The Rise Of AI Fakes And How To Spot Them

In recent months, we have seen many examples of this kind of content spread across the internet. Such as images of Donald Trump being arrested, and even AI-generated ‘deepfake music’ of popular artists singing other people’s songs.

Worryingly, a trend with deepfake music has been to predominantly reproduce the voices of Black artists. Roberto Nickson, an AI commentator, was able to write an entire verse that believably sounded like Kanye West. What are the implications of being able to put words into the mouths of your favorite Black artists?

So how does it work? Put simply, generative AI uses an algorithm that has been trained on a huge dataset of pictures and corresponding captions. Once trained, it can combine the ideas, features, and styles from these images to create new original pictures. These tools can make increasingly realistic images, and it’s only getting easier to use them. This means we’re being exposed to more and more fake pictures every day, but we don’t yet know how to deal with them.

You might say that this is nothing new. Photoshop has existed for years after all. Generative AI is different from Photoshop for one key reason. With Photoshop, you can manipulate an image, but the image needs to exist in the first place. With generative AI, you can create entirely new images of events that never happened. We’re beginning to do this with videos as well.

This is the crux of the very real threat that we are staring down. It starts with an amusing photo of the Pope wearing a stylish coat. But where does it end? Social media platforms are already struggling to deal with online misinformation. How will we navigate an online future where you literally can not believe your eyes? 

Could you spot an AI fake?

Would you know what to look for to spot if an image has been generated by AI?

1. Look for a watermark or description

Sometimes people will tell you that the images they’ve shared were generated by an AI in the title, description, or comments. Likewise, some AI image generators will put a watermark on their images to show that they are not real. However, not all do, and some people might remove them deliberately. So don’t rely on this tip alone.

2. The tells are in the details

The pictures as a whole can often look very convincing. However, AI images often look hyper-realistic, that is, instead of looking candid, they look somewhat artistic and staged. Even more telling are the small details like hands, eyes, glasses frames, shadows, and reflections where AI image generators make mistakes. For example, hands can have too many fingers, or glasses frames can be mismatched. If there are words in the image such as street signs, they will usually be misspelled or nonsense.

3. Look at the source

Existing images are often used for reference by AI image generators. You can use reverse image search, like Google Images, to find similar images online and compare them with the suspected fake.

The future will be synthesized

There is enormous potential with generative AI, for good and also for evil. On the good side, it can be a tool that enhances productivity, creativity, and innovation. But there are serious ethical and social challenges that are posed by this new technology. The most pressing of which is misinformation.

Read: Black Women, Bias, And Burnout: Let’s Talk About Responsible AI

Social networks like Twitter, TikTok, and Facebook already struggle to moderate the mass amount of content on their platforms. Now we must prepare for the deluge of AI-generated images, audio, and videos that have already started duping people. 

Until we develop tools to help us detect and prevent misuse of generative AI, we will need to improve our AI literacy and be more skeptical of what we see online. The future will be synthesized, but it’s up to us to make sure it’s not distorted.