From Midjourney, stable diffusion at the end of 2022, to Open AI's Chat GPT, DALL· E, Sora, Google's Gemini, AI has shown its unparalleled ability to write articles, draw, make pictures, videos, and more.
Watching AI show amazing skills again and again, I don't know if you have ever been worried, since AI can generate pictures and videos that are close to real people, as ordinary people, can we still tell whether a picture or a video is real in the future?
Such concerns are not unfounded. Let's do a quiz and see if you can find the only one of the four images below.
▲ These pictures were generated by the "This Person Does Not Exist" website: thispersondoesnotexist.com
Of course, in addition to portraits, AI can also generate a wide variety of landscape photos, as well as CG images that are themselves made with a computer. And to distinguish these photos, it is even more difficult for ordinary people.
Thankfully, in most cases, the ability to tell if an image is made by AI or not doesn't have a real impact on our lives. We have seen most of the pictures on the Internet.
And in recent years, there have also been some fake news made with AI images.
▲AI-generated image of Trump's arrest, screenshot from X (formerly Twitter)
While most of this fake news is hoaxes, if it floods the internet, it can cast doubt on the credibility of news events. After all, seeing isn't necessarily believing, so how do we know what's true and what's not?
▲AI-generated picture "Musk and robot girlfriend", screenshot from X (original Twitter)
How to tell if a picture is "real or fake"?
Since there are a lot of real and fake images on the Internet, how can we tell if an image is generated by AI? In fact, there are some simple ways to help us tell.
1. Use common sense to find things that don't make sense
The most direct and effective way to do this is to use our "common sense".
For the past year or two, AI-generated human hand features have been the focus of complaints. While some models, such as Midjourney V5, Stable Diffusion, have some optimizations for hands, if you look at the images in the AI fake news, you will still find some "weird hands".
▲DALL· E draws the hand, and you can still see the "unreasonable" hand features
In addition to the flaws in the hands, AI-generated images may also have some errors that don't make sense.
For example, in the aforementioned "Trump arrest map", there are several strange places.
▲ Illogical aspects of AI-generated images
First, a strangely oriented foot appeared behind Trump that belonged to no one. In addition, Trump's left arm was twisted into an abnormal shape. The passers-by behind him looked indifferent in the face of such a "major" incident, and the chasing police officers did not seem to be focusing on Trump.
Therefore, the common sense of the human world, as well as our knowledge of the physical world, is a huge weapon for us to identify AI images. If there are some details in the image that defy common sense and the laws of physics, it can basically be assumed that it is an AI-generated image.
2. Check the source of the picture
But some of the common sense errors in some images are hidden in very detailed places that are not easy to distinguish at first glance. There is also a simple way to judge this type of image.
Nowadays, search engines have an "image query" function, and many online websites can also provide similar functions. You can upload "suspicious images" directly. By looking up an image, you can find a website that has posted the image.
If for a very important news event, but the publishing media is invariably personal media or marketing accounts, then the picture is most likely fake. If the image has been published by reliable media and the content of the report is confirmed, then such an image may be true.
▲ The existing website and the image source query tool can directly search for the source of the picture, and the picture is taken from the Internet
3. With the help of identification websites
At present, there are many websites that can directly query the EXIF information of the image (recording the camera's lens, aperture, shutter and other parameters).
If someone sends you an image to show off their vacation or photography skills, you can upload the image directly.
If you can see the camera parameters, then there is a high probability that the picture is real, at least it means that it was indeed taken with the camera. And if you don't have this information, it's possible that it's AI-generated, but it's also possible that it's been compressed.
In addition to this, there are websites dedicated to identifying AI images, such as "AI or not websites". These websites can also identify whether a given picture is AI-generated or human-created, and while not 100% accurate, it can serve as a reference.
▲AI or not 对于两张图片的判断结果，图片截取自AI or not网站
Of course, for most news events, the heat will pass quickly, and it will not have a substantial impact on the lives of ordinary people. But the next "deepfake" technology to mention will have a direct impact on everyone.
Beware!AI Scam: Deepfake
Deepfake is a word that is synthesized by deep learning and fake. Its meaning is easy to understand, which is to use deep learning technology to fake some pictures and videos, which are generally dominated by humans.
In fact, deepfake technology has been around long before the advent of AI tools like GPT and midjourney. One of its most typical applications is AI face swapping.
In 2019, the role of "Cui Liulang" in the costume drama "The Twelve Hours of Chang'an" underwent an AI face change. In addition, in 2020's "Three Thousand Crows", "The Great Pediatrician" and other film and television dramas, AI face-swapping technology is also used.
And the recent "short drama to the sea", deepfake technology has also played an important role, using AI face swapping technology, you can replace Chinese faces with foreign faces with one click, eliminating the cost of remakes.
▲One-click face change in the short drama of going to sea, the picture is taken from X (original Twitter)
Now some AI face swapping tools can even complete face replacement based on a photo.
These technologies can be used for both daily entertainment and commercial activities, such as creating digital avatars of anchors to achieve 24-hour uninterrupted live streaming and sales.
But if deepfakes get into the hands of bad actors, they can become a tool for them to do bad things.
For example, scammers can use images collected on the web to create AI face-swapping videos and use such videos to commit fraud.
At the beginning of this year, the police in Hong Kong, China, received a report that a scammer impersonated the chief financial officer of a company through AI face swapping, and asked the company's financial transfer of 200 million Hong Kong dollars (about 180 million yuan).
In addition, there are scammers who use AI to change their faces, impersonating the victim's friends, children, parents, and other characters to defraud people. Seeing one's own family and friends in the video will make the victim let down their guard.
When the bad guys get their hands on deepfakes, they can have a real impact on the lives of ordinary people, and anyone can be deceived.
Confronted with deepfake content
What are we going to do?
So how exactly do you screen deepfakes? There are things that AI institutions and individuals can do about this.
1. For AI companies
AI agencies are also working to reduce the misuse of deepfake technology.
In the process of using some image generation models, such as Open AI's DALL· In the E model, you can't get the AI to process a specific photo, and there's no way to generate a specific face.
Most AI-generated models block specific words to prevent the model from being used to generate malicious images or videos. For example, Midjourney banned the word arrest after the proliferation of fake news images of "Trump arrested."
In addition, most institutions, including Open AI, Meta, midjourney, etc., will add watermarks to the generated image data, while formal institutions that provide AI face swapping will also embed watermarks in the video, which is different from real videos.
Of course, criminals, driven by profits, will rack their brains to remove the traces of forgery. Their goal is to make people think that the pictures and videos are real.
Therefore, for everyone, we also need to know some precautions.
2. For individuals
For us ordinary people, it is unlikely that we will have special identification software in our hands, and it is unlikely that we will analyze the flaws in the picture frame by frame, but there are still some simple methods that can help us analyze and judge.
For example, because there are some AI face-swapping videos that criminals make in advance.
If the person in the video mentions asking you to transfer money, you can ask the person in the video to do certain actions, such as turning around, and holding their hands in front of the camera for a while and waving.
In this way, it is not only possible to tell if the other party is a pre-recorded video, but at the same time, shaking the hand in front of the face will also interfere with some simple AI face swapping tools, and the "dummy" may reveal flaws.
In addition, you can also ask the other person to answer some messages that only you know, such as "Where was the last time you met?" or you can deliberately ask the other person about the children to see if they are impersonated by scammers, knowing that they do not have children.
After all, the other party has already mentioned that you will transfer money, and it is not too much to ask the other party to do a few actions and answer a few questions to verify your identity. And your own money bag, of course, you have to protect yourself.
In addition, you can keep in mind the "six laws" and "eight whatevers" principles mentioned by the Ministry of Public Security for fraud prevention.
"Six Laws"
1. When it comes to bank cards, they will all hang up;
2. When it comes to winning the lottery, all hang up;
3. When it comes to the "telephone transfer public procuratorate law" that requires remittance, it will be hung up;
4. All text messages that let you click on the link will be deleted;
5. Links sent by WeChat strangers will not be clicked;
6. References to "safe accounts" are all frauds.
"Eight Whatevers"
1. Those who claim to be required by the public procuratorate and the law to remit money;
2. Anything that allows you to remit money to a "safe account";
3. If you are notified of winning the lottery and receiving the prize, you will be asked to pay the money first;
4. Anyone who notifies the "family" that the accident must be remitted first;
5. Anyone who asks for bank card information and verification code on the phone;
6. Anyone who allows you to open online banking to be inspected;
7. Those who claim to be leaders ask for remittances;
8. Any unfamiliar website needs to register bank card information.
bibliography
[1]https://thispersondoesnotexist.com/
[2]http://www.news.cn/legal/20240226/da985bba58bf4ea2a65b872d06aec6a5/c.html
[3]https://www.scientificamerican.com/article/how-to-tell-if-a-photo-is-an-ai-generated-fake/
[4]https://about.fb.com/news/2024/02/labeling-ai-generated-images-on-facebook-instagram-and-threads/
[5]https://www.aiornot.com/
Author: Xiao Wei Popular science creator
Source: Popular Science China