TechnologyThe keys to detecting fake people created by AI on dating sites and social networks

The keys to detecting fake people created by AI on dating sites and social networks

That person with whom you are talking on social networks, in a virtual chat, on Tinder or, in general, through a platform or website on the Internet, could not be who they say they are or, even, not exist. Although it seems so. Although it answers your questions. Even if I showed you a picture of him.

Among the many uses that artificial intelligence allows are also theft, crime, identity theft and fraud. According to an Israeli study, it only takes 9 AI-generated faces to impersonate between 40% and 60% of the identities of the world’s population.

The objective can be to make a bad joke or get millions of interactions, but it can also be about a scam that seeks to get money of people through Tinder, Only Fans, social networks and applications or platforms on-line.

“ChatGPT makes it very easy for cybercriminals to impersonate different people without extensive literary knowledge. [Pueden] take advantage of the linguistic model to generate more personalized messages and trick people into falling victim to cyberattacks,” explains Martin Cheek, a fraud and cybersecurity expert at SmartSearch, to Daily Mail.

Is ChatGPT secure? This is how cybercriminals are taking advantage of AI

Added to this is that human beings are unable to distinguish what is real from what is artificial 78% of the time, at least according to a study carried out by the Royal Society in collaboration with the University of Oxford and Brown University.

How can you tell the difference between a real person and a ‘fake person’ created by artificial intelligence? How to know when the AI ​​is behind a photo or a conversation? As the expression says, the difference (or the devil, in its literal translation from English) is in the detail.

Specifically, some of the details that it is recommended to pay attention to to know if it is a real person or an AI They are the background and the shadows, the language and certain parts of the body, such as the eyes, hands, hair, teeth, language and inconsistencies in general, according to different people, including cybersecurity experts contacted by Daily Mail.

Details on the face, such as eyes and skin

Photo of a man created by AI based on instructions from 'Business Insider Spain'.
Photo of a man created by AI based on instructions from ‘Business Insider Spain’.

Stability/Business Insider Spain

Some of these fake images of people have been generated in recent years by websites such as This Person Does Not Exist (This person does not exist), which allows the production of images generated by GANs (Generative Antagonistic Networks).

Although artificial intelligence and tools like ChatGPT or GPT-4 have emerged or have become fashionable in recent months, cases with this type of photo have been appearing for years. In 2021, a team of computer scientists from the University at Buffalo in New York (United States) created an AI application that sought to provide a way to detect deepfakes (deep fake) through the specular reflections of the eyes of a portrait.

To do this, they made use of what they called portrait adjustment and took into account different factors, such as the imaginary line that crosses the center of the eyes and whether their position is parallel to the camera; whether reflections from the environment are present in both eyes and their distance from the light source.

Thus, the scientists detected some “inconsistencies” in the eyes synthesized by AI, such as different geometric shapes or locations in different regions of said specular brightness in both eyes, although they admitted that the method needed to be perfected. Their inexpressiveness, their different orientation or the strange reflection of light in the iris can also attract attention, experts say to Daily Mail.

The outlet also points to other parts of the face that can give away if it’s an AI-generated image, such as slightly misaligned faces and strange looking, and small differences in skin tone and coloringsuch as spots, uneven tones, or flickering at the edges of the face.

Last year, a Twitter user warned in a thread that went viral about some elements that can help detect that an image of a person is not real, but artificially created:

  • The faces are very basic and he always has a neutral smile, just like the gesture.
  • The symmetry is almost perfect, especially in the eyes.
  • The texture of the skin is reminiscent of wax figures, although this has been perfected in recent times.

Hair

Young blonde girl created by AI from indications of 'Business Insider Spain'.
Young blonde girl created by AI from indications of ‘Business Insider Spain’.

Stability/Business Insider Spain

In response to the previously mentioned thread, by Lemonauta, several users of the social network claimed to have observed additional details that could help identify these fake human beings created by bots, such as non-circular pupils, teeth that mix with others, or hairs that stick together. blend with the environment.

“No one has perfect hair, except the fake people in the AI-generated images,” he says. Daily Mail, which warns of the absence of irregular or unattached strands of hair and of “small imperfections” in the hair that should appear if it were a real person. This is an example that puts the middle of a person image created by an artificial intelligence.

Hands

Image created by AI of a woman, with indications of 'Business Insider Spain'.
Image created by AI of a woman, with indications of ‘Business Insider Spain’.

Stability/Business Insider Spain

“In the case of artificial intelligence, there are quite a few things that it is quite bad at, especially the hands. Sometimes you can see distortions in them, and sometimes they can be too big or have an extra finger,” Cyril Noel-Tagoe, principal security researcher at British cybersecurity firm Netacea, told Daily Mail.

The hands drawn or created by an AI can have oddities and inconsistencies derived from the fact that the tool has been fed by a multitude of images that it uses to look for patterns, but sometimes without knowing that it must meet certain requirements related to its appearance (the shape of the hand, how is each one of the fingers…) or with the number (5 fingers).

Although the faces may be better achieved, that of the hands is one of the aspects that most attract the attention of a false person image. Business Insider Spain has used an AI tool (Stability – DreamStudio) to generate images of an adult man with a beard and an adult woman, smiling, with brown hair and “pretty hands”, and in all of them hands appear disfigured, odd, or with extra fingers.

Image of a fake woman created by AI with indications of 'Business Insider Spain'.
Image of a fake woman created by AI with indications of ‘Business Insider Spain’.

Stability/Business Insider Spain

background and shadows

Image created by 'This person does not exist' from data selected by 'Business Insider Spain'.
Image created by ‘This person does not exist’ from data selected by ‘Business Insider Spain’.

This person does not exist/Business Insider Spain

“AI-generated images often have some telltale signs that they’re fake, so it’s important to pay close attention. If the person’s outline is blurry, for example, or if small details like shadows don’t look real, you’ll know that the image is not,” explains Vonny Gamot, a senior manager at McAfee, to the British outlet.

Like the person, the shadows generated by the artificial intelligence are also false, since it does not have real lighting either. So, the shadows, the background and the outline of the person They can be signs that alert that the photo is false.

Lemonauta also recommends looking at the background of an image to see if it is created by AI: this is usually out of focus, since the algorithm is intended to generate faces, while confusing textures or strange objects may appear in the rest of the scene.

The language

Photo of a man created by AI based on instructions from 'Business Insider Spain'.
Photo of a man created by AI based on instructions from ‘Business Insider Spain’.

Stability/Business Insider Spain

ChatGPT and other AI tools are trained on a wide mix of texts that can go back several years or decades, so the words or phrases they use can sometimes seem like something out of another era.

Yeah sound strange, unnatural or old-fashioned and contain short phrases and repetition of termsit is possible that whoever is talking to you is an artificial intelligence, explains Daily Mail after contacting experts such as Thomas Platt, a specialist in bots at Netacea.

“If you are concerned that a criminal could use ChatGPT, you should be aware of the limitations of the AI. Ask him about current affairsLike what was on TV last night. These AIs are trained on historical data, so if you ask about yesterday’s episode of Love Islandit will be very difficult,” he advises.

The account activation date

If you are talking in a chat on-line or on social networks with a person and you think that there could be an AI behind it —or someone driving an AI—, you can do two more things, according to experts.

On the one hand, ask him to prove who he is in some way —for example, with a video call—; for another, see when you activated your account: If you joined last week, it could be another red flag.

Other details

Young blonde girl created by AI from indications of 'Business Insider Spain'.
Young blonde girl created by AI from indications of ‘Business Insider Spain’.

Stability/Business InsiderSpain

The woman in this photo above does not exist, nor is this one real. In addition to the above, other tips for spotting AI-created fake person images involve looking at body parts such as teeth He is suspicious if they are too perfect and regular.

Evan Ratliff, editor of Business Insider who has chased ‘fake humans’ on the internet, recommends looking carefully to start notice “certain oddities”: a close-setting ear, strange wrinkles around one eye, a slight crease on one cheek, or a shirt collar that sticks out much more on one side than the other.

Other examples that Ratliff points to in fake personas created by AI are things like wearing only one earring, having one side of the face shaved more than the other, or wearing rimless glasses.

For her part, Lemonita points to the framing of the photo, which is usually the same and typical of photos for identity documents, since the protagonists do not usually tilt or rotate their heads.

LEAVE A REPLY

Please enter your comment!
Please enter your name here

Latest Posts

Read More
More