How to Detect Fake AI-Generated Images

Published March 22, 2024

Artificial Intelligence (AI) advancements have resulted in a surge of fake images spreading across the internet. These deceptive visuals, powered by generative AI tools, can now be created by anyone, even without technical know-how. Tools like DALL-E, Midjourney, and OpenAI's Sora have made the generation of convincing deepfakes more accessible than ever. Fake images can have benign appearances, but there's potential for misuse in scams, identity theft, and political manipulation.

Detecting Deepfake Content

In the early days of deepfake technology, imperfections in falsified images were more noticeable, making them easier to spot. However, AI has improved significantly, and spotting fakes has become much more difficult. Subtle signs, like an overly smooth or shiny skin tone in photos, can hint at AI manipulation. Yet creative touches can successfully mask these giveaways.

Look Closer at Faces

Face-swapping is a common deepfake technique. Experts suggest examining facial edges for discrepancies in skin tone or blurriness. When viewing a video, check if lip movements are perfectly synchronized with the spoken words. Pay attention to the teeth as well—they might look blurry or not align with a person’s real-life appearance.

Consider the Context

Contextual clues can often reveal deepfakes. Public figures doing something outlandish or inconsistent with their character may be a telltale sign. Skepticism is warranted especially when no other credible sources corroborate an unlikely image or video.

Using AI to Spot Fakes

Some companies have adopted the approach of using AI to combat AI. Microsoft has developed an authentication tool that provides a confidence score on whether content has been manipulated. Similarly, Intel’s FakeCatcher uses algorithms to determine a photo's authenticity. However, many such tools are restricted to a limited audience precisely to avoid helping malefactors improve their deceptive techniques.

Challenges in Fake Content Detection

While helpful pointers exist, the rapid advancement of AI means these tips might soon be outdated. Moreover, expecting average internet users to become digital Sherlocks could instill a false sense of security as detecting fakes becomes an increasingly complex task even for experts.

AI, deepfake, detection