Researchers discover the impact of hostile images on AI and human perception

Recent experiments published in Nature Communications have uncovered a link between altered images affecting artificial intelligence (AI) systems and their impact on human perception. Exploring this connection highlights the need to ensure the safety of AI and its impact on our reality.

Researchers have found that subtle changes to digital images, specifically designed to fool computer vision, can also distort people's perceptions. These changes, known as ad attacks, can fool AI into misclassifying content, but also impact a person's ability to perceive the world around them.

Interestingly, even the smallest distortions that do not affect pixels more than two levels on a scale from 0 to 255 can effectively mislead the AI, highlighting its vulnerability.

This research is not limited to the digital world. Physical objects, such as displaying a stop sign as a speed limit sign, may also be vulnerable. Researchers are actively looking for ways to protect against such attacks, given their potential security implications.

In addition, scientists conducted experiments that revealed the impact of hostile changes on human perception. Participants were presented with pairs of images, each of which was subject to an enemy attack. For example, a source image classified as “vase” was altered to fool the AI ​​into seeing “cat” or “truck.” Participants were then asked target questions, such as “Which image looks more like a cat?”

Even though neither image resembled a cat, participants consistently showed a perceptual bias toward one of the altered images.

This finding highlights that people are also susceptible to altered images, even if the changes affect some pixels. This impact on human perception highlights the need for further research in this area.

It is important to note that the impact of hostile changes extends beyond machine perception, affecting human decision-making. This raises the need not only to protect AI, but also to understand how technology impacts people.

The need to develop safer and more reliable technologies is becoming increasingly important in our fast-paced world, where the boundaries between impacts on machines and people are becoming increasingly blurred.