Deepfakes: what they are and how to recognize them

By February 8, 2024 App World

I deepfake are nothing more than deceptive video content generated through artificial intelligence-based synthesis of human images. They are constructed by overlaying videos and photos through a machine learning technique. Deepfakes are spread mostly via social media to direct public opinion toward specific political and ideological topics or secondary themes of merchandising and advertising for profit.

In some cases deepfake content is difficult to recognize because it is so well done that it looks real. They often have as their subject matter public figures of different industries to make the source appear trustworthy. Even in Italy, many personalities from show business and beyond have had to deny false content in which their faces were illicitly used to promote commercial products.

Origin and history of deepfakes

In recent decades theartificial intelligence has made great strides in processing information to generate results that would require human intelligence. Thanks to the widespread use of increasingly sophisticated and powerful computers, for the past decade or so people have been using so-called Big Data to extract information. In this way, even smaller enterprises can take advantage of the use of more complex algorithms based on numerical analysis theories combined with advanced statistics.

Even more recently, further studies have been undertaken to simulate the neural networks, or patterns of neurons connected to each other and trained with example data. As computers have evolved technologically, the very complexity of such networks has increased in level, inserting new layers of hidden neurons to increase the difficulty of processing. These networks are called "deep" networks, and this is where deepfakes originated.

Deepfakes are thus deep fake content as they are generated with the use of deep learning, that is, algorithms that are based on deep neural networks and aim to learn human characteristics to simulate. The word deepfake first appeared in 2017 in Reddit and has been revived internationally by journalist Samantha Cole. It mainly involves fake videos, pictures or audio in which the person in question says or does things that in reality he or she never said or did. Basically, one face is superimposed on another body or a different audio is superimposed on a video.

Examples of the use of deepfakes

Today, deepfakes are used for four main purposes. In fact, they are made in academia to study the phenomenon, for fun and goliardia, for journalistic purposes, and above all to carry out criminal acts to the detriment of others.

The first example of an academic deepfake dates back to 2017 and is named Synthesizing Obama. University of Washington scholars extracted an audio of an old speech by the president and then created different videos in which Obama spoke those words with perfect realism and synchronism.

Of a more playful and entertaining type are the deepfake videos that feature actor Nicholas Cage, whose face is inserted into films he never made. In this case they are mostly experiments without ulterior motives, carried out by young programmers who want to learn these techniques. Of another category are the deepfake criminals showing Hollywood personalities involved in pornographic videos they have never had anything to do with. The goal is obviously extortion, and such episodes are becoming increasingly common.

Finally, there are the deepfakes made by media outlets to draw attention to a phenomenon that can have worrying implications. How can we not mention the fake Christmas message of the British channel Channel 4 in which a fake Queen Elizabeth addressed the subjects by saying improbable things. At the end of this video, the broadcaster showed behind-the-scenes footage of the service to give viewers an understanding of how deepfakes are made.

How to create deepfakes and how to expose them

The Reddit user who first mentioned deepfake also created FakeApp, the first application of its kind. Subsequently, other open source programs were developed such as. FaceSwap e DeepFaceLab, but all of these software are moderately complex to use as they require computers of medium to high RAM power and a good graphics card to train the neural networks.

One of the most common software to produce deepfakes is Reface. It is actually quite basic and offers rather simple and not always accurate results. With this app you can superimpose faces on objects, replace the face of a video using that of a photo, animate the face of an image, or make an image speak with lip movements consistent with audio.

More complex and sophisticated software, on the other hand, can generate highly realistic results, and it is often very complicated to guess whether the content is fake or real. In many cases there are videos that are easy to unmask because they contain glitch, or jumps in the reproduction of certain details, but with technological development these defects are much rarer today.

The MIT (Massachusetts Institute of Technology) in one of his article explained what should be the main flaws or discrepancies that are useful in determining whether a video is authentic or a deepfake. One should pay close attention to the details of faces such as cheeks, forehead, eyes, and eyebrows. Also, it is good to pay attention to the glasses and reflections, that is, whether they are consistent with the lighting.

Other elements to evaluate on the face are mustache, hair, moles and the size and color of the lips, as well as the blink of the eyelashes. Additional components to pay attention to are hands, which can often have a finger a plus or unnatural joints. Finally, MIT suggests checking skin color shades and light effects on various parts of a person's body.

Fighting and countering deepfakes

Now increasingly realistic and well-made, deepfakes are often used to commit criminal acts or exploit the image of well-known personalities to sponsor products and brands illicitly. The problem is also becoming relevant in our country, and some national and international bodies have taken action to inform a wider audience.

La European Commission published guidelines on the use of Artificial Intelligence in which a precise definition of deepfake is given (Article 52, paragraph 3). This article of the text calls for due transparency in indicating videos that have been manipulated. In this sense, Italy has not been unprepared and the Privacy Guarantor compiled a fact sheet to educate citizens on the topic and increase awareness.

Come possiamo aiutarti?

Digital Strategies
Web & Graphic Design
Software Development
Conversion Marketing
Più servizi

Il tempo è la risorsa più importante che abbiamo

Indicaci budget e tempi di realizzazione per aiutarci a valutare il tuo progetto.

In quanto tempo vorresti vedere il tuo progetto realizzato?

INDIETRO
CONFERMA

Raccontaci tutti i dettagli

Name*

Company

Email*

Phone*

Tell us about your project*