Heart of Machine Reporting, Heart of Machine Editorial Department
The hottest topic on the Internet in recent days is "Look at him when he's old". Wireless Lab, headquartered in St. Petersburg, Russia, has set off a wave of social media, with its application "FaceAPP" gaining worldwide popularity in a very short period of time.
This application uses AI algorithm, which can instantly upload people's face photos "young" or "aging". It's the source of all the evil that has flooded the Internet these days with photos of stars aged 17/80.
Soccer players Ronaldo and Messi in 2049-50 season.
However, as FaceAPP based on artificial intelligence algorithm becomes the focus of attention, some people begin to worry that their privacy will be violated in the tide of "face-changing for all".
Application of overnight explosion
You've probably seen the images FaceAPP has processed. FaceApp, a photo editing app launched in 2017 as an image filter product, has again gained popularity in social media in recent days due to its dramatic upgrade in effect, and this week it topped the free lists of Apple and Android App Stores.
This application can automatically detect faces. It only needs to upload a photo to achieve a variety of effects, such as changing facial expressions, getting younger, getting older and changing gender. Of course, it also needs the most basic and practical beauty filter.
Using the FaceAPP to simulate, the Apple CEO Tim
Many Hollywood stars have completed the "age challenge" and many stars have even taken part in it voluntarily.
The Heart of the Machine also downloaded FaceApp from the App Store and tried out the free version. From the use point of view, the speed of uploading photos is relatively slow, but uploading once can carry out various operations, such as changing smile, hair color, hairstyle, age, beard, gender and so on. Each category will have some free style, but there are also many pay style. For example, in the option of changing beard, there are eight styles, ranging from no beard at all to beard, and only "trendy beard" is used free of charge.
As shown below, we took Demo faces to test the effects of various styles.
The first one is the original one, and the second one is the transformation effect of different styles. We can find that the effect is still very good, and we can achieve the same effect with our own photos. In addition to the above features, we have deliberately tried the effect of age change, and the free version also supports all age changes.
"It's collecting our data."
App, which uses artificial intelligence algorithms, has been welcomed by everyone, but this does not seem to be the case in American politics, which has recently been cracking down on privacy leaks. Because the application was developed by Russians, the Democratic National Committee (DNC) quickly issued a cautious warning to the Democratic presidential campaign team in 2020.
Chuck Schumer, the Senate Minority Leader, even formally asked the FBI and the Federal Trade Commission to launch a national security investigation into the application.
"This app allows users to transform different styles of people in photos, such as aging the characters in photos. Unfortunately, this novel attempt is not without risks: FaceApp was developed by Russians. "DNC Chief Security Officer Bob Lord warned.
Schumer wrote: "It would be very disturbing to provide sensitive personal information to a foreign force actively engaged in cyber hostilities against the United States. "
Although FaceAPP does upload images to the server for processing, this application only uploads specific images that need to be modified, not the entire album. Yaroslav Goncharov, CEO of FaceApp, told the Verge in 2017 that although the company sometimes stores images for short periods to improve AI model performance, most of them are quickly removed from the server.
FaceAPP issued a statement Wednesday on the allegations of privacy leaks.
FaceAPP performs most photo processing in the cloud, uploading only photos selected by users for editing, and never uploading other images to the cloud.
FaceAPP stores uploaded images in the cloud, mainly for performance and traffic considerations. FaceAPP wants to ensure that users do not upload photos repeatedly for each edit operation, and that most of the images are deleted from the server within 48 hours of uploading.
FaceAPP accepts the user's request to remove its image from the server, although the current support team is already at full load, but the requests will have a priority. If you want to send this type of request faster, the user can be in Settings-
All of FaceApp's functions are available without login, so 99% of users are not logged in, and FaceApp can't access any personal identity information.
FaceApp will not sell or share any user data to third parties.
Even if the R&D team is in Russia, the user data will not be transferred to Russia.
Finally, FaceApp said that after gaining access to the album, the software will only upload the selected image, which users can quickly check through the network sniffer tool.
Can AI really predict our future?
It seems that concerns about the violation of user privacy by artificial intelligence applications have been inflated, but we may still need a more transparent management mechanism.
With the progress of technology, AI's method of changing face has become more and more mature. Today we can not only make people younger in photos and change their expressions, but also synthesize a person's voice, and even generate a video for him or her that does not exist. "Deepfake," which has previously attracted attention, can produce false photos of celebrities, then make false news, or let politicians say things they have never said. Now the technology can be easily used by ordinary people.
Like many other applications, FaceApp invokes the application program interface (API) provided by Facebook and Google to achieve its filter effect through a neural network.
But is AI really reliable? What's the basic principle of face conversion? Suppose you stare at one person's video for 100 hours, then show you another person's picture, and then let you draw the picture just now from memory, you will draw it very much like the first person. In this case, machine learning algorithms such as Deepface usually use Autoencoder to generate similar images based on the original image. In the process of processing, the encoder compresses the picture, while the decoder restores the picture.
In order to change face, artificial intelligence algorithm also needs a lot of face image training after labeling. So, AI "predicts" what you will look like in the next 30 years is a superposition of the most probable results in the database, which seems reliable, but not rigorous.
Finally, there are faces that artificial intelligence can't handle. There are netizens on the famous Hollywood actor Morgan
It doesn't look like any difference.