Spotting A.I. Generative Material With Its Inherent Flaws

Generative A.I. has been a blessing for making tasks easier, the process of work faster and consequently making room for leisure, but in parallel, artificial intelligence also has become the biggest source of cheating and a way out for people not interested in doing research and pretending to be hard workers. This ongoing advancements in A.I. has given headaches to professionals, companies and academies round the world, unable to distinguish man-written original works and A.I. generative.

Fortunately, like all technologies, A.I. operates with hidden flaws which takes off the mask of originality from so-called-original work. A.I. has a tendency of using same words frequently in a specific context without any apparent need of using those words successively. By comparing the frequency of these words in publications of the pre-A.I.-language-model era, A.I. generative works are easily marked out. Similarly, A.I. generative videos exhibit so many peculiarities not found in man-made videos.

Frequent Words in A.I. Generative Texts

The researchers from University of Tubingen and Northwestern University analyzed 14 million research papers published between 2010 and 2014 to find out the frequency of all words over the years.

1) Delving In

The researchers realized that the word ‘delve’ has been used 25 times more than it was used in the pre-A.I. era. The same goes for the words ‘showcasing’ and ‘underscores’ whose usage surged nine times in the previous three years. Similarly, the usage of words like ‘potential’, ‘finding’, and ‘crucial’ increased by 4.1%, 2.7% and 2.6% respectively.

2) An Intricate Play

The phrase, an intricate play, as well as the words ‘comprehensive’ and ‘pivotal’, also saw a surge in their usage in the past two years, clearly pointing towards the help taken from A.I. language models.

Some might claim this to be coincidental, having nothing to do with A.I., for languages go through evolutionary process, making some words obsolete and others more demanding. But the history of language usage vividly depicts that a sudden surge in the usage of words has mostly been related to a major event, specifically an epidemic or pandemic, like coronavirus. A similar surge was observed after the outbreak of ‘zika’ and ‘ebola’ viruses, making the words, zika, ebola, pandemic, epidemic and lockdown, more common.
The sudden surge in specific words and phrases in the post-A.I. era, is contrary to the trend observed during the pandemics. Unlike during the pandemics, there is no relation between the words or phrases and their sudden surge in usage. The only reasonable argument is that those works containing the words successively are the products of A.I. that frequently used the words as per its modeling.

Clues for Specialists

After using A.I. language models, like ChatGPT, an English specialist realizes that A.I. models always utilize the same patterns, moods and words mostly. For example, the imperative mood is their favorite while answering questions, and they utilize it successively. Similarly, a highly formal language permeates an A.I.-written text, with hardly any informal or less formal word, like ‘pal’, used. Moreover, figurative language or idioms are rarely employed by A.I. models for the obvious reason that figurative language involves a degree of emotions of human beings, and that its usage requires deep linguistic and contextual comprehension lacked by A.I.

The measure to detect A.I. in works has now become a necessity due to A.I.'s frequent conjuring up inauthentic references, making false claims regarding different matters and providing inadequate summaries.
As A.I. becomes more advanced year after year, it will most likely get over this panacea as well. Moreover, with research on the flaws of A.I. gets more intense, it will also be providing A.I. users how to filter out flaws from A.I. generative works and remain undetected.

A.I. Generative Videos

A.I. generative videos are more flawed than A.I. texts because they are visual, making them easily detectable by naked eyes.

1) Gibberish Text

Artificial intelligence, even being a language model, does not have the capacity to comprehend a language as humans do. For this reason, you will find either there is no text in A.I. generative videos/images or there are absolutely indelible words written on different buildings or objects appearing in A.I. videos.

2) Slow and Fast Movement of Frames

A.I. generative videos frames move either too fast to hide flaws, or too slow for the same purpose. All flaws come to the front on pausing these videos at small intervals.

3) Weirdness

The most detectable flaw is the weirdness that one encounters in A.I. videos. No laws of science work in A.I. generative world. Sometimes the construction of buildings is weird as well as their alignment; some videos show people with no faces; in some people magically disappear; and some videos show humans mingling their bodies into one another, along with super weird hands.

4) Unnaturalness

A.I cannot depict natural human emotions as we naturally show. If you want a strange smile and a strange fluttering of hair, A.I. generative world is the perfect place for you. People depicted in A.I. videos try to imitate human emotions and gestures, but fail horribly.

5) Too Perfect or Too Imperfect

There are two possibilities in A.I. generative videos and images: either they depict imperfectness of human emotions, gestures and body parts, or perfectness of designs of cities and automobiles, too perfect to exist in the real world.

6) Context

Context matters and is the key to find out the authenticity of videos. Does the video in question have any other versions? What is the source of it, a news channel or a verified X account? Is any authentic source using the video as a reference?

When fulfilling a prompt given by us, A.I. does not really care about how human beings look or behave, rather it just creates its conception of humans or their behavior by mingling pixels together, resulting in a mess.

The flaws make A.I. generative content and videos detectable. No doubt, in the coming years, the rapid progress in the realm of A.I. might reduce these flaws to smithereens. But a clue will always be there for a diligent eye.

Image: DIW-Aigen

Read next: 

• AI Could Promote Rise In Distrust Of Digital Information, Google Study Warns

• The Future Of AI In Business: Microsoft Debunks Misconceptions Linked To This Groundbreaking Tech

• AI Bubble Close To Hitting Tipping Point As New Report Says Companies Need $600 Billion Each Year To Pay For Infrastructure
Previous Post Next Post