Unveiling Why Deepfake Faces Look Astonishingly Real! Believing Before Doubting the Fake — The "Hyperrealism" Created by AI Faces

Unveiling Why Deepfake Faces Look Astonishingly Real! Believing Before Doubting the Fake — The "Hyperrealism" Created by AI Faces

Why AI Faces Appear "More Real Than Real"—Unexpected Pitfalls Revealed by Brain Imaging

When you see a face on social media, it doesn't feel off. The expression is natural, and there's a certain familiarity. You think, "This looks like a regular person," and you trust the profile or message.

However, that sense of reassurance might be the entry point to a mistaken judgment.

A research team from the University of Florida reported a phenomenon where AI-generated faces are rated as more realistic than actual human photos. This tendency, known as "deepfake hyperrealism," goes beyond simply failing to detect fakes.

The features we use as clues for "realness" might actually lead us to choose fakes.


A More Troublesome Reversal of Judgment Than "Not Seeing Through"

When people hear "deepfake," they might think of videos where a celebrity's face is superimposed onto another video. The study in question dealt with static images of AI-generated faces.

According to the university's announcement, the research team prepared 40 photos of real people and 40 AI-generated faces. Participants were asked to evaluate their realism while their brain activity was monitored.

The results showed that both younger and older participants generally rated AI faces as more realistic than actual photos. This tendency was stronger among older participants.

It's important to note that this doesn't mean "everyone is fooled by every AI image." The bias in evaluations was observed under specific images and experimental conditions.

Nevertheless, the weakness of relying on one's own senses becomes apparent. The judgment of "it looks natural, so it must be real" is not as reliable as one might think.


How Did the Brain's "Area Involved in Detecting Incongruity" React?

The research team focused on a region deep in the brain called the "anterior insula," which is involved in processing important information and potentially unreliable cues.

The study found that activity in this region tended to decrease when viewing AI faces compared to real photos. There was also a correlation between lower activity and rating faces as more realistic.

However, it's not accurate to consider the anterior insula as a simple "fake detection sensor." This region alone doesn't determine authenticity, and finding differences in activity doesn't prove it caused judgment errors.

The results provide clues to understanding the phenomenon of naturally accepting AI faces. It doesn't go as far as saying "AI shuts down the brain's alarm system."


"Familiar Face-Likeness" Turns into Realism

So why do non-existent faces look so natural?

One explanation offered by Professor Natalie Ebner, who led the study, is the "typicality" of faces. Averaging many facial features weakens distinctive traits, bringing them closer to the typical face people imagine. Such faces are easier to process and may be perceived as more likable.

Of course, this doesn't mean AI simply overlays multiple photos. It's about how humans perceive the features that appear in generated faces.

In everyday life, we use the feeling of "familiarity" and "easy understanding" as shortcuts in judgment. However, sometimes this ease of processing is mistaken as evidence of an image's authenticity.

According to this explanation, it's not just unnatural images that require caution. Faces that are seamlessly viewed and reassuring also need the same level of scrutiny.


Focusing Only on the Elderly Misses the Point of the Study

In this study, the tendency to rate AI faces as more realistic was stronger among older participants. However, no significant difference in anterior insula activity was found between younger and older participants.

This means the explanation that "older adults believe AI faces because their brain region's response weakens with age" wasn't directly supported. The detailed reasons for age differences need further investigation.

Moreover, the same bias was observed in younger participants, which should not be overlooked.

Being familiar with smartphones doesn't equate to correctly judging AI images. Thinking "I want my parents to be cautious, but I'm fine" can lead to neglecting one's own checks.

When sharing with family, explaining "human senses have these weaknesses" rather than saying "older people are easily deceived" might make it easier for everyone to see it as their own issue.


On Social Media, Calls for Caution About "Confidence in Detection" and Verification of Sources

Discussions about the realism of AI faces continue on social media. Here, instead of direct reactions to the current paper, summarized opinions from related threads on the overseas forum Reddit are presented.

 

One opinion is that overconfidence in the ability to detect images is dangerous. Posts pointed out that not only the sophistication of fakes but also the assumption "I can tell" is problematic. Some posts candidly admitted a lack of confidence in distinguishing them.

The second opinion is that it's not enough for individuals to just scrutinize. There were calls for mechanisms to verify watermarks or image provenance at the time of generation, as well as display standards.

The third concern is about misuse for impersonation. There are worries that posing as a fictitious person to approach others could become easier with realistic face images.

On the other hand, there was also a view that this could be an opportunity to spread the habit of not unconditionally trusting images online.

These are individual opinions from a single thread and do not represent a survey of societal awareness. However, the question of whether "just looking more closely" is enough as a countermeasure aligns with the current study.


In Japan, Consider the Difference Between "Real Faces" and "Trustworthy People"

Bringing this study into the context of daily life in Japan raises the issue of how to handle profile photos on social media or the faces of people met online.

For example, the person you're consulting about investments uses a photo with a calm expression. The person who messages you daily shows a friendly smile. While these impressions might start a conversation, they don't serve as proof of identity or the accuracy of the conversation.

Even if the image is a real photo and not AI-generated, the person using it might not be the person in the photo. Moreover, even if it is the person's photo, it doesn't mean their story is trustworthy.

"Is the image real?" "Who is the person?" "Is the content of the conversation accurate?" Each needs to be verified separately.

Being aware of this distinction shifts the focus from observing faces to verifying information.


Develop a Habit of Verification Before "Seeing Through"

The implication for daily life from this study is the attitude of verifying information outside the image, especially at the moment you feel reassured by a realistic face.

Specifically, the following actions can be helpful:

  • Do not judge identity based solely on a photo. Verify affiliations and backgrounds through independent channels, not just the information sent by the other party.
  • Verify contacts claiming to be acquaintances using previously known contact details. Even if a new account's photo closely resembles the person, do not trust it solely based on that.
  • Pause judgment on stories that rush you to send money or provide personal information. Separate the impression of the person's face from the reasonableness of their demands.
  • Do not treat AI assessment tool results as identity guarantees. Image assessment and verifying the other party's credibility are separate tasks.

Researchers emphasize knowing about these cognitive biases as the first step toward countermeasures. However, this experiment did not verify how much actual fraud damage can be reduced by raising awareness.

It's important not to think "knowing will definitely prevent it," but to have a verification procedure after knowing.


Realism Is No Longer Proof of Identity

This study questions not only the advancement of AI but also what we usually base our trust in others on.

Natural expressions, friendliness, familiar facial features—these impressions are important in building relationships with people. However, they do not guarantee the existence of the person behind the screen or the correctness of their statements.

What is needed going forward is not to doubt every face but not to end verification with just the face.

What you verify next after feeling "it looks real" becomes increasingly significant as AI faces become more common.


Sources

  1. Eurasia Review:Introduction to the University of Florida's research, including an overview of the experiment, explanations by researchers, and results on age differences and brain activity.
    https://www.eurasiareview.com/16112026-deepfake-faces-appear-hyperreal-these-brain-scans-may-explain-why/

  2. University of Florida: Research Introduction
    Composition of face images, interpretation of hyperrealism, anterior insula response, and explanations by researchers for caution.
    https://news.ufl.edu/2026/11/deepfake-faces/University of Florida

  3. Social Cognitive and Affective Neuroscience: Research Paper
    "When Fakes Feel Real: Behavioral and Neural Mechanisms of Deepfake Hyperrealism in Younger and Older Adults." Reports on evaluation tendencies in younger and older adults and the relationship with anterior insula activity. Note: The number of participants differs between the abstract and press articles, so this article avoids specifying the number.
    https://doi.org/10.1093/scan/nsag086academic.oup.com

  4. Reddit: Related Discussions on AI-Generated Faces
    Reference source for opinions on caution about confidence in detection, concerns about impersonation, and calls for verification of provenance and display standards, as introduced in the SNS section. Not direct reactions to the current paper.
    https://www.reddit.com/r/artificial/comments/1rb0619/fake_faces_generated_by_ai_are_now_too_good_to_be/