AI detectors usually work by combining tests.

For example, machine learning models analyse different characteristics of texts, such as style, sentence structure and length, vocabulary or the hypothetical repetition of unnatural patterns to identify which ones may have been produced by Artificial Intelligence.

In a specific case, such as sentence length, humans tend to alternate between longer and shorter sentences, while AI models replicate structures that are more homogeneous.

When it comes to vocabulary, people have a tendency to frequently use expressions or words with which they are more familiar in writing.

In contrast, Artificial Intelligence models display different patterns depending on the data they have been trained with.

Another issue analysed goes beyond the merely linguistic and studies the creativity, coherence, meaning or level of depth that texts may have, since, in general, those written by human hands can have a greater degree of originality or a higher level of detail and specificity.

One of the capabilities of AI-powered text analysis and detection tools is that they can detect specific fragments that they find suspicious, not just the text as a whole.

Limitations in detection

It should be noted that, despite everything we have analysed throughout this article, no detector is infallible: they can produce both false positives (considering human texts as written by AI) and false negatives (conversely, failing to identify texts that are actually written by AI).

To this should be added possible subsequent ‘manual’ editing of the text using AI, a capability that, paradoxically, AI models themselves also have: ‘humanising’ texts to make them appear more like they were written by a person.

Therefore, the recommendation is to combine AI detection tools with human resources, especially in contexts where authenticity is crucial, such as academic environments.

Is using AI to write texts plagiarism?

One of the debates generated regarding the use of AI-created texts is the possibility of plagiarism.

A recent study by the University of the Basque Country published in the journal Interactive Learning Environments refutes that the use of AI is a trigger for plagiarism and focuses on the fact that ‘the problem is not the technology itself, but how it is used and the academic context in which it is used’.

However, it should be pointed out that—and this is not exclusive to the use of AI models—if the text in question generated by the tool is presented as one’s own creation, without substantial modifications or proper attribution, it could indeed constitute plagiarism.

On the contrary, the appropriate use of this technology, understood as the editing of the text or the correct indication of sources, does not constitute plagiarism.

Share it on your social networks


Communication

Contact our communication department or requests additional material.